Artificial intelligence-based adaptive cognitive training and real-time monitoring system

Abstract

This invention discloses an artificial intelligence-based adaptive cognitive training and real-time monitoring system, belonging to the field of cognitive impairment training technology. It includes: a multimodal data acquisition and preprocessing module for acquiring, preprocessing, and fusing multimodal data from patients to form multimodal fused data; a cognitive state analysis and evaluation module for analyzing and evaluating the multimodal fused data to determine the patient's cognitive state analysis and evaluation results; an adaptive cognitive training module for developing a cognitive training plan for patients and adaptively adjusting it based on their cognitive training progress; and a real-time monitoring, early warning, and control module for real-time monitoring and automatic early warning when abnormalities are detected. This invention solves the problem of existing systems failing to achieve precise intervention and real-time tracking, resulting in poor cognitive impairment training effects. This invention enables precise intervention and real-time tracking of patients, effectively delaying cognitive function decline, improving the effectiveness of cognitive impairment training, and enhancing patients' quality of life.

Description

Technical Field

[0001] This invention relates to the field of cognitive impairment training technology, specifically to an adaptive cognitive training and real-time monitoring system based on artificial intelligence. Background Technology

[0002] Alzheimer's disease is a degenerative disease of the central nervous system that mainly occurs in the elderly or pre-elderly. The main characteristics of the disease include progressive cognitive impairment and behavioral impairment. Cognitive impairment training is an important means of slowing the progression of Alzheimer's disease. Its core goal is to improve patients' cognitive functions such as memory, attention, language ability and executive function through specific training.

[0003] Existing technologies cannot achieve precise intervention and real-time tracking of Alzheimer's patients, nor can they effectively slow down the decline of patients' cognitive function, resulting in poor cognitive impairment training effects and reduced quality of life for patients. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based adaptive cognitive training and real-time monitoring system that enables precise intervention and real-time tracking of Alzheimer's patients, effectively slows down cognitive decline, improves the training effect of cognitive impairment and the quality of life of patients, and solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: An AI-based adaptive cognitive training and real-time monitoring system includes: The multimodal data acquisition and preprocessing module is configured as follows: Patient multimodal data is collected based on terminal devices, and the collected patient multimodal data is preprocessed and fused from different modalities to form patient multimodal fused data; The cognitive state analysis and assessment module is configured as follows: A patient cognitive state analysis and evaluation model is constructed based on deep learning technology. The fused multimodal fusion data of patients is analyzed and evaluated based on the patient cognitive state analysis and evaluation model to determine the patient cognitive state analysis and evaluation results. The adaptive cognitive training module is configured as follows: A cognitive training plan is developed based on the results of the patient's cognitive status analysis and assessment, and the plan is adaptively adjusted according to the patient's cognitive training progress. The real-time monitoring, early warning, and control module is configured as follows: The system monitors patients' adaptive cognitive training in real time, automatically issues warnings when abnormalities are detected, and promptly manages and controls the patients' adaptive cognitive training.

[0006] Preferably, based on the collection of patient multimodal data by the terminal device, the following operations are performed: The system uses cameras to monitor patients' reaction time, accuracy, error types, time required to complete tasks, and touchscreen gesture trajectories during game interactions, thereby acquiring patient game interaction data. The system uses cameras to capture patients' facial expressions, head postures, and eye gaze patterns to obtain patients' visual data. The system uses cameras to monitor the content, tone, speed, clarity, and frequency of pauses in patients' voice responses, thereby acquiring patient voice interaction data. Based on wearable devices, the patient's heart rate, heart rate variability and skin conductance are collected to assess the patient's stress state and emotional fluctuations during training and to obtain the patient's physiological signal data. Patient multimodal data is generated based on patient game interaction data, patient visual data, patient voice interaction data, and patient physiological signal data.

[0007] Preferably, the collected patient multimodal data is preprocessed by performing the following operations: Based on Python, the system cleans patient multimodal data, removes noise that is irrelevant to the patient's adaptive cognitive training, and identifies outliers in the patient multimodal data. Outliers in patient multimodal data are evaluated based on interpretability, persistence, and relevance to determine whether they are useful for adaptive cognitive training of patients. When outliers are useful for the patient's adaptive cognitive training, they are corrected; when outliers are not useful for the patient's adaptive cognitive training, they are deleted. Z-Score standardization is used to standardize patient multimodal data, transforming it into a distribution with a mean of 0 and a standard deviation of 1. This eliminates differences in the dimensions and ranges of values ​​among patient multimodal data, resulting in standardized patient multimodal data.

[0008] Preferably, the patient's multimodal data is fused using different modalities, and the following operations are performed: Align patient multimodal data based on timestamps so that data from different modalities in the patient multimodal data refer to the same event segment in time; Based on principal component analysis, feature extraction is performed on the multimodal data of patients. The high-dimensional data in the multimodal data of patients is reduced by orthogonal transformation. Uncorrelated principal components are extracted to maximize the data variance. Feature vectors related to the patient's adaptive cognitive training are extracted from the multimodal data of patients to determine the multimodal feature data of patients. Based on the attention mechanism, the multimodal feature data of patients are fused, and feature data from different modalities are fused and jointly analyzed to form a comprehensive patient cognitive state representation vector, thus determining the multimodal fused data of patients.

[0009] Preferably, based on the attention mechanism, the patient's multimodal feature data is fused, and feature data from different modalities are fused and jointly analyzed to form a comprehensive patient cognitive state representation vector, determining the patient's multimodal fused data, including: Each modality feature vector is mapped to a unified semantic space through a fully connected layer to obtain the mapped target feature vector; The target feature vectors are concatenated into a comprehensive patient cognitive state representation vector; Based on the gating mechanism, the dynamic gating weights of the target feature vector are calculated according to the comprehensive patient cognitive state representation vector; The target feature vectors of each modality are weighted and summed according to the dynamic gating weights to obtain the patient's multimodal fusion data.

[0010] Preferably, a patient cognitive state analysis and assessment model is constructed based on deep learning technology, and the following operations are performed: Collect patients' cognitive multimodal historical data and divide the data into training and test sets in a 7:3 ratio. Based on deep learning technology, a training set is used to train the deep learning model, enabling the deep learning model to autonomously learn the behavior of analyzing and evaluating patients' cognitive states from the training set, and automatically analyze and evaluate patients' cognitive states in terms of memory, executive ability, attention and language ability, and determine the patient cognitive state analysis and evaluation model. The patient cognitive state analysis and evaluation model was tested using a test set, and its performance was evaluated. When the patient cognitive state analysis and evaluation model could not automatically analyze and evaluate the patient's cognitive state, the parameters of the model were adjusted. After continuous iterative optimization, the optimal patient cognitive state analysis and evaluation model was formed.

[0011] Preferably, the fused patient multimodal fusion data is analyzed and evaluated based on the patient cognitive state analysis and evaluation model, and the following operations are performed: The fused multimodal data of the patient is input into the patient cognitive state analysis and assessment model. The model automatically analyzes the fused multimodal data and automatically assesses the patient's cognitive state, including memory, executive function, attention, and language ability, to determine the patient's cognitive state analysis and assessment results.

[0012] Preferably, a cognitive training plan is developed based on the patient's cognitive state analysis and assessment results, and the plan is adaptively adjusted according to the patient's cognitive training progress, by performing the following operations: Based on the results of the patient's cognitive status analysis and assessment, a patient cognitive training program is developed, and the patient's cognitive impairment is trained and intervened according to the patient cognitive training program; Among them, it is presented in a gamified form, including virtual shopping, card matching and puzzles, and cognitive training for patients is carried out based on gamification; It can identify the patient's emotional state and fatigue level in real time during cognitive training, and automatically adjust the difficulty of the cognitive training program according to the emotional state and fatigue level, so that the patient's challenge level is always maintained in the optimal anxiety zone, so that the patient will not feel bored due to the cognitive training program being too easy, nor will they feel frustrated due to the cognitive training program being too difficult.

[0013] Preferably, the system identifies the patient's emotional state and fatigue level in real time during cognitive training, automatically adjusts the difficulty of the cognitive training program based on the emotional state and fatigue level, and maintains the patient's challenge level in the optimal anxiety zone, performing the following operations: Based on the type of game used in the training intervention, obtain quantitative records of fatigue levels; Based on the supporting records corresponding to the quantitative fatigue level records, perform quantitative error root cause analysis to obtain the error logic type; If the error logic type is fatigue factor extraction logic error, then the corresponding fatigue degree quantification record will be removed. If the error logic type is factor quantization logic error, obtain the first factor quantization logic set; The first factor quantification logic set of the full factor quantification logic set that removes the quantification records of the same fatigue level is used to obtain the second factor quantification logic set. The first target factor quantization logic set is determined based on the intersection of the second factor quantization logic sets. Based on the first factor quantification logic set and its corresponding corroborating error deviation, error support relationship analysis is performed to obtain the analysis value; Based on the convergence of the first factor quantization logic with the same factor quantization logic type in the analysis values, determine the second target factor quantization logic set; The fatigue factor extraction logic and factor quantification logic corresponding to the game type are determined based on the first target factor quantification logic set and the second target factor quantification logic set. The degree of fatigue is determined based on the fatigue factor extraction logic and factor quantification logic corresponding to the game type. Determine the patient's level of challenge with the current game based on their emotional state and fatigue level; If the challenge level is below the optimal anxiety zone, increase the current game difficulty; If the challenge level is higher than the optimal anxiety zone, then lower the current game difficulty. The optimal anxiety zone is defined as being higher than the patient's current ability level in the current game. The method for determining this higher-level zone is as follows: Calculate the ratio of the patient's most recent checkpoint speed to their average checkpoint speed in the most recent checkpoint passage; Calculate the product of the preset horizontal range and the reciprocal of the speed ratio, and round down to obtain the value above the horizontal range.

[0014] Preferably, the patient's adaptive cognitive training is monitored in real time, with automatic alerts issued when abnormalities are detected, and the patient's adaptive cognitive training is promptly managed and controlled by performing the following operations: The system monitors patients’ adaptive cognitive training in real time and displays their daily cognitive training progress and emotional fluctuation trends in chart form. It also generates structured medical reports regularly to assist doctors in clinical diagnosis and efficacy evaluation. When real-time monitoring detects that a patient's cognitive training mood remains low or that the patient fails to complete cognitive training multiple times, an early warning alarm is automatically triggered, and the patient's adaptive cognitive training is promptly managed, forming a closed-loop management system for the patient's adaptive cognitive training.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention collects multimodal patient data using terminal devices, preprocesses the collected multimodal data, and fuses different modalities to form multimodal fused patient data. A patient cognitive state analysis and evaluation model is constructed based on deep learning technology. The fused multimodal fused patient data is then analyzed and evaluated according to this model to determine the patient cognitive state analysis and evaluation results. Based on these results, a patient cognitive training program is developed, and the program is adaptively adjusted according to the patient's cognitive training progress. The adaptive cognitive training is monitored in real time, with automatic warnings issued when abnormalities are detected. This closed-loop management system enables precise intervention and real-time tracking of Alzheimer's patients, effectively slowing cognitive decline, improving cognitive impairment training outcomes, and enhancing patients' quality of life. Attached Figure Description

[0016] Figure 1 This is a block diagram of the adaptive cognitive training and real-time monitoring system of the present invention; Figure 2 This is a flowchart of the adaptive cognitive training and real-time monitoring system of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] To address the current limitations of precise intervention and real-time tracking for Alzheimer's patients, which hinder effective slowing of cognitive decline and leads to poor cognitive impairment training outcomes and reduced quality of life, please refer to [link to relevant documentation]. Figures 1-2 This embodiment provides the following technical solution: An AI-based adaptive cognitive training and real-time monitoring system includes: a multimodal data acquisition and preprocessing module, a cognitive state analysis and evaluation module, an adaptive cognitive training module, and a real-time monitoring, early warning, and control module. The multimodal data acquisition and preprocessing module is used to collect multimodal patient data based on terminal devices, and to preprocess and fuse different modalities of the collected multimodal patient data to form multimodal fused patient data.

[0019] In this embodiment, based on the patient's multimodal data collected by the terminal device, the following operations are performed: The system uses cameras to monitor patients' reaction time, accuracy, error types, time required to complete tasks, and touchscreen gesture trajectories during game interactions, thereby acquiring patient game interaction data. The system uses cameras to capture patients' facial expressions (indifference, confusion, pleasure), head posture, and eye gaze trajectory (whether focused or wandering) to obtain patients' visual data. The system uses cameras to monitor the content, tone, speed, clarity, and frequency of pauses in patients' voice responses, thereby acquiring patient voice interaction data. Based on wearable devices, the patient's heart rate, heart rate variability and skin conductance are collected to assess the patient's stress state and emotional fluctuations during training and to obtain the patient's physiological signal data. Patient multimodal data is generated based on patient game interaction data, patient visual data, patient voice interaction data, and patient physiological signal data.

[0020] In this embodiment, the collected patient multimodal data is preprocessed by performing the following operations: Based on Python, the system cleans patient multimodal data, removes noise that is irrelevant to the patient's adaptive cognitive training, and identifies outliers in the patient multimodal data. Outliers in patient multimodal data are evaluated based on interpretability, persistence, and relevance to determine whether they are useful for adaptive cognitive training of patients. When outliers are useful for the patient's adaptive cognitive training, they are corrected; when outliers are not useful for the patient's adaptive cognitive training, they are deleted. Z-Score standardization is used to standardize patient multimodal data, transforming it into a distribution with a mean of 0 and a standard deviation of 1. This eliminates differences in the dimensions and ranges of values ​​among patient multimodal data, resulting in standardized patient multimodal data.

[0021] In this embodiment, data fusion of different modalities is performed on the patient's multimodal data, and the following operations are performed: Align patient multimodal data based on timestamps so that data from different modalities in the patient multimodal data refer to the same event segment in time; Principal component analysis was used to extract features from patient multimodal data. Orthogonal transformation was used to reduce the dimensionality of high-dimensional data in the patient multimodal data, and uncorrelated principal components were extracted to maximize data variance. Feature vectors related to patient adaptive cognitive training were extracted from the patient multimodal data to determine the patient multimodal feature data, including the mean, standard deviation and median of reaction time, the sliding window average of accuracy, the change sequence of error type, posture features, texture features, facial expression and posture representation features, Mel frequency cepstral coefficients, pitch, energy, speech rate and pause duration feature statistics, peak value of skin conductance response, rise time and recovery time.

[0022] Based on the attention mechanism, the multimodal feature data of patients are fused, and feature data from different modalities are fused and jointly analyzed to form a comprehensive patient cognitive state representation vector, thus determining the multimodal fused data of patients.

[0023] The cognitive state analysis and assessment module is used to construct a patient cognitive state analysis and assessment model based on deep learning technology, analyze and assess the fused multimodal fusion data of the patient based on the patient cognitive state analysis and assessment model, and determine the patient cognitive state analysis and assessment results.

[0024] In this embodiment, multimodal feature data of patients is fused based on an attention mechanism. Feature data from different modalities are fused and jointly analyzed to form a comprehensive patient cognitive state representation vector, and the patient multimodal fused data is determined, including: Each modality feature vector Each feature vector is mapped to a unified semantic space through a fully connected layer, resulting in the mapped target feature vector. : ; in, For the first Modal feature vectors; For the first Feature vectors mapped from modal feature vectors; =1, 2, ..., ; The total number of modes; It is the first The weight matrix of each modal eigenvector. In Represents the transpose of a matrix. , It is the first The dimension of each modal feature vector To unify the dimensions of the hidden layers; It is the first Bias vectors of modal eigenvectors ; The target feature vectors are concatenated into a comprehensive patient cognitive state representation vector. : ; in, , This represents a vertical splicing operation; Based on the gating mechanism, the dynamic gating weights of the target feature vector are calculated according to the comprehensive patient cognitive state representation vector. : ; in, Here is the weight matrix for the attention mechanism. ; This is the weight vector for the attention mechanism. ; This is a bias term for the attention mechanism. ; It is the hyperbolic tangent activation function; for Activation function; The target feature vectors of each modality are weighted and summed based on dynamic gating weights to obtain the patient's multimodal fusion data. : ; in, .

[0025] For example, : Modality 1: Electroencephalogram (EEG) features ; Modality 2: MRI image features ; Modal 3: Clinical scale scoring ; Define a unified feature space dimension ,So: , ; , ; , ; , , ; The mapped target feature vector is: , , ; Comprehensive patient cognitive state representation vector for: ; The calculation results of the dynamic gating weights are as follows: , , Patient multimodal fusion data .

[0026] This invention introduces an attention mechanism that can dynamically adjust the contribution weights of each modality, thereby forming a more representative comprehensive representation, improving the fusion accuracy of multimodal fusion data, and further improving the accuracy of subsequent patient cognitive state analysis and assessment.

[0027] In this embodiment, a patient cognitive state analysis and assessment model is constructed based on deep learning technology, and the following operations are performed: Collect patients' cognitive multimodal historical data and divide the data into training and test sets in a 7:3 ratio. Based on deep learning technology, a training set is used to train the deep learning model, enabling the deep learning model to autonomously learn the behavior of analyzing and evaluating patients' cognitive states from the training set, and automatically analyze and evaluate patients' cognitive states in terms of memory, executive ability, attention and language ability, and determine the patient cognitive state analysis and evaluation model. The patient cognitive state analysis and evaluation model was tested using a test set, and its performance was evaluated. When the patient cognitive state analysis and evaluation model could not automatically analyze and evaluate the patient's cognitive state, the parameters of the model were adjusted. After continuous iterative optimization, the optimal patient cognitive state analysis and evaluation model was formed.

[0028] In this embodiment, the fused patient multimodal fusion data is analyzed and evaluated based on the patient cognitive state analysis and evaluation model, and the following operations are performed: The fused multimodal data of the patient is input into the patient cognitive state analysis and assessment model. The model automatically analyzes the fused multimodal data and automatically assesses the patient's cognitive state, including memory, executive function, attention, and language ability, to determine the patient's cognitive state analysis and assessment results.

[0029] It should be noted that memory ability is based on the accuracy of delayed recall tasks; executive ability is based on the strategies and efficiency of tasks such as planning and classification; attention is based on reaction time and omission rate in continuous operation tests; and language ability is based on speech analysis of naming and repetition.

[0030] Specifically, memory training improves memory through methods such as number memorization and object recall. For example, memory tests can be conducted using playing cards, with the difficulty gradually increased. Images, objects, or associations can be used to aid memory, such as associating the content to be memorized with familiar things.

[0031] Specifically, attention training involves using visual tracking exercises, such as focusing on a moving ball or pen tip, to improve concentration; and arranging for patients to listen to and count specific sounds in a noisy environment to train their ability to resist distractions.

[0032] Specifically, language skills training involves improving language expression and comprehension through games such as word chain games, reading, and writing, and encouraging patients to repeat daily conversations or tell stories to enhance language fluency.

[0033] The adaptive cognitive training module is used to develop a cognitive training plan for patients based on the analysis and evaluation results of their cognitive state, and to adaptively adjust the cognitive training plan according to the patients' cognitive training progress.

[0034] In this embodiment, a cognitive training plan is formulated based on the patient's cognitive state analysis and evaluation results, and the plan is adaptively adjusted according to the patient's cognitive training progress, by performing the following operations: Based on the results of the patient's cognitive status analysis and assessment, a patient cognitive training program is developed, and the patient's cognitive impairment is trained and intervened according to the patient cognitive training program; Among them, it is presented in a gamified form, including virtual shopping, card matching and puzzles, and cognitive training for patients is carried out based on gamification; It can identify the patient's emotional state and fatigue level in real time during cognitive training, and automatically adjust the difficulty of the cognitive training program according to the emotional state and fatigue level, so that the patient's challenge level is always maintained in the optimal anxiety zone, so that the patient will not feel bored due to the cognitive training program being too easy, nor will they feel frustrated due to the cognitive training program being too difficult.

[0035] It should be noted that real-time identification of the patient's emotional state (such as frustration, anxiety, and pleasure) and fatigue level during cognitive training is used to dynamically adjust the difficulty of cognitive training and prevent patients from giving up due to excessive frustration.

[0036] In this embodiment, the patient's emotional state and fatigue level are identified in real time during cognitive training. The difficulty of the cognitive training program is automatically adjusted based on the patient's emotional state and fatigue level to maintain the patient's challenge level within the optimal anxiety zone. The following operations are performed: Based on the type of game used in the training intervention, quantitative records of fatigue levels are obtained.

[0037] Specifically, the game types are those used for cognitive training (virtual shopping, card matching, and puzzles). The fatigue level quantification record is a historical record of manually quantifying patients' fatigue levels based on their playing performance in different game types. For example, historical records of manually quantifying patients' fatigue levels based on card matching speed and accuracy (for instance, a card matching speed of 1.5 seconds / pair and an accuracy rate of 80% corresponds to a fatigue value of 40; a card matching speed of 0.75 seconds / pair and an accuracy rate of 90% corresponds to a fatigue value of 20).

[0038] Based on the supporting records corresponding to the quantified fatigue level records, a root cause analysis of the quantified error is performed to obtain the error logic type.

[0039] Specifically, corroborating records are auxiliary evidence data that support the verification of fatigue level quantification results, such as facial expression videos, eye-tracking data, and physiological signal waveforms. Root cause analysis of quantification errors refers to the systematic analysis of the fundamental reasons for the differences between fatigue quantification results and the actual state. Error logic types include fatigue factor extraction logic errors and factor quantification logic errors. Fatigue factor extraction logic errors refer to the error type that incorrectly identifies fatigue-related features from the original multimodal data. Factor quantification logic errors refer to the error type that correctly extracts fatigue features, but the rules for converting these features into quantified values ​​are incorrect.

[0040] If the error logic type is fatigue factor extraction logic error, then the corresponding fatigue degree quantification record will be removed.

[0041] Specifically, if the error is determined to originate from the feature extraction stage (such as misjudging emotional fluctuations as fatigue signals), the record should be directly removed to avoid contaminating subsequent analysis.

[0042] If the error logic type is factor quantization logic error, obtain the first factor quantization logic set.

[0043] Specifically, the first factor quantization logic set is a set of factor quantization rules determined for the error type of factor quantization logic error. When the error originates from the quantization transformation stage rather than feature extraction, the system will obtain a preliminary set of quantization rules to prepare for subsequent corrections.

[0044] The first factor quantification logic set is removed from the full factor quantification logic set of the same fatigue level quantification records to obtain the second factor quantification logic set.

[0045] Specifically, the full factor quantification logic set is the set of all factor quantification rules corresponding to each fatigue level quantification record. The first factor quantification logic set, which excludes the same fatigue level quantification record from the full factor quantification logic set, refers to the set portion of the full factor quantification logic set that corresponds to the first factor quantification logic set, i.e., the second factor quantification logic set is the set of reasonable factor quantification rules corresponding to each fatigue level quantification record.

[0046] The first target factor quantization logic set is determined based on the intersection of the second factor quantization logic sets.

[0047] Specifically, the first target factor quantification logic set is a set of common rules found from the second factor quantification logic set of the same game type, which represents a reliable set of rules that can be directly obtained from historical records.

[0048] Based on the first factor quantification logic set and its corresponding corroborating error deviation, error support relationship analysis is performed to obtain the analysis value.

[0049] Specifically, corroborating error deviation refers to the degree of difference between the corroborating record and the quantification result. Error support relationship analysis refers to analyzing the degree and direction of support for error deviation from different factor quantification logics within the same first factor quantification logic set. For example, if the card matching speed is 1.5 seconds / pair, the card matching accuracy is 80%, and the verified fatigue value is 40, then quantifying the card matching speed of 1.5 seconds / pair as a fatigue value of 60 gives it a support level of 10, with the support direction being the direction of increasing fatigue value. Quantifying the card matching accuracy of 80% as a fatigue value of 50 gives it a support level of 5, with the support direction also being the direction of increasing fatigue value. The analyzed values ​​represent the factor quantification logic, the degree of support for error deviation, and the direction of support.

[0050] Based on the convergence of the analysis values ​​to the first factor quantization logic of the same factor quantization logic type, determine the second target factor quantization logic set.

[0051] Specifically, convergence of the first factor quantization logic of the same type based on the analysis values ​​refers to iterative optimization of the first factor quantization logic of the same type using the analysis values, so that the degree of support deviation of the support error in the analysis values ​​tends to 0 and the support direction tends to be unbiased. The second target factor quantization logic set is the set of factor quantization logics obtained through the above iterative optimization.

[0052] The fatigue factor extraction logic and factor quantification logic corresponding to the game type are determined based on the first target factor quantification logic set and the second target factor quantification logic set.

[0053] Specifically, the extraction logic of the corresponding quantified factors in the first target factor quantification logic set and the second target factor quantification logic set is the fatigue factor extraction logic, and the first target factor quantification logic or the second target factor quantification logic is the corresponding factor quantification logic.

[0054] The degree of fatigue is determined based on the fatigue factor extraction logic and factor quantification logic corresponding to the game type.

[0055] Specifically, when determining the degree of fatigue based on the fatigue factor extraction logic and factor quantification logic corresponding to the game type, the real-time fatigue value is quantified based on the real-time game process, the fatigue factor extraction logic and factor quantification logic corresponding to the game type, and the real-time fatigue value is the determined degree of fatigue.

[0056] Determine the patient's level of challenge with the current game based on their emotional state and fatigue level.

[0057] Specifically, the challenge level refers to the degree of match between the current game difficulty and the patient's current ability level. The current ability level is the highest game difficulty that the patient can just complete without anxiety. The degree of match includes: how much the current game difficulty is higher than the highest game difficulty that can be completed without anxiety, how much the current game difficulty is equal to the highest game difficulty that the patient can complete without anxiety, and how much the current game difficulty is lower than the highest game difficulty that the patient can complete without anxiety.

[0058] If the challenge level is below the optimal anxiety zone, increase the current game difficulty; If the challenge level is higher than the optimal anxiety zone, then lower the current game difficulty. The optimal anxiety zone is defined as being higher than the patient's current ability level in the current game. The method for determining this higher-level zone is as follows: Calculate the ratio of the patient's most recent passage speed to the average passage speed of the most recent passage.

[0059] Specifically, the most recent completion speed is the patient's most recent successful completion speed, for example, 3 minutes / level. The average completion speed is the historical average completion speed of all game participants who recently completed a level, for example, 6 minutes / level. The speed ratio is: most recent completion speed / average completion speed, for example, 0.5.

[0060] Calculate the product of the preset horizontal range and the reciprocal of the speed ratio, and round down to obtain the value above the horizontal range.

[0061] Specifically, the preset level range is one game difficulty level. The product of the preset level range and the reciprocal of the speed ratio is calculated and rounded down. For example, continuing the example above, the preset level range is 1, the speed ratio is 0.5, and the level range above is: The optimal anxiety zone should be 2 difficulty levels higher than the current ability level (the higher level range indicates how many difficulty levels the optimal anxiety zone should be higher than the current ability level).

[0062] This invention introduces quantitative records and supporting records of fatigue levels for different game types to perform root cause analysis of quantitative errors. Based on the error logic type obtained from the analysis, the necessity of subsequent analysis is determined. If the error logic type is a fatigue factor extraction logic error, the corresponding record is directly removed. When the error logic type is a factor quantification logic error, a first factor quantification logic set is obtained for this error type. The set corresponding to the first factor quantification logic set in the full factor quantification logic set of the same fatigue level quantification record is removed to obtain a second factor quantification logic set. The intersection of these sets is used to find a set of commonly existing reliable rules (the first target factor quantification logic set). Error support relationship analysis is performed based on the first factor quantification logic set and its corresponding supporting error deviation to obtain analysis values. The first factor quantification logic of the same factor quantification logic type is converged based on the analysis values ​​to determine the second target factor quantification logic set. The first and second target factor quantification logic sets are integrated to determine the fatigue factor extraction logic and factor quantification logic corresponding to the game type, and the real-time game process is quantified to determine the fatigue level. Based on the patient's emotional state and fatigue level, the challenge level of the current game is determined. The challenge level is compared with the optimal anxiety zone to adjust the difficulty of the current game. The optimal anxiety zone is higher than the patient's current ability level in the current game. The higher-level range is dynamically adjusted based on the patient's most recent success, which improves the suitability and rationality of cognitive training.

[0063] The real-time monitoring and early warning control module is used to monitor the patient's adaptive cognitive training in real time, automatically issue early warnings when abnormalities are detected, and promptly control the patient's adaptive cognitive training.

[0064] In this embodiment, the patient's adaptive cognitive training is monitored in real time, and an automatic warning is issued when abnormalities are detected. The patient's adaptive cognitive training is then promptly controlled, and the following operations are performed: The system monitors patients’ adaptive cognitive training in real time and displays their daily cognitive training progress and emotional fluctuation trends in chart form. It also generates structured medical reports regularly to assist doctors in clinical diagnosis and efficacy evaluation. When real-time monitoring detects that a patient's cognitive training mood remains low or that the patient fails to complete cognitive training multiple times, an early warning alarm is automatically triggered, and the patient's adaptive cognitive training is promptly managed, forming a closed-loop management system for the patient's adaptive cognitive training.

[0065] In summary, this approach enables precise intervention and real-time tracking of Alzheimer's patients, effectively slowing cognitive decline and improving the effectiveness of cognitive impairment training and quality of life.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive cognitive training and real-time monitoring system based on artificial intelligence, characterized in that, include: The multimodal data acquisition and preprocessing module is configured as follows: Patient multimodal data is collected based on terminal devices, and the collected patient multimodal data is preprocessed and different modal data are fused to form patient multimodal fused data; The cognitive state analysis and assessment module is configured as follows: A patient cognitive state analysis and evaluation model is constructed based on deep learning technology. The fused multimodal fusion data of patients is analyzed and evaluated based on the patient cognitive state analysis and evaluation model to determine the patient cognitive state analysis and evaluation results. The adaptive cognitive training module is configured as follows: A cognitive training plan is developed based on the results of the patient's cognitive status analysis and assessment, and the plan is adaptively adjusted according to the patient's cognitive training progress. The real-time monitoring, early warning, and control module is configured as follows: The system monitors patients' adaptive cognitive training in real time, automatically issues warnings when abnormalities are detected, and promptly manages and controls patients' adaptive cognitive training. Based on the patient's cognitive status analysis and assessment results, a cognitive training plan is developed for the patient, and the plan is adaptively adjusted according to the patient's cognitive training progress. The following operations are performed: Based on the results of the patient's cognitive status analysis and assessment, a patient cognitive training program is developed, and the patient's cognitive impairment is trained and intervened according to the patient cognitive training program; Among them, it is presented in a gamified form, including virtual shopping, card matching and puzzles, and cognitive training for patients is carried out based on gamification; The system identifies the patient's emotional state and fatigue level in real time during cognitive training, automatically adjusts the difficulty of the cognitive training program based on these factors, and maintains the patient's challenge level within the optimal anxiety zone. The system performs the following operations: Based on the type of game used in the training intervention, obtain quantitative records of fatigue levels; Based on the supporting records corresponding to the quantitative fatigue level records, perform quantitative error root cause analysis to obtain the error logic type; If the error logic type is fatigue factor extraction logic error, then the corresponding fatigue degree quantification record will be removed. If the error logic type is factor quantization logic error, obtain the first factor quantization logic set; The first factor quantification logic set of the full factor quantification logic set that removes the quantification records of the same fatigue level is used to obtain the second factor quantification logic set. The first target factor quantization logic set is determined based on the intersection of the second factor quantization logic sets. Based on the first factor quantification logic set and its corresponding corroborating error deviation, error support relationship analysis is performed to obtain the analysis value; Based on the convergence of the first factor quantization logic with the same factor quantization logic type in the analysis values, determine the second target factor quantization logic set; The fatigue factor extraction logic and factor quantification logic corresponding to the game type are determined based on the first target factor quantification logic set and the second target factor quantification logic set. The degree of fatigue is determined based on the fatigue factor extraction logic and factor quantification logic corresponding to the game type. Determine the patient's level of challenge with the current game based on their emotional state and fatigue level; If the challenge level is below the optimal anxiety zone, increase the current game difficulty; If the challenge level is higher than the optimal anxiety zone, then lower the current game difficulty. The optimal anxiety zone is defined as being higher than the patient's current ability level in the current game. The method for determining this higher-level zone is as follows: Calculate the ratio of the patient's most recent checkpoint speed to their average checkpoint speed in the most recent checkpoint passage; Calculate the product of the preset horizontal range and the reciprocal of the speed ratio, and round down to obtain the value above the horizontal range.

2. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 1, characterized in that, Perform data fusion of different modalities on patient multimodal data, and perform the following operations: Align patient multimodal data based on timestamps so that data from different modalities in the patient multimodal data refer to the same event segment in time; Based on principal component analysis, feature extraction is performed on the multimodal data of patients. The high-dimensional data in the multimodal data of patients is reduced by orthogonal transformation. Uncorrelated principal components are extracted to maximize the data variance. Feature vectors related to the patient's adaptive cognitive training are extracted from the multimodal data of patients to determine the multimodal feature data of patients. Based on the attention mechanism, the multimodal feature data of patients are fused, and feature data from different modalities are fused and jointly analyzed to form a comprehensive patient cognitive state representation vector, thus determining the multimodal fused data of patients.

3. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 2, characterized in that, This study fuses multimodal feature data of patients based on an attention mechanism, integrating and jointly analyzing feature data from different modalities to form a comprehensive patient cognitive state representation vector. The resulting multimodal fused patient data includes: Each modality feature vector is mapped to a unified semantic space through a fully connected layer to obtain the mapped target feature vector; The target feature vectors are concatenated into a comprehensive patient cognitive state representation vector; Based on the gating mechanism, the dynamic gating weights of the target feature vector are calculated according to the comprehensive patient cognitive state representation vector; The target feature vectors of each modality are weighted and summed according to the dynamic gating weights to obtain the patient's multimodal fusion data.

4. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 2, characterized in that, The fused multimodal fusion data of patients was analyzed and evaluated based on the patient cognitive state analysis and evaluation model, and the following operations were performed: The fused multimodal data of the patient is input into the patient cognitive state analysis and assessment model. The model automatically analyzes the fused multimodal data and automatically assesses the patient's cognitive state, including memory, executive function, attention, and language ability, to determine the patient's cognitive state analysis and assessment results.

5. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 3, characterized in that, Based on the collection of multimodal patient data using terminal devices, perform the following operations: The system uses cameras to monitor patients' reaction time, accuracy, error types, time required to complete tasks, and touchscreen gesture trajectories during game interactions, thereby acquiring patient game interaction data. The system uses cameras to capture patients' facial expressions, head postures, and eye gaze patterns to obtain patients' visual data. The system uses cameras to monitor the content, tone, speed, clarity, and frequency of pauses in patients' voice responses, thereby acquiring patient voice interaction data. Based on wearable devices, the patient's heart rate, heart rate variability and skin conductance are collected to assess the patient's stress state and emotional fluctuations during training and to obtain the patient's physiological signal data. Patient multimodal data is generated based on patient game interaction data, patient visual data, patient voice interaction data, and patient physiological signal data.

6. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 4, characterized in that, The collected multimodal patient data was preprocessed by performing the following operations: Based on Python, the system cleans patient multimodal data, removes noise that is irrelevant to the patient's adaptive cognitive training, and identifies outliers in the patient multimodal data. Outliers in patient multimodal data are evaluated based on interpretability, persistence, and relevance to determine whether they are useful for adaptive cognitive training of patients. When outliers are useful for the patient's adaptive cognitive training, they are corrected; when outliers are not useful for the patient's adaptive cognitive training, they are deleted. Z-Score standardization is used to standardize patient multimodal data, transforming it into a distribution with a mean of 0 and a standard deviation of 1. This eliminates differences in the dimensions and ranges of values ​​among patient multimodal data, resulting in standardized patient multimodal data.

7. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 5, characterized in that, A patient cognitive state analysis and assessment model based on deep learning technology is constructed, and the following operations are performed: Collect patients' cognitive multimodal historical data and divide the data into training and test sets in a 7:3 ratio. Based on deep learning technology, a training set is used to train the deep learning model, enabling the deep learning model to autonomously learn the behavior of analyzing and evaluating patients' cognitive states from the training set, and automatically analyze and evaluate patients' cognitive states in terms of memory, executive ability, attention and language ability, and determine the patient cognitive state analysis and evaluation model. The patient cognitive state analysis and evaluation model was tested using a test set, and its performance was evaluated. When the patient cognitive state analysis and evaluation model could not automatically analyze and evaluate the patient's cognitive state, the parameters of the model were adjusted. After continuous iterative optimization, the optimal patient cognitive state analysis and evaluation model was formed.

8. The artificial intelligence-based adaptive cognitive training and real-time monitoring system according to claim 7, characterized in that, The system monitors the patient's adaptive cognitive training in real time, automatically issues warnings when abnormalities are detected, and promptly manages the patient's adaptive cognitive training by performing the following operations: The system monitors patients’ adaptive cognitive training in real time and displays their daily cognitive training progress and emotional fluctuation trends in chart form. It also generates structured medical reports regularly to assist doctors in clinical diagnosis and efficacy evaluation. When real-time monitoring detects that a patient's cognitive training mood remains low or that the patient fails to complete cognitive training multiple times, an early warning alarm is automatically triggered, and the patient's adaptive cognitive training is promptly managed, forming a closed-loop management system for the patient's adaptive cognitive training.

Images