Cognitive impairment intervention system and method based on general artificial intelligence

Abstract

The invention discloses a cognitive impairment intervention system and method based on general artificial intelligence, and the method comprises the steps: S1, collecting the physiological, behavior and cognitive state data of a patient, and carrying out the data preprocessing; s2, inputting a dual convolutional neural network to extract spatial-temporal features, and generating comprehensive cognitive state features; s3, time sequence clustering analysis is carried out, the key fluctuation period of the cognitive state is recognized, the decline trend is predicted, and the risk level is adjusted; s4, analyzing the emotional state of the patient in combination with a dynamic emotion recognition algorithm of micro-expression time sequence mapping, and optimizing an intervention scheme based on a user-defined emotion tag; s5, adjusting the intervention scheme by using an incremental learning method according to real-time feedback of the patient; and S6, optimizing the dual convolutional neural network by long-term training data, improving feature extraction and intervention precision, and realizing personalized long-term intelligent intervention. According to the method, multi-modal data and adaptive optimization are fused, accurate detection, personalized intervention and long-term optimization are realized, the rehabilitation efficiency is improved, and the life quality is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical and health information technology, and particularly to a cognitive impairment intervention system and method based on general artificial intelligence. Background Art

[0002] Cognitive impairment is an important health problem affecting the elderly population globally and some special groups, including various types such as Alzheimer's disease, vascular cognitive impairment, and Parkinson's disease-related cognitive impairment. With the intensification of population aging, the prevalence of cognitive impairment has been increasing year by year, bringing a heavy burden to the quality of life of patients, the family care burden, and the medical system. Early and accurate diagnosis and timely and effective intervention are the keys to improving the daily living ability of patients and delaying the progression of the disease. However, current cognitive intervention means still have many limitations and are difficult to meet the individualized, long-term, and intelligent rehabilitation needs.

[0003] Current cognitive interventions mainly rely on professional personnel for one-on-one guidance or use generalized training software to provide fixed practice models. These methods can improve the cognitive ability of patients to a certain extent, but there are significant limitations. First, one-on-one professional training requires a large amount of human resource investment, making it difficult to achieve large-scale promotion. Moreover, patients lack continuous training guidance in the home environment, resulting in unstable rehabilitation effects. Second, existing cognitive training programs usually adopt fixed training tasks without considering individual differences of patients, such as the disease stage, specific areas of cognitive function impairment, emotional state, etc., leading to a lack of pertinence in the training plan and affecting the intervention effect. In addition, many cognitive training systems only evaluate based on static data, lacking real-time adjustment of the dynamic state of patients and being unable to optimize the intervention plan according to training feedback, thus making it difficult to achieve long-term personalized optimization.

[0004] In recent years, artificial intelligence technology has developed rapidly in the field of medical and health, especially showing great potential in cognitive function assessment and intervention. Cognitive impairment screening methods based on machine learning and deep learning have initially achieved automated analysis, improving the accuracy and efficiency of diagnosis. However, in the field of cognitive function intervention, the current application of artificial intelligence still faces many challenges. For example, most intelligent intervention systems only rely on a single type of data input (such as language ability tests, simple question-and-answer interactions), failing to fully combine multi-modal information such as physiological data (such as brain waves, heart rate changes), behavioral data (such as gait analysis, activity patterns), and micro-expression data (such as emotional fluctuations), resulting in inaccurate cognitive modeling of the patient's state. In addition, current cognitive training systems usually rely on a preset task library and fail to make full use of deep learning technology for personalized task generation and dynamic adjustment, making it difficult to carry out refined intervention for individual differences of patients.

[0005] Therefore, how to provide a cognitive impairment intervention system and method based on general artificial intelligence is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] An object of the present invention is to provide a cognitive impairment intervention system and method based on general artificial intelligence. The present invention makes full use of artificial intelligence technologies such as dual convolutional neural networks, time series clustering analysis, micro-expression time series mapping, and incremental learning, details the process of real-time monitoring of patients' physiological, behavioral, and cognitive state data, and realizes targeted intervention solutions based on multi-modal data fusion, personalized emotion recognition, and adaptive cognitive training optimization, with the advantages of high accuracy, strong personalized adaptation ability, dynamic adjustment of training strategies, and long-term intervention optimization.

[0007] According to an embodiment of the present invention, a cognitive impairment intervention method based on general artificial intelligence includes the following steps:

[0008] S1. Real-time collect the physical data of the patient through a monitoring device and perform preprocessing. The physical data includes physiological data, behavioral data, and cognitive state data. The preprocessing includes data denoising, normalization, and feature extraction;

[0009] S2. Input the preprocessed physical data into a dual convolutional neural network, and extract data features from the time dimension and the space dimension respectively. The dual convolutional neural network includes two parallel convolutional modules, one of which is used to extract low-dimensional time series data features, and the other is used to extract high-dimensional space data features. Combine the two data features through a shared convolutional layer and generate a comprehensive cognitive state feature;

[0010] S3. Perform time series clustering analysis on the generated comprehensive cognitive state feature, use the adaptive clustering radius method to identify the key fluctuation period of the patient's cognitive state, and predict the decline trend of the patient's cognitive ability based on the periodic change. When the amplitude of the periodic change exceeds the set threshold, trigger a multi-level cognitive impairment risk assessment mechanism and adjust the risk level;

[0011] S4. During the patient's intervention, combine a dynamic emotion recognition algorithm based on micro-expression time series mapping to analyze the patient's emotional state, use a custom emotion label mapping rule to convert the emotional state into digital parameters, and associate them with the cognitive impairment risk level to generate a dynamically updated personalized intervention cognitive plan;

[0012] S5. According to the patient's actual response during the intervention process, use the incremental learning method to adjust the intervention cognitive plan in real time and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures;

[0013] S6. Continuously optimize the dual convolutional neural network, use the patient's treatment feedback as training data, continuously optimize feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

[0014] Optionally, the body data includes physiological data, behavioral data, and cognitive state data, and the preprocessing includes data denoising, normalization, and feature extraction;

[0015] Optionally, the S2 specifically includes:

[0016] S21. Receive the preprocessed patient body data, and construct the input data matrix X = [X t , X s , where X is composed of a time series data matrix and a spatial data matrix, X t is the time series data matrix, and X s is the spatial data matrix;

[0017] S22. Input X t into the time series convolutional module of the dual convolutional neural network, use a variable stride multi-scale convolutional kernel to extract time series features, and define the time series feature extraction mapping function:

[0018]

[0019] Among them, F t (X t ) is the time series feature mapping result, k t is the number of time series convolutional kernels, W t,i is the weight matrix of the i-th time series convolutional kernel, is the i-th time series window data, b t is the bias term, and σ is the non-linear activation function;

[0020] The time series convolutional kernel adopts a dynamic stride mechanism, and each W t,i adaptively adjusts the convolutional window size to adapt to the cognitive behavior characteristics of different time scales;

[0021] S23. Input X s into the spatial convolutional module of the dual convolutional neural network, use an asymmetric receptive field multi-resolution convolution to extract spatial features, and define the spatial feature mapping function:

[0022]

[0023] Among them, F s (X s ) is the spatial feature mapping result, k s is the number of spatial convolutional kernels, W s,j is the weight matrix of the j-th spatial convolutional kernel, It represents the data of the j-th group of spatial feature windows, which are the feature data intercepted at spatial position j, b s is the bias term;

[0024] The spatial convolution kernel adopts an asymmetric receptive field, that is, the convolution kernel sizes in different directions are different, enabling the system to adapt to the changes in the cognitive state features in different directions;

[0025] S24. Through the weight-sharing adaptive fusion mechanism, fuse F t (X t ) and F s (X s ), and define the calculation of the fused feature:

[0026] F(X) = α t F t (X t ) + α s F s (X s );

[0027]

[0028] Among them, F(X) is the final fused feature result, α t and α s are the dynamically learned weight parameters, γ t and γ s are the learnable parameters used to adjust the contribution ratios of the temporal features and the spatial features;

[0029] S25. Normalize the fused feature F(X) and perform dimensionality reduction to construct the final comprehensive cognitive state feature:

[0030]

[0031] Among them, is the normalized feature data, μ is the individual feature mean, σ 2 is the feature variance of the patient group, ∈ is the smoothing factor to ensure the stability of the data distribution and avoid the influence of individual differences on the generalization ability of feature extraction, Z is the comprehensive cognitive state feature vector, and P is the personalized feature projection matrix to ensure that the features after dimensionality reduction can still represent the key cognitive state information of the patient.

[0032] Optionally, the specific steps of S3 include:

[0033] S31. Based on the comprehensive cognitive state feature vector Z, construct a temporal feature matrix:

[0034] Q = [Z1, Z2,..., Z T T ; ​

[0035] Among them, Q is the time-series feature matrix, representing the comprehensive cognitive state of the patient over T time steps, where T is the length of the time step, and Z t is the comprehensive cognitive state feature vector at time step t, representing the physiological, behavioral, and cognitive state data of the patient;

[0036] S32. Based on each time step t, calculate the weighted cognitive state change rate of the time-series feature matrix:

[0037]

[0038] Among them, R t is the weighted cognitive state change rate at time step t, d is the dimension of the time-series feature, that is, the number of features in the time-series feature matrix Q, and Q t,i is the i-th dimensional eigenvalue at time t, representing different cognitive state information of the patient, and Q t-1,i is the i-th dimensional eigenvalue at time t - 1, and ω i is the feature weight coefficient, α is the generalized distance norm index, used to adjust the sensitivity between different feature change rates, and T is the length of the time step;

[0039] S33. Based on the weighted cognitive state change rate, calculate the local density of each time step:

[0040]

[0041] Among them, ρ t is the local density at time step t, used to measure whether this time step belongs to the period of drastic fluctuations in the cognitive state, T is the length of the time step, and R j is the cognitive state change rate at time step j, and δ is the adaptive clustering radius, used to control the calculation range of the local density;

[0042] Set the local density threshold λ, and identify the key fluctuation period set C = {t|ρ t > λ}, where C is the key fluctuation period set, containing all time steps with local density exceeding the threshold, that is, the time points where the cognitive state changes drastically;

[0043] S34. Based on the key fluctuation period set, calculate the decline trend of cognitive ability:

[0044] S t = γS t-1 +(1 - γ)R t , S0 = 0, t ∈ C;

[0045] Among them, S t is the decline trend index of cognitive ability at time step t, used to predict long-term cognitive state changes, and S t-1is the decline trend index of the previous time step, and γ is the historical trend weight factor used to balance short-term and long-term trends;

[0046] S35. When the fluctuation amplitude of the cognitive state exceeds the set threshold θ, adjust the cognitive impairment risk level:

[0047]

[0048] Among them, L is the adjusted cognitive impairment risk level, L0 is the initial risk level, and β is the risk adjustment factor used to control the dynamic adjustment amplitude of the risk level. and are respectively the maximum and minimum values of the decline trend of cognitive ability, used to measure the periodic fluctuation amplitude of the patient's cognitive state.

[0049] Optionally, the S4 specifically includes:

[0050] S41. Based on the time series feature matrix Q, adopt a multi-layer adaptive trajectory mapping method to map the time series features to the micro-expression feature matrix:

[0051]

[0052] Among them, M t is the micro-expression feature vector at time step t, M t-j is the micro-expression feature vector at time step t - j, representing the facial state in the past j time steps, H is the length of the historical time series window. is the autoregressive coefficient, indicating the influence of the historical micro-expression state M t-j on the current state, φ i is the time series feature contribution coefficient, d is the dimension of the time series features, that is, the number of features in the time series feature matrix Q, Q t,i is the i-th dimensional time series feature value at time step t, k is the number of micro-expression key points, θ l is the trajectory change weight, measuring the movement amplitude of facial muscles, ∈ t is the random perturbation term used to model the influence of unobserved factors on micro-expressions, T is the time step length, ΔP t,l is the relative movement trajectory of the l-th micro-expression key point, and ΔP t,l = P t,l - P t-1,l where P t,l is the coordinate of the l-th micro-expression key point at time step t, and P t-1,l is the coordinate of the l-th micro-expression key point at time step t - 1;

[0053] S42. Based on the relative movement trajectory change, define the emotional change rate and use the emotional change rate as the non-linear movement change rate of the micro-expression feature at time step t:

[0054]

[0055] Among them, is the emotional change rate at time step t, b is the number of facial features, that is, the dimension of facial key points, M t,i is the i-th dimensional micro-expression feature vector at time step t, M t-1,i is the i-th dimensional micro-expression feature vector at time step t - 1, ∈ is a small value to prevent the denominator from being zero to ensure calculation stability, and β is a non-linear adjustment exponent to enhance the sensitivity to small emotional fluctuations;

[0056] S43. Adopt a custom emotional label mapping rule to map the emotional change rate to an emotional label:

[0057]

[0058] Among them, E t is the emotional label score at time step t, which measures the current possible emotional state of the patient, C is the number of custom emotional categories, λ c is the confidence adjustment coefficient for category c, μ c is the mean of category c, representing the central value of this category, σ c is the standard deviation of category c, representing the range of variation of this category;

[0059] S44. Based on the emotional label score, use the information entropy measurement method to calculate the emotional instability index:

[0060]

[0061] Among them, is the emotional instability index at time step t, E t,c is the emotional label score corresponding to category c at time step t, log2 is the logarithmic function, and η is the micro-expression change adjustment coefficient, combining the contribution of micro-expression features to emotional fluctuations;

[0062] S45. Based on the emotional instability index and the cognitive impairment risk level L, calculate the personalized emotion intervention parameter:

[0063]

[0064] Among them, A t is the personalized emotion intervention parameter at time step t, Γ is the emotion intervention gain matrix, controlling the intervention intensity of different emotional states, λ L is the cognitive impairment risk influence coefficient, is the risk level of cognitive impairment under the emotion label at time step t, and ξ is the emotion-cognition coupling regulation parameter, which measures the coupling effect of emotional fluctuation and cognitive fluctuation. is the mean value of the emotional stability index. is the mean value of the cognitive ability decline trend, S t is the cognitive ability decline trend index at time step t.

[0065] S46. According to the personalized emotion intervention parameters, the personalized intervention cognitive plan is adjusted in real time and dynamically updated in real time to ensure the adaptability and pertinence of the intervention measures.

[0066] The cognitive dysfunction intervention system based on general artificial intelligence according to the embodiment of the present invention includes:

[0067] A data collection and preprocessing module, which is used to collect the physiological data, behavior data and cognitive state data of patients in real time through monitoring devices, and perform denoising, normalization and feature extraction on the collected data.

[0068] A spatio-temporal feature extraction module, which is used to input the preprocessed body data into a dual convolutional neural network, extract low-dimensional time-series data features from the time dimension, extract high-dimensional spatial data features from the space dimension, and merge the two data features through a shared convolutional layer to generate comprehensive cognitive state features.

[0069] A time-series clustering analysis module, which is used to perform time-series clustering analysis on the comprehensive cognitive state features, use the adaptive clustering radius method to identify the key fluctuation periods of the patient's cognitive state, and predict the cognitive ability decline trend of the patient based on the periodic changes. When the amplitude of the periodic changes exceeds the set threshold, a multi-level cognitive impairment risk assessment mechanism is triggered and the risk level is adjusted.

[0070] A dynamic emotion recognition and mapping module, which is used to analyze the patient's emotion state by combining a dynamic emotion recognition algorithm based on micro-expression time-series mapping during the patient's intervention, and use a custom emotion label mapping rule to convert the emotion state into digital parameters, associate with the cognitive impairment risk level, and generate a dynamically updated personalized intervention cognitive plan.

[0071] A personalized intervention adjustment module, which is used to adjust the intervention cognitive plan in real time according to the patient's actual reaction during the intervention process by using the incremental learning method, and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures.

[0072] A long-term learning and optimization module, which is used to continuously optimize the dual convolutional neural network based on the patient's treatment feedback, continuously optimize the feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

[0073] The beneficial effects of the present invention are:

[0074] First, the present invention uses a dual convolutional neural network to extract multi-modal feature data of patients from the temporal and spatial dimensions, and combines a time series clustering analysis method to accurately identify the key fluctuation periods of the cognitive state of patients, thereby effectively predicting the decline trend of cognitive ability. This technology breaks through the limitation of traditional cognitive assessment relying on regular tests, enabling the system to real-time perceive the cognitive changes of patients, improving the accuracy of early diagnosis and the timeliness of intervention.

[0075] Secondly, the present invention combines a dynamic emotion recognition algorithm based on micro-expression time series mapping to analyze the emotional state of patients during the intervention process, and uses a custom emotion label mapping rule to quantify the emotional state and associate it with the risk level of cognitive impairment, thereby optimizing the intervention plan. Compared with the defect of existing cognitive training systems that only focus on the completion of cognitive tasks and ignore the emotional changes of patients, the present invention can ensure that the training tasks are carried out when the patient's emotion is stable, and adjust the training difficulty and content according to the emotional fluctuations, improving the comfort and effectiveness of the intervention, and avoiding the decline of training efficiency caused by unstable emotions.

[0076] Finally, the present invention uses an incremental learning method to enable the system to continuously optimize the personalized intervention plan according to the training feedback of patients. By continuously adjusting the training content, training duration and difficulty level, the system can dynamically optimize for the cognitive ability, recovery speed and emotional state of different patients, realizing the upgrade of long-term personalized intervention plans. This continuous learning mechanism solves the deficiency that existing systems cannot make long-term adaptive adjustments according to individual needs, enabling patients to obtain more accurate and effective cognitive rehabilitation training during the long-term intervention process, ultimately improving the daily living ability, delaying the disease progression and improving the quality of life. Brief Description of the Drawings

[0077] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0078] Figure 1 is a flow chart of a cognitive impairment intervention method based on general artificial intelligence proposed by the present invention. Detailed Embodiments

[0079] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic way, so they only show the components related to the present invention.

[0080] Reference Figure 1 , a cognitive impairment intervention method based on general artificial intelligence, includes the following steps:

[0081] S1. Real-time collect the patient's body data through monitoring devices and perform preprocessing. The body data includes physiological data, behavioral data, and cognitive state data. The preprocessing includes data denoising, normalization, and feature extraction;

[0082] S2. Input the preprocessed body data into a dual convolutional neural network to extract data features from the time dimension and the space dimension respectively. The dual convolutional neural network contains two parallel convolutional modules, one for extracting low-dimensional time-series data features and the other for extracting high-dimensional spatial data features. Combine the two types of data features through a shared convolutional layer and generate comprehensive cognitive state features;

[0083] S3. Perform time-series clustering analysis on the generated comprehensive cognitive state features, use the adaptive clustering radius method to identify the key fluctuation periods of the patient's cognitive state, and predict the decline trend of the patient's cognitive ability based on the periodic changes. When the amplitude of the periodic changes exceeds the set threshold, trigger a multi-level cognitive impairment risk assessment mechanism and adjust the risk level;

[0084] S4. During the patient's intervention, combine the dynamic emotion recognition algorithm based on micro-expression time-series mapping to analyze the patient's emotional state, use the custom emotion label mapping rule to convert the emotional state into digital parameters, and associate them with the cognitive impairment risk level to generate a dynamically updated personalized intervention cognitive plan;

[0085] S5. According to the patient's actual response during the intervention process, adjust the intervention cognitive plan in real time through the incremental learning method and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures;

[0086] S6. Continuously optimize the dual convolutional neural network, use the patient's treatment feedback as training data, continuously optimize feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

[0087] In this embodiment, the body data includes physiological data, behavioral data, and cognitive state data. The preprocessing includes data denoising, normalization, and feature extraction;

[0088] In this embodiment, the specific content of S2 is as follows:

[0089] S21. Receive the preprocessed patient body data and construct an input data matrix X = [X t , X s , where X is composed of a time-series data matrix and a spatial data matrix. X t is the time-series data matrix, and X s is the spatial data matrix;

[0090] S22. Take X tInput the temporal convolution module of the dual convolutional neural network, use variable-stride multi-scale convolutional kernels to extract temporal features, and define the temporal feature extraction mapping function:

[0091]

[0092] Among them, F t (X t ) is the temporal feature mapping result, k t is the number of temporal convolutional kernels, W t,i is the weight matrix of the i-th temporal convolutional kernel, is the data of the i-th temporal window, b t is the bias term, and σ is the non-linear activation function;

[0093] The temporal convolutional kernel adopts a dynamic stride mechanism, and each W t,i adaptively adjusts the size of the convolutional window to adapt to the cognitive behavior characteristics of different time scales;

[0094] S23. Input X s into the spatial convolution module of the dual convolutional neural network, use asymmetric receptive field multi-resolution convolution to extract spatial features, and define the spatial feature mapping function:

[0095]

[0096] Among them, F s (X s ) is the spatial feature mapping result, k s is the number of spatial convolutional kernels, W s,j is the weight matrix of the j-th spatial convolutional kernel, is the data of the j-th group of spatial feature windows, representing the feature data intercepted at spatial position j, b s is the bias term;

[0097] The spatial convolutional kernel adopts an asymmetric receptive field, that is, the convolutional kernel sizes in different directions are different, enabling the system to adapt to the changes in cognitive state characteristics in different directions;

[0098] S24. Through the weight sharing adaptive fusion mechanism, fuse F t (X t ) and F s (X s ), and define the fused feature calculation:

[0099] F(X) = α t F t (X t ) + α s F s (X s );

[0100]

[0101] Among them, F(X) is the final fused feature result, α t and α s are dynamically learned weight parameters, γ t and γ s are learnable parameters used to adjust the contribution ratio of temporal features and spatial features;

[0102] S25. Normalize the fused feature F(X) and perform dimensionality reduction to construct the final comprehensive cognitive state feature:

[0103]

[0104] Among them, is the normalized feature data, μ is the individual feature mean, σ 2 is the feature variance of the patient group, ∈ is a smoothing factor to ensure stable data distribution and avoid the influence of individual differences on the generalization ability of feature extraction, Z is the comprehensive cognitive state feature vector, and P is the personalized feature projection matrix to ensure that the features after dimensionality reduction can still represent the key cognitive state information of the patient.

[0105] In this embodiment, S3 specifically includes:

[0106] S31. Based on the comprehensive cognitive state feature vector Z, construct a temporal feature matrix:

[0107] Q = [Z1, Z2, …, Z T T ;

[0108] Among them, Q is the temporal feature matrix, representing the comprehensive cognitive state of the patient at T time steps, T is the time step length, and Z t is the comprehensive cognitive state feature vector at time step t, representing the physiological, behavioral, and cognitive state data of the patient;

[0109] S32. Based on each time step t, calculate the weighted cognitive state change rate of the temporal feature matrix:

[0110]

[0111] Among them, R t is the weighted cognitive state change rate at time step t, d is the dimension of the temporal feature, that is, the number of features in the temporal feature matrix Q, Q t,i is the i-th dimensional feature value at time t, representing different cognitive state information of the patient, Q t-1,i is the i-th dimensional feature value at time t - 1, ω i ​is the feature weight coefficient, α is the generalized distance norm exponent used to adjust the sensitivity between different feature change rates, T is the time step length;

[0112] S33. Calculate the local density at each time step based on the weighted cognitive state change rate:

[0113]

[0114] where ρ t is the local density at time step t, used to measure whether this time step belongs to the period of violent fluctuations in the cognitive state, T is the time step length, R j is the cognitive state change rate at time step j, and δ is the adaptive clustering radius used to control the calculation range of the local density;

[0115] Set the local density threshold λ, and identify the key fluctuation period set C = {t|ρ t > λ}, where C is the key fluctuation period set, containing all time steps with local density exceeding the threshold, that is, the time points when the cognitive state changes violently;

[0116] S34. Calculate the cognitive ability decline trend based on the key fluctuation period set:

[0117] S t = γS t-1 + (1 - γ)R t , S0 = 0, t ∈ C;

[0118] where S t is the cognitive ability decline trend index at time step t, used to predict the long-term cognitive state change, S t-1 is the decline trend index of the previous time step, and γ is the historical trend weight factor used to balance the short-term and long-term trends;

[0119] S35. When the fluctuation amplitude of the cognitive state exceeds the set threshold θ, adjust the cognitive impairment risk level:

[0120]

[0121] where L is the adjusted cognitive impairment risk level, L0 is the initial risk level, β is the risk adjustment factor used to control the dynamic adjustment amplitude of the risk level, and are the maximum and minimum values of the cognitive ability decline trend respectively, used to measure the periodic fluctuation amplitude of the patient's cognitive state.

[0122] In this embodiment, the specific content of S4 includes:

[0123] S41. Based on the temporal feature matrix Q, adopt a multi-layer adaptive trajectory mapping method to map the temporal features to the micro-expression feature matrix:

[0124]

[0125] Among them, M t is the micro-expression feature vector at time step t, M t-j is the micro-expression feature vector at time step t−, representing the facial states in the past j time steps, H is the length of the historical temporal window, is the autoregressive coefficient, representing the influence of the historical micro-expression state M t-j on the current state, φ i is the temporal feature contribution coefficient, d is the dimension of the temporal features, that is, the number of features in the temporal feature matrix Q, Q t,i is the i-th dimensional temporal feature value at time step t, k is the number of micro-expression key points, θ l is the trajectory change weight, measuring the movement amplitude of facial muscles, ∈ t is the random perturbation term, used to model the influence of unobserved factors on micro-expressions, T is the time step length, ΔP t,l is the relative movement trajectory of the l-th micro-expression key point, and ΔP t,l =P t,l −P t-1,l , where P t,l is the coordinate of the l-th micro-expression key point at time step t, P t-1,l is the coordinate of the l-th micro-expression key point at time step t−1;

[0126] S42. Based on the relative movement trajectory change, define the emotion change rate, and use the emotion change rate as the non-linear movement change rate of the micro-expression feature at time step t:

[0127]

[0128] Among them, is the emotion change rate at time step t, b is the number of facial features, that is, the dimension of facial key points, M t,i is the i-th dimensional micro-expression feature vector at time step t, M t-1,i is the i-th dimensional micro-expression feature vector at time step t−1, ∈ is a small value to prevent the denominator from being zero, ensuring calculation stability, and β is the non-linear adjustment exponent, enhancing the sensitivity to small emotion fluctuations;

[0129] S43. Adopt a custom emotion label mapping rule to map the emotion change rate to emotion labels:

[0130]

[0131] Among them, Et is the emotional label score at time step t, measuring the patient's current possible emotional state, C is the number of custom emotional categories, λ c is the confidence adjustment coefficient for category c, μ c is the mean of category c, representing the central value of this category, σ c is the standard deviation of category c, representing the range of variation of this category;

[0132] S44. Based on the emotional label score, use the information entropy measurement method to calculate the emotional instability index:

[0133]

[0134] where is the emotional instability index at time step t, E t,c is the emotional label score corresponding to category c at time step t, log2 is the logarithmic function, η is the micro-expression change adjustment coefficient, combining the contribution of micro-expression features to emotional fluctuations;

[0135] S45. Based on the emotional instability index and the cognitive impairment risk level L, calculate the personalized emotional intervention parameter:

[0136]

[0137] where A t is the personalized emotional intervention parameter at time step t, Γ is the emotional intervention gain matrix, controlling the intervention intensity of different emotional states, λ L is the cognitive impairment risk influence coefficient, is the cognitive impairment risk level under the emotional label at time step t, ξ is the emotion-cognition coupling adjustment parameter, measuring the coupling effect of emotional fluctuations and cognitive fluctuations, is the mean of the emotional stability index, is the mean of the cognitive ability decline trend, S t is the cognitive ability decline trend index at time step t;

[0138] S46. According to the personalized emotional intervention parameter, adjust the personalized intervention cognitive plan in real time and perform dynamic real-time update to ensure the adaptability and pertinence of the intervention measures.

[0139] The cognitive dysfunction intervention system based on general artificial intelligence includes:

[0140] The data acquisition and preprocessing module is used to collect the patient's physiological data, behavioral data and cognitive state data in real time through monitoring devices, and perform denoising, normalization and feature extraction on the collected data;

[0141] A spatio-temporal feature extraction module, which is used to input the preprocessed body data into a dual convolutional neural network, extract low-dimensional temporal data features from the time dimension, extract high-dimensional spatial data features from the space dimension, and merge the two types of data features through a shared convolutional layer to generate comprehensive cognitive state features;

[0142] A temporal clustering analysis module, which is used to perform temporal clustering analysis on the comprehensive cognitive state features, use the adaptive clustering radius method to identify the key fluctuation periods of the patient's cognitive state, and predict the decline trend of the patient's cognitive ability based on the periodic changes. When the amplitude of the periodic changes exceeds the set threshold, a multi-level cognitive impairment risk assessment mechanism is triggered, and the risk level is adjusted;

[0143] A dynamic emotion recognition and mapping module, which is used to, during the period when the patient is receiving intervention, analyze the patient's emotional state in combination with a dynamic emotion recognition algorithm based on micro-expression temporal mapping, and convert the emotional state into digital parameters using a custom emotion label mapping rule, associate with the cognitive impairment risk level, and generate a dynamically updated personalized intervention cognitive plan;

[0144] A personalized intervention adjustment module, which is used to, according to the patient's actual reaction during the intervention process, use the incremental learning method to adjust the intervention cognitive plan in real time and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures;

[0145] A long-term learning and optimization module, which is used to continuously optimize the dual convolutional neural network based on the patient's treatment feedback, continuously optimize the feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

[0146] Example 1:

[0147] To verify the feasibility of the present invention in implementation, the present invention is applied to the rehabilitation training of cognitive impairment patients in the neurology department of a certain tertiary hospital. This experimental study lasted for 12 months. The test subjects were 80 patients with mild to moderate cognitive dysfunction, aged between 55 and 80 years old. All patients were diagnosed as mild to moderate Alzheimer's disease (AD) or vascular cognitive impairment (VaD) by professional doctors. The experimental scenario was the intelligent rehabilitation center of the hospital. The patients participated in the intervention training at the specified time every day. Each training session lasted for 45 minutes and was carried out 5 times a week. The experimental locations included the cognitive training laboratory of the hospital and the patient's home environment. The experimental equipment included physiological data monitoring equipment, facial recognition cameras, behavior tracking sensors, and the AI cognitive intervention system of the present invention.

[0148] During the training process, the system first collects the patient's physiological data (such as heart rate, blood oxygen saturation, brain waves), behavioral data (such as gait, movement coordination), and cognitive state data (such as memory test results, language expression ability) in real time through monitoring devices. After denoising, normalizing, and feature extraction, all the data is input into a dual convolutional neural network to extract low-dimensional time-series data features from the time dimension, high-dimensional spatial data features from the spatial dimension, and generate comprehensive cognitive state features. The system performs time-series clustering analysis on the patient's cognitive state, identifies key fluctuation periods, and predicts the trend of the patient's cognitive ability decline by combining the adaptive clustering radius method. When the system detects a significant fluctuation in the cognitive state, for example, a certain patient's short-term memory test score drops by more than 15% within three consecutive days, or the speech fluency score decreases by more than 10%, the system automatically triggers the risk assessment mechanism, adjusts the risk level from low risk to moderate risk, and increases the intensity of relevant intervention tasks.

[0149] During the intervention process, the present invention uses a dynamic emotion recognition algorithm based on micro-expression time-series mapping to analyze the patient's emotional state in real time. By collecting the patient's facial micro-expression features through a camera and combining custom emotion label mapping rules, the system can accurately identify the patient's emotional changes. For example, when a certain patient is performing a memory task, the system detects that the facial tension increases by 20%, the eye movement speed decreases by 15%, and the facial muscle stiffness increases by 10%, and the emotional state changes from "concentrated" to "anxious". At this time, the system automatically reduces the task difficulty and guides the patient into the relaxation training session to avoid the decline of training effect due to excessive stress. Compared with the phenomenon that some patients give up training due to anxiety or frustration caused by the fixed task mode of the traditional training system, the present invention significantly improves the training persistence and completion rate through an intelligent emotion adjustment strategy.

[0150] In addition, the system adopts an incremental learning mechanism to continuously optimize the intervention plan based on the patient's training feedback. The training system records the patient's training data, including task completion time, correct rate, emotional change trend, etc., and continuously adjusts the training content based on these data. For example, during the first three months of training for a certain patient, the correct rate of their language memory task has increased by 30%. The system detects that their language ability has reached a relatively high level, so it automatically adjusts the training content, reduces the proportion of language memory tasks, and increases the difficulty of logical reasoning tasks to better meet the recovery needs of this patient. In contrast, for another patient with continuously weak spatial cognitive ability, the system automatically increases the frequency of three-dimensional spatial cognitive training and recommends more real-scene interaction tasks in combination with behavioral data analysis. The experimental result data is as follows:

[0151] Table 1 Comparison of Experimental Data

[0152]

[0153]

[0154] The experimental results of the present invention show that the cognitive function score (MMSE) of patients using this system increased by an average of 4.5 points after 6 months, while that of the traditional training group only increased by 2.1 points, showing a significant advantage. At the same time, the personalized intervention strategy of the system enabled the training completion rate of patients to reach 92.5%, an increase of 18% compared to the traditional training system. In terms of the patients' ability to take care of themselves (ADL score), the patients using this system increased by an average of 7.8 points, while the traditional training group increased by 4.3 points. In terms of emotional stability, the patients adopting the present invention had a 25% reduction in the anxiety emotion detection score, while the improvement rate of anxiety emotions by the traditional method was only 12%. These data indicate that the present invention can not only improve cognitive function, but also effectively improve the compliance of training, optimize the psychological state of patients, so as to achieve more efficient and accurate personalized cognitive intervention.

[0155] In summary, the present invention shows significant advantages in the intervention process of patients with cognitive impairment. It can provide a personalized cognitive intervention plan based on real-time data, dynamically adjust the difficulty of training tasks, optimize the intervention experience in combination with the emotional state, and continuously optimize the training content using the incremental learning mechanism, enabling patients to adapt to personalized training in the long term and improving the training effect and compliance. This system not only reduces the dependence on professional intervention, but also provides a more accurate, efficient and personalized solution in the long-term rehabilitation process of patients, providing a more intelligent and scientific method for the intervention of cognitive function disorders.

[0156] The above is only a preferred specific embodiment 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 of the present invention and its inventive concept, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A method for intervening in cognitive dysfunction based on general artificial intelligence, characterized in that, It includes the following steps: S1. Real-time collect the patient's body data through monitoring devices and perform preprocessing. The body data includes physiological data, behavioral data, and cognitive state data. The preprocessing includes data denoising, normalization, and feature extraction; S2. Input the preprocessed body data into a dual convolutional neural network, extract data features from the time dimension and the space dimension respectively. The dual convolutional neural network contains two parallel convolutional modules, one of which is used to extract low-dimensional time-series data features, and the other is used to extract high-dimensional spatial data features. Combine the two types of data features through a shared convolutional layer and generate comprehensive cognitive state features; S3. Perform time-series clustering analysis on the generated comprehensive cognitive state features, use the adaptive clustering radius method to identify the key fluctuation periods of the patient's cognitive state, and predict the decline trend of the patient's cognitive ability based on the periodic changes. When the amplitude of the periodic changes exceeds the set threshold, trigger a multi-level cognitive impairment risk assessment mechanism and adjust the risk level; S4. During the patient's intervention, combine the dynamic emotion recognition algorithm based on micro-expression time-series mapping to analyze the patient's emotional state, use a custom emotion label mapping rule to convert the emotional state into digital parameters, and associate them with the cognitive impairment risk level to generate a dynamically updated personalized intervention cognitive plan; S5. According to the patient's actual response during the intervention process, use the incremental learning method to adjust the intervention cognitive plan in real time and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures; S6. Continuously optimize the dual convolutional neural network, use the patient's treatment feedback as training data, continuously optimize feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

2. The cognitive dysfunction intervention method based on general artificial intelligence according to claim 1, wherein, The body data includes physiological data, behavioral data, and cognitive state data. The preprocessing includes data denoising, normalization, and feature extraction.

3. The cognitive dysfunction intervention method based on general artificial intelligence according to claim 1, characterized in that, The specific content of S2 includes: S21. Receive the preprocessed patient body data and construct an input data matrix X = [X t , X s , where X consists of a time series data matrix and a spatial data matrix, X t is the time series data matrix, and X s is the spatial data matrix; S22. Input X t into the temporal convolutional module of the dual convolutional neural network, and use variable-stride multi-scale convolutional kernels to extract temporal features. Define the temporal feature extraction mapping function: Among them, F t (X t ) is the result of temporal feature mapping, k t is the number of temporal convolutional kernels, W t,i is the weight matrix of the i-th temporal convolutional kernel, is the data of the i-th temporal window, b t is the bias term, and σ is the non-linear activation function; The temporal convolutional kernel adopts a dynamic stride mechanism, and each W t,i adaptively adjusts the size of the convolutional window to adapt to the cognitive behavior characteristics of different time scales; S23. Input X s into the spatial convolution module of the dual convolutional neural network, extract spatial features using asymmetric receptive field multi-resolution convolution, and define the spatial feature mapping function: Among them, F s (X s ) is the spatial feature mapping result, k s is the number of spatial convolution kernels, W s,j is the weight matrix of the j-th spatial convolution kernel, is the j-th group of spatial feature window data, representing the feature data intercepted at spatial position j, b s is the bias term; The spatial convolution kernel adopts an asymmetric receptive field, that is, the convolution kernel sizes in different directions are different, so that the system can adapt to the changes in cognitive state features in different directions; S24. Through the weight-sharing adaptive fusion mechanism, fuse F t (X t ) and F s (X s ), and define the calculation of the fusion feature: F(X) = α t F t (X t ) + α s F s (X s ); Among them, F(X) is the final fused feature result, α t and α s are dynamically learned weight parameters, γ t and γ s are learnable parameters used to adjust the contribution ratio of temporal features and spatial features; S25. Perform normalization processing on the fused feature F(X) and perform dimensionality reduction to construct the final comprehensive cognitive state feature: Among them, is the normalized feature data, μ is the individual feature mean, and σ 2 is the feature variance of the patient group, ∈ is the smoothing factor to ensure the stability of the data distribution and avoid the influence of individual differences on the generalization ability of feature extraction. Z is the comprehensive cognitive state feature vector, and P is the personalized feature projection matrix to ensure that the features after dimensionality reduction can still represent the key cognitive state information of the patient.

4. The cognitive dysfunction intervention method based on general artificial intelligence according to claim 1, characterized in that The specific content of S3 includes: S31. Based on the comprehensive cognitive state feature vector Z, construct a time-series feature matrix: Q = [Z1, Z2, …, Z T T ;​ Among them, Q is the temporal feature matrix, representing the comprehensive cognitive state of the patient over T time steps, T is the length of the time step, and Z t is the comprehensive cognitive state feature vector at time step t, representing the physiological, behavioral, and cognitive state data of the patient; S32. Based on each time step t, calculate the weighted cognitive state change rate of the time-series feature matrix: Among them, R t is the weighted cognitive state change rate at time step t, d is the dimension of the temporal feature, that is, the number of features in the temporal feature matrix Q, and Q t,i is the i-th dimensional eigenvalue at time t, representing different cognitive state information of the patient, and Q t-1,i is the i-th dimensional eigenvalue at time t - 1, and ω i is the feature weight coefficient, α is the generalized distance norm exponent used to adjust the sensitivity between different feature change rates, and T is the time step length; S33. Based on the weighted cognitive state change rate, calculate the local density of each time step: Among them, ρ t is the local density at time step t, which is used to measure whether this time step belongs to the period of drastic fluctuations in the cognitive state. T is the time step length, and R j is the change rate of the cognitive state at time step j, and δ is the adaptive clustering radius, which is used to control the calculation range of the local density; Set the local density threshold λ, and identify the set of critical fluctuation periods C = {t|ρ t > λ}, where C is the set of critical fluctuation periods, which contains all time steps with local density exceeding the threshold, that is, the time points when the cognitive state changes drastically; S34. Based on the set of key fluctuation periods, calculate the decline trend of cognitive ability: S t = γS t-1 + (1 - γ)R t , S0 = 0, t ∈ C; Among them, S t is the cognitive ability decline trend index at time step t, used to predict the long-term change of cognitive state, and S t-1 is the decline trend index of the previous time step, and γ is the historical trend weight factor, used to balance the short-term and long-term trends; S35. When the fluctuation amplitude of the cognitive state exceeds the set threshold θ, adjust the cognitive impairment risk level: Among them, L is the adjusted risk level of cognitive impairment, L0 is the initial risk level, and β is the risk adjustment factor, which is used to control the dynamic adjustment range of the risk level. and are the maximum and minimum values of the trend of cognitive ability decline respectively, which are used to measure the periodic fluctuation range of the patient's cognitive state.

5. The cognitive dysfunction intervention method based on general artificial intelligence according to claim 1, characterized in that, The specific content of S4 includes: S41. Based on the time-series feature matrix Q, use the multi-layer adaptive trajectory mapping method to map the time-series features to the micro-expression feature matrix: Among them, M t is the micro-expression feature vector at time step t, and M t-j is the micro-expression feature vector at time step t - j, representing the facial state in the past j time steps. H is the length of the historical time series window. is the autoregressive coefficient, indicating the influence of the historical micro-expression state M t-j on the current state. φ i is the time series feature contribution coefficient, d is the dimension of the time series feature, that is, the number of features in the time series feature matrix Q. Q t,i is the value of the i-th dimension of the time series feature at time step t, k is the number of micro-expression key points, and θ l is the trajectory change weight, measuring the movement amplitude of facial muscles. ∈ t is the random perturbation term, used to model the influence of unobserved factors on micro-expressions. T is the time step length, and ΔP t,l is the relative movement trajectory of the l-th micro-expression key point, and ΔP t,l = P t,l - P t-1,l , where P t,l is the coordinate of the l-th micro-expression key point at time step t, and P t-1,l is the coordinate of the l-th micro-expression key point at time step t - 1; S42. Based on the relative motion trajectory change, define the emotion change rate and use the emotion change rate as the non-linear motion change rate of the micro-expression feature at time step t: Among them, is the emotional change rate at time step t, b is the number of facial features, that is, the dimension of facial key points, M t,i is the i-th dimensional micro-expression feature vector at time step t, M t-1,i is the i-th dimensional micro-expression feature vector at time step t - 1, ∈ is a small value to prevent the denominator from being zero to ensure calculation stability, and β is a non-linear adjustment exponent to enhance the sensitivity to small emotional fluctuations; S43. Use a custom emotion label mapping rule to map the emotion change rate to an emotion label: Among them, E t is the emotion label score at time step t, measuring the possible current emotional state of the patient, C is the number of custom emotion categories, λ c is the confidence adjustment coefficient for category c, μ c is the mean of category c, representing the central value of this category, σ c is the standard deviation of category c, representing the range of variation of this category; S44. Calculate the emotional instability index using the information entropy measurement method based on the emotional label scores: Among them, is the emotional instability index at time step t, E t,c is the emotional label score corresponding to category c at time step t, log2 is the logarithmic function, and η is the micro-expression change adjustment coefficient, combining the contribution of micro-expression features to emotional fluctuations; S45. Based on the emotional instability index and the cognitive impairment risk level L, calculate personalized emotional intervention parameters: Among them, A t is the personalized emotion intervention parameter at time step t, Γ is the emotion intervention gain matrix that controls the intervention intensity of different emotional states, λ L is the cognitive impairment risk impact coefficient, is the cognitive impairment risk level under the emotion label at time step t, ξ is the emotion-cognition coupling regulation parameter that measures the coupling effect of emotional fluctuations and cognitive fluctuations, is the mean value of the emotion stability index, is the mean value of the cognitive ability decline trend, S t is the cognitive ability decline trend index at time step t; S46. According to the personalized emotional intervention parameters, adjust the personalized intervention cognitive plan in real time and update it dynamically to ensure the adaptability and pertinence of the intervention measures.

6. A cognitive dysfunction intervention system based on general artificial intelligence, which executes the method for intervening in cognitive dysfunction based on general artificial intelligence according to any one of claims 1 to 5, characterized in that, Including: A data collection and preprocessing module, which is used to collect the patient's physiological data, behavioral data, and cognitive state data in real time through monitoring devices, and perform denoising, normalization, and feature extraction on the collected data; A spatio-temporal feature extraction module, which is used to input the preprocessed body data into a dual convolutional neural network, extract low-dimensional time-series data features from the time dimension, extract high-dimensional spatial data features from the space dimension, and merge the two data features through a shared convolutional layer to generate comprehensive cognitive state features; A time-series clustering analysis module, which is used to perform time-series clustering analysis on the comprehensive cognitive state features, identify the key fluctuation periods of the patient's cognitive state using the adaptive clustering radius method, and predict the decline trend of the patient's cognitive ability based on the periodic changes. When the amplitude of the periodic changes exceeds the set threshold, trigger a multi-level cognitive impairment risk assessment mechanism and adjust the risk level; A dynamic emotion recognition and mapping module, which is used to analyze the patient's emotional state during the intervention period by combining a dynamic emotion recognition algorithm based on micro-expression time-series mapping, and convert the emotional state into digital parameters using a custom emotional label mapping rule, associate it with the cognitive impairment risk level, and generate a dynamically updated personalized intervention cognitive plan; A personalized intervention adjustment module, which is used to adjust the intervention cognitive plan in real time using the incremental learning method according to the patient's actual reaction during the intervention process, and feedback it to the patient himself to ensure the timeliness and pertinence of the intervention measures; A long-term learning and optimization module, which is used to continuously optimize the dual convolutional neural network based on the patient's treatment feedback, continuously optimize the feature extraction and intervention strategies, and finally achieve long-term personalized intervention effects.

Images