A cognitive impairment training scheme determination method and system based on cognitive assessment

Abstract

This invention provides a method and system for determining a cognitive impairment training program based on cognitive assessment. The method includes collecting multimodal cognitive assessment data of the target; performing feature enhancement on eye-tracking assessment data to obtain enhanced features; extracting physical and cognitive features based on scene assessment data to obtain physical and cognitive features; classifying cognitive states based on enhanced features, physical features, cognitive features, and scale assessment data to obtain cognitive domain scores; and selecting the optimal program from an initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program. This invention achieves collaborative optimization scheduling of multiple patients, multiple tasks, and multiple stimulus parameters, and can dynamically allocate limited resources according to the degree of cognitive impairment and intervention potential of each patient, forming a globally optimal configuration in terms of resource allocation optimization, task priority ranking, and stimulus parameter linkage.

Description

Technical Field

[0001] This invention belongs to the technical field of medical devices and artificial intelligence, and specifically relates to a method and system for determining a cognitive impairment training program based on cognitive assessment. Background Technology

[0002] Cognitive impairment is an abnormality of higher neurological functions caused by various factors such as neurodegenerative diseases, stroke, and traumatic brain injury. It is mainly manifested as a decline in multiple dimensions of cognitive function, including memory, attention, executive function, and visuospatial ability, severely impacting patients' quality of life and social participation. Currently, the diagnosis and treatment of cognitive impairment still faces the following technical bottlenecks: The assessment dimensions are too limited to accurately quantify cognitive deficits: Traditional cognitive assessments rely on scales (such as MMSE and MoCA) and doctors' subjective judgments, lacking objective means to quantify patients' cognitive processing (such as attention allocation, visual scanning patterns, and information processing speed), making it difficult to achieve early and accurate identification in the stage of mild cognitive impairment.

[0003] Training programs are generalized and lack targeting and dynamic adaptation: existing cognitive training is mostly based on fixed paradigms and fails to dynamically adjust the difficulty according to the individual cognitive deficit characteristics of patients, real-time cognitive load and physiological feedback during training, thus limiting the training effect.

[0004] Electrical stimulation intervention is disconnected from cognitive training and lacks a synergistic closed loop: Although transcranial electrical stimulation (tDCS, tACS, etc.) has been proven to modulate neuroplasticity, most existing systems are independent devices, and their stimulation parameters cannot form a dynamic linkage with the patient's real-time cognitive state and training task progress, resulting in poor "training-stimulation" synergy and unstable intervention effects.

[0005] Insufficient depth of multimodal data fusion and lack of dynamic modeling of cognitive state: Although existing systems attempt to fuse multimodal data such as eye movement, EEG, and behavior, they mostly use simple splicing or shallow fusion, failing to simulate the temporal interaction between visual information and the physical stimulation of physiological signals (such as eye movement) and cognitive processing, resulting in insufficient modeling accuracy of the patient's true cognitive state. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method and system for determining cognitive impairment training programs based on cognitive assessment, which solves the technical problems in the prior art.

[0007] In a first aspect, the present invention provides the following technical solution: a method for determining a cognitive impairment training program based on cognitive assessment, comprising: Collect multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data, and eye-tracking assessment data; The eye-tracking assessment data is augmented to obtain enhanced features; Physical and cognitive features are extracted based on the scene evaluation data to obtain physical and cognitive features; Cognitive state classification is performed based on the enhanced features, physical features, cognitive features, and scale assessment data to obtain a cognitive domain score; An initial cognitive impairment training program library is determined based on the cognitive domain score. The initial cognitive impairment training program library includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. The optimal program is selected from the initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program.

[0008] Compared with existing technologies, the beneficial effects of this invention are as follows: In the cognitive assessment dimension, this invention breaks through the limitations of traditional scale assessments by integrating multimodal data from eye tracking, scene interaction, and scale assessments to construct an attention-based fusion model. This model strictly follows the temporal pattern of "physical stimulation preceding cognitive processing" in visual information processing, sequentially simulating the direct stimulation of eye movements by physical visual features and the feedback regulation of eye movements by cognitive features. This achieves precise quantification of micro-cognitive processes such as patient attention allocation and visual scanning patterns. Simultaneously, a feature enhancement method is introduced, extracting low-frequency stable patterns related to cognitive states through one-dimensional channel attention and discrete Fourier transform, significantly improving the early identification sensitivity of mild cognitive impairment. In terms of treatment plan generation and dynamic optimization, this invention constructs a fully closed-loop diagnostic system of "assessment-modeling-training-stimulation-optimization-feedback." This invention decomposes long-term treatment goals into actionable, phased objectives, addressing the issues of sparse rewards and long-term credit allocation. It employs a two-layer network architecture to ensure precise execution of the underlying strategy, overcoming monotonicity constraints through value function decomposition, allowing local disadvantages to be compensated for by achieving global long-term optimality, and enabling dynamic adaptive adjustment of training difficulty and stimulus parameters. In terms of collaboration, this invention aims to maximize global cumulative rewards, achieving collaborative optimization scheduling across multiple patients, tasks, and stimulus parameters. It can dynamically allocate limited resources based on each patient's cognitive impairment and intervention potential, forming a globally optimal configuration in resource allocation optimization, task priority ranking, and stimulus parameter linkage. It supports deployment in multiple scenarios, including hospitals, communities, and homes. The terminal interaction module adopts an age-friendly design, balancing the needs of precision medicine and primary care screening, demonstrating good scalability and clinical applicability.

[0009] Preferably, the step of performing feature enhancement on the eye-tracking assessment data to obtain enhanced features specifically includes: The eye movement assessment data is preprocessed and features are extracted to obtain eye movement feature data; The eye-tracking feature data were processed using global average pooling. The data from each channel is aggregated to obtain a global channel feature vector. : ; In the formula, This is a global average pooling process. Number of channels; Weights are generated from the global channel feature vector using a fully connected layer to obtain adjusted weights. : ; ; In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; Based on the adjusted weight Determine the adjustment of eye movement feature data : ; In the formula, This is a channel-by-channel dot product operation; The adjusted eye-tracking feature data is subjected to a discrete Fourier transform to obtain a frequency domain transform result. The low-frequency component in the frequency domain transform result is extracted, and the low-frequency component is subjected to an inverse Fourier transform to obtain the enhanced feature.

[0010] Preferably, the step of extracting physical and cognitive features based on the scene evaluation data to obtain physical and cognitive features includes: Extract cognitive assessment scene image frames from the scene assessment data, and calculate the image difference between two adjacent frames. : ; In the formula, The color partition index for the histogram. The first , The color histogram vector of the frame; Two adjacent frames with image differences less than a preset difference value are merged to obtain several initial keyframes; Calculate image similarity between initial keyframes : ; In the formula, These are the initial keyframes. The image mean, These are the initial keyframes. variance For the initial keyframe Covariance between These are the first and second constants, respectively. Initial keyframes with image similarity greater than a preset similarity are merged to obtain the final keyframes; Extract the physical features of the final keyframe, including edge direction features, texture features, shape features, and global features; The final keyframe is input into a pre-trained large language model and prompt words are set to obtain semantic description text; The semantic description text is segmented and stop word removed to obtain processed text. A pre-trained text feature extraction model is then used to extract features from the processed text to obtain text feature vectors. The text feature vectors are then subjected to dimensionality reduction to obtain cognitive features.

[0011] Preferably, the step of classifying cognitive states based on the enhanced features, the physical features, the cognitive features, and the scale assessment data to obtain a cognitive domain score includes: Calculate the physical characteristics For the enhanced features Influence characteristics : ; ; In the formula, This is a vector concatenation operation. To output the projection matrix, For the first One point of attention, For attention mechanisms, The first A person's attention As a query vector As a key vector, As the projection matrix corresponding to the value vector This is a multi-head attention mechanism; Based on the aforementioned influence characteristics Determine the characteristics of physical stimuli : ; Calculate the cognitive features in relation to the cognitive response features. : ; Based on the cognitive response characteristics Determine fusion features : ; The cognitive domain score is determined based on the fusion features and the scale assessment data.

[0012] Preferably, the step of determining the cognitive domain score based on the fusion features and the scale assessment data includes: The fused features are then subjected to residual connections and fused output to obtain the fused features. : ; In the formula, For layer normalization operation, It is a feedforward neural network consisting of two fully connected layers; The fused features are input into the classification layer for classification to output an initial cognitive score. : ; In the formula, These are the weight matrix and bias term of the classification layer, respectively; The initial cognitive score is weighted and fused with the scale assessment data to obtain the cognitive domain score.

[0013] Preferably, the step of selecting the optimal training scheme from the initial cognitive impairment training scheme library based on the improved reinforcement learning algorithm to output the optimal cognitive impairment training scheme includes: Acquire training data, and extract state space data, observation space data, and action space data from the training data; Define the first strategy The first strategy is executed every certain number of time steps to achieve the long-term goal. and the global state in the state space data. Input into the first strategy to obtain the sub-target. : ; Determine the first value function of the first strategy, and update the first strategy by maximizing the first value function: ; ; In the formula, For mathematical expectation, For the parameters of the first strategy Find the gradient. Let the objective function of the first strategy be... For the first value function, As a discount factor, For a moment The value corresponding to the first value function; Define the second strategy With the first network The second strategy is executed in the first network, and the second strategy is executed at each time step, taking local observations in the observation space. and the sub-target The input to the second strategy is used to obtain specific actions in the action space data. : ; Determine the second value function of the second strategy, and update the second strategy by maximizing the second value function: ; ; In the formula, For the parameters of the second strategy Find the gradient. Let the objective function of the second strategy be... For the second value function, The consistency constraint coefficient is... In order to actually achieve the sub-goals, For a moment The corresponding total training reward value; Calculate the total training reward value and total Q value : ; ; In the formula, To dynamically adjust the weights, For sub-target values, The cognitive domain scores are before and after training, respectively. As the reward coefficient, , They are time points Task completion rate, fatigue level This is an adjustment factor for the Q value; Define the second network Through the second network Decompose the total Q value into a reward function. With superior and inferior functions : ; ; Based on the aforementioned return function With the aforementioned superiority / inferiority function Determine the mixed Q value : ; ; In the formula, For weight calculation networks; The loss functions of the first network and the second network are determined based on the hybrid Q-value, and the first network and the second network are updated based on the loss functions: ; ; In the formula, This indicates that samples are collected from the experience pool. The mathematical expectation, for The next state, For the ideal mixed Q value of the second network output, This indicates that samples are collected from the experience pool. The mathematical expectation, The loss functions for the first network and the second network are respectively. These are the parameters of the second network; Based on the enhanced features, physical features, cognitive features, scale assessment data, cognitive domain scores, and the initial cognitive impairment training scheme library, the target global state and target observation data are determined. Based on the updated first strategy, second strategy, first network, second network, target global state, and target observation data, the optimal cognitive impairment training scheme is determined.

[0014] Secondly, the present invention provides the following technical solution: a system for determining a cognitive impairment training program based on cognitive assessment, the system comprising: The acquisition module is used to acquire multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data, and eye-tracking assessment data. An enhancement module is used to enhance the features of the eye-tracking assessment data to obtain enhanced features; The extraction module is used to extract physical and cognitive features based on the scene evaluation data to obtain physical and cognitive features; The classification module is used to classify cognitive states based on the enhanced features, the physical features, the cognitive features, and the scale assessment data to obtain a cognitive domain score. The filtering module is used to determine an initial cognitive impairment training program library based on the cognitive domain score. The initial cognitive impairment training program library includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. The optimal program is selected from the initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program.

[0015] Preferably, the enhancement module is used for: The eye movement assessment data is preprocessed and features are extracted to obtain eye movement feature data; The eye-tracking feature data were processed using global average pooling. The data from each channel is aggregated to obtain a global channel feature vector. : ; In the formula, This is a global average pooling process. Number of channels; Weights are generated from the global channel feature vector using a fully connected layer to obtain adjusted weights. : ; ; In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; Based on the adjusted weight Determine the adjustment of eye movement feature data : ; In the formula, This is a channel-by-channel dot product operation; The adjusted eye-tracking feature data is subjected to a discrete Fourier transform to obtain a frequency domain transform result. The low-frequency component in the frequency domain transform result is extracted, and the low-frequency component is subjected to an inverse Fourier transform to obtain the enhanced feature.

[0016] Thirdly, the present invention provides the following technical solution: a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for determining a cognitive impairment training program based on cognitive assessment.

[0017] Fourthly, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described method for determining a cognitive impairment training scheme based on cognitive assessment. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a method for determining a cognitive impairment training program based on cognitive assessment, provided in Embodiment 1 of the present invention; Figure 2 This is a structural block diagram of the cognitive impairment training scheme determination system based on cognitive assessment provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the hardware structure of a computer provided for another embodiment of the present invention.

[0020] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0021] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0022] Example 1 In Embodiment 1 of the present invention, as Figure 1 As shown, a method for determining a training program for cognitive impairment based on cognitive assessment includes: S1. Collect multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data, and eye-tracking assessment data; Specifically, the scale assessment data here are the patient's basic scores in each cognitive domain collected through electronic scales (MMSE, MoCA, etc.); the scenario assessment data are the patient's behavioral data such as completion time, number of errors, and decision path collected through gamified scenario tasks; and the eye movement assessment data are the patient's fixation trajectory, saccade amplitude, pupil diameter changes, blink frequency, and other parameters collected through eye movement tracking devices during the assessment task. S2. Perform feature enhancement on the eye movement assessment data to obtain enhanced features; Step S2 includes: S21. Perform data preprocessing and feature extraction on the eye movement assessment data to obtain eye movement feature data; Specifically, the raw eye-tracking data contains noise, missing values, and blink artifacts, and the following preprocessing is required: deleting sampling points with a confidence level of less than 80% from the eye tracker, filling short-term data gaps caused by brief occlusion (missing duration <100ms) using cubic spline interpolation, and removing abnormal data points with pupil diameters exceeding the physiological range of [1mm, 8mm]. Subsequently, blinking events were identified based on the pupil diameter change rate: when the pupil diameter decreased by more than 30% within 5 consecutive sampling points and then quickly recovered, it was determined to be a blink; linear interpolation was used to fill in the data during the blinking period to preserve signal continuity; at the same time, the blinking frequency was recorded as an auxiliary feature to assess the patient's fatigue state.

[0023] Regarding the feature extraction part, using a 1-second time window (with a step size of 0.5 seconds to achieve a 50% overlapping sliding window), the following 24 temporal features are extracted, including: Fixation characteristics: fixation count (total number of fixations within the window), fixation speed: maximum, minimum, average, fixation duration: maximum, average; Salivation characteristics: number of saccades, saccade speed: minimum value, average value; Pupil characteristics: Left pupil diameter: maximum, minimum, mean, standard deviation, variance; Right pupil diameter: maximum, minimum, mean, standard deviation, variance; Mean pupil diameter: maximum, minimum, mean, standard deviation, variance; The features are then normalized to obtain the eye-tracking feature data.

[0024] S22. Utilize global average pooling to process the eye-tracking feature data. The data from each channel is aggregated to obtain a global channel feature vector. : ; In the formula, This is a global average pooling process. Number of channels; S23. Weights are generated on the global channel feature vector through a fully connected layer to obtain adjusted weights. : ; ; In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; S24. Based on the adjusted weights Determine the adjustment of eye movement feature data : ; In the formula, This is a channel-by-channel dot product operation; Specifically, in the above content, the channel attention mechanism enables the model to dynamically focus on eye movement feature channels (such as pupil diameter change rate, fixation time, etc.) that contribute more to cognitive state discrimination, and suppress redundant channel interference.

[0025] S25. Perform a discrete Fourier transform on the adjusted eye movement feature data to obtain a frequency domain transform result, extract the low-frequency component in the frequency domain transform result and perform an inverse Fourier transform on the low-frequency component to obtain the enhanced feature. Specifically, cognitive and emotion-related eye movement changes are usually characterized by low-frequency dominant patterns (such as slow pupil dilation and long-term fluctuations in gaze stability). Therefore, preserving low-frequency components and suppressing high-frequency noise, frequency domain enhancement can effectively extract stable emotional and cognitive patterns across time scales in eye movement signals, while filtering out high-frequency physiological noise (such as microsaccades and sampling noise) to improve feature robustness.

[0026] S3. Based on the scene evaluation data, extract physical and cognitive features to obtain physical and cognitive features; Specifically, when the target is receiving a cognitive task, the video / scene content exerts a dual regulation on eye movements through "physical stimulation" and "cognitive processing," extracting physical and cognitive features respectively.

[0027] Step S3 includes: S31. Extract cognitive assessment scene image frames from the scene assessment data and calculate the image difference between two adjacent frames. : ; In the formula, The color partition index for the histogram. The first , The color histogram vector of the frame.

[0028] S32. Merge two adjacent frames whose image difference is less than a preset difference value to obtain several initial keyframes.

[0029] S33. Calculate the image similarity between the initial keyframes. : ; In the formula, These are the initial keyframes. The image mean, These are the initial keyframes. variance For the initial keyframe Covariance between These are the first and second constants, respectively.

[0030] S34. Merge the initial keyframes whose image similarity is greater than the preset similarity to obtain the final keyframe; S35. Extract the physical features of the final keyframe, including edge direction features, texture features, shape features, and global features; Specifically, the edge direction features are HOG features, the texture features are LBP, GLCM (contrast, energy, correlation, homogeneity), and Gabor (4-direction × 2-scale) features, the shape features are Hu moment features, and the global features are GIST features.

[0031] S36. Input the final keyframe into the pre-trained large language model and set prompt words to obtain semantic description text; Specifically, the large language model here is the DeepSeek model.

[0032] S37. The semantic description text is segmented and stop word removed to obtain processed text. The pre-trained text feature extraction model is then used to extract features from the processed text to obtain text feature vectors. The text feature vectors are then subjected to dimensionality reduction to obtain cognitive features. Specifically, text features are extracted using the XLM-Roberta-base model, and then the PCA dimensionality reduction method is used to reduce the dimensionality of the text feature vectors to obtain cognitive features. Cognitive features simulate the top-down feedback signal of the anterior frontal cortex after semantic processing of visual information, which complements the physical features.

[0033] S4. Based on the enhanced features, the physical features, the cognitive features, and the scale assessment data, perform cognitive state classification to obtain a cognitive domain score; Specifically, step S4 includes: S41. Calculate the physical characteristics. For the enhanced features Influence characteristics : ; ; In the formula, This is a vector concatenation operation. To output the projection matrix, For the first One point of attention, For attention mechanisms, The first A person's attention As a query vector As a key vector, As the projection matrix corresponding to the value vector This is a multi-head attention mechanism.

[0034] S42, Based on the aforementioned influence characteristics Determine the characteristics of physical stimuli : .

[0035] Specifically, regarding the physical stimulus characteristics, it simulates the process by which the primary visual cortex processes physical information such as edges, colors, and brightness, and then directly stimulates the oculomotor nerve through the superior colliculus, reflecting a rapid perception-driven pathway from the bottom up.

[0036] S43. Calculate the cognitive feature in response to the cognitive response feature. : ; S44. Based on the cognitive response characteristics Determine fusion features : ; Specifically, regarding fusion features, the process of processing semantic information cognitively in higher visual brain regions (such as the anterior frontal cortex) and then regulating eye movement signals through feedback pathways embodies a "top-down" cognitive regulation mechanism. The two-layer sequence strictly follows the temporal rule of "physical stimulation precedes cognitive processing" in cognitive neuroscience.

[0037] S45. Determine the cognitive domain score based on the fusion features and the scale assessment data; Step S45 includes: S451. Perform residual connection and fusion output on the fused features to obtain the fused features. : ; In the formula, For layer normalization operation, It is a feedforward neural network consisting of two fully connected layers.

[0038] S452. Input the fused features into the classification layer for classification to output an initial cognitive score. : ; In the formula, These are the weight matrix and bias term of the classification layer, respectively.

[0039] S453. The initial cognitive score is weighted and fused with the scale assessment data to obtain the cognitive domain score; Specifically, the initial cognitive score corresponds to the quantitative scores of six cognitive domains: attention, memory, executive function, processing speed, visuospatial and social cognition. Similarly, the scale assessment data is also quantified by combining the initial cognitive score and the scale assessment data with a weight ratio of 6:4 to obtain the cognitive domain score.

[0040] S5. Based on the cognitive domain score, determine the initial cognitive impairment training program library, which includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. Based on the improved reinforcement learning algorithm, select the optimal program from the initial cognitive impairment training program library to output the optimal cognitive impairment training program. Specifically, the cognitive training program includes: a training task matching (working memory training, planning training, cognitive flexibility training, and social cognitive training) from a training resource library based on the patient's impaired cognitive domain, and setting the training frequency, duration, and difficulty level; an eye-tracking training program designing personalized eye-tracking training tasks (fixation stability training, saccadic accuracy training, and visual search efficiency training) to address issues such as poor fixation stability and insufficient saccadic accuracy identified in eye-tracking assessments; a VR training program matching immersive VR training scenarios (virtual supermarket, home organization, street navigation, etc.) based on the patient's interests and cognitive needs; and an electrical stimulation program setting the stimulation mode (tDCS, tACS, tPCS, tRNS, CES), stimulation target, current intensity (0.5~2mA), frequency (1~200Hz), and stimulation duration.

[0041] Step S5 includes: S51. Obtain training data, and extract state space data, observation space data and action space data from the training data; Specifically, state-space data includes cognitive scores, training completion rate, electrical stimulation response, and equipment resource usage; observation-space data includes eye movement characteristics, training performance, and physiological feedback; and action-space data includes training task selection and electrical stimulation parameter adjustment.

[0042] S52, Define the First Strategy The first strategy is executed every certain number of time steps to achieve the long-term goal. and the global state in the state space data. Input into the first strategy to obtain the sub-target. : ; Specifically, for the first strategy, it can break down long-term goals (such as "improving by 15% in 3 months") into phased, actionable sub-goals (such as "improving memory by 3 points this week"), while providing intermediate rewards to solve the problem of sparse rewards, evaluating the long-term value of different sub-goals, and selecting the optimal direction.

[0043] S53. Determine the first value function of the first strategy, and update the first strategy by maximizing the first value function: ; ; In the formula, For mathematical expectation, For the parameters of the first strategy Find the gradient. Let the objective function of the first strategy be... For the first value function, As a discount factor, For a moment The value corresponding to the first value function; The discount factor is 0.99. For the update of the first strategy, its purpose is to learn how to select sub-objectives, which can be expressed as "in what direction should the strategy be optimized".

[0044] S54, Define the second strategy With the first network The second strategy is executed in the first network, and the second strategy is executed at each time step, taking local observations in the observation space. and the sub-target The input to the second strategy is used to obtain specific actions in the action space data. : ; Specifically, the second strategy is used to ensure that the execution of actions does not deviate from the direction of the sub-goals given by the higher level, maximize immediate rewards within the framework of the sub-goals, dynamically adjust actions according to the real-time status of the goals, and immediately adjust actions when cognitive overload or training fatigue is detected.

[0045] S55. Determine the second value function of the second strategy, and update the second strategy by maximizing the second value function: ; ; In the formula, For the parameters of the second strategy Find the gradient. Let the objective function of the second strategy be... For the second value function, The consistency constraint coefficient is... In order to actually achieve the sub-goals, For a moment The corresponding total training reward value; Specifically, the purpose of the second strategy update is to learn how to perform specific actions (action execution), and the second strategy is specifically executed by the first network, which is an Actor network. This first network is used to map the real-time state of the target to specific diagnostic and treatment actions, thereby achieving personalized and precise intervention. S56. Calculate the total training reward value. and total Q value : ; ; In the formula, To dynamically adjust the weights, For sub-target values, The cognitive domain scores are before and after training, respectively. As the reward coefficient, , They are time points Task completion rate, fatigue level This is an adjustment factor for the Q value; Specifically, the reward value here is the global reward at time t, and the Q value is also the global Q value at time t. The score is 0.6. Task completion can be determined by the percentage of training task completed, while fatigue is calculated based on the fluctuation of pupil diameter and blink frequency. ,in The warm-up time step is 10,000 steps in this application.

[0046] S57, Define the second network Through the second network Decompose the total Q value into a reward function. With superior and inferior functions : ; ; Specifically, the second network, the Critic network, employs a value function decomposition structure, which decomposes the state-action value into a reward function and a merit-disadvantage function. This achieves a dynamic correlation between state rewards and action merit-disadvantage. The reward function reflects the "baseline value" of the current state, which is the expected cumulative reward that can be obtained by performing the optimal action in a certain state. The merit-disadvantage function reflects the "additional value" of choosing a certain action relative to the optimal action. A positive value indicates that the action is better than the average level, while a negative value indicates that it is worse than the average level.

[0047] S58. Based on the aforementioned reward function With the aforementioned superiority / inferiority function Determine the mixed Q value : ; ; In the formula, This is a weighted calculation network.

[0048] S59. Determine the loss functions of the first network and the second network based on the hybrid Q-value, and update the first network and the second network based on the loss functions: ; ; In the formula, This indicates that samples are collected from the experience pool. The mathematical expectation, for The next state, For the ideal mixed Q value of the second network output, This indicates that samples are collected from the experience pool. The mathematical expectation, The loss functions for the first network and the second network are respectively. These are the parameters of the second network; Specifically, in this step, the first and second networks are updated alternately. This allows the first network to learn to output the optimal action, while the second network learns to accurately evaluate the value of actions. By determining the loss function, the Q-value of the second network's output is made to approximate the ideal value defined by the Bellman optimality equation. When this error is minimized, the second network can accurately evaluate the long-term value of actions. The first network is trained to "output actions that yield high Q-values." If an action has a high Q-value, the gradient of the loss function encourages the first network to increase the output probability of that action; conversely, it decreases it. In this way, the first network gradually learns to select the optimal action.

[0049] S510. Based on the enhanced features, the physical features, the cognitive features, the scale assessment data, the cognitive domain score, and the initial cognitive impairment training scheme library, determine the target global state and target observation data. Based on the updated first strategy, second strategy, first network, second network, the target global state, and the target observation data, determine the optimal cognitive impairment training scheme. Specifically, after obtaining the updated two networks and the policy, the above steps are repeated by replacing the training data with the target global state and target observation data to output the optimal cognitive impairment training scheme, as follows: The first strategy selects the optimal sub-goal based on the global state of the current target, reassessing the rationality of the sub-goal every 50 time steps. If the patient's progress is slow or their state changes, the sub-goal direction is automatically switched. The second strategy uses a pre-trained second network and outputs the best action based on the observation data of the current target and the optimal sub-goal. When the following situations are detected, the first network automatically adjusts the action (because the network has learned to output appropriate actions in similar states): Cognitive overload: significant decrease in eye movement saccade speed, increased pupil diameter fluctuation → reduce training difficulty, reduce electrical stimulation intensity; Training fatigue: decreased task completion rate, prolonged reaction time → switch training type, increase rest interval; Slow progress: no significant improvement in cognitive score for multiple consecutive cycles → adjust stimulation mode (e.g., switch from tDCS to tACS), increase training frequency. In fact, for the second strategy of this application, the mixed Q value output can be used not only for action selection, but also for the following aspects: ranking of the merits of different options: comparing the mixed Q values ​​of different actions to quantitatively evaluate the expected efficacy of each candidate option; efficacy prediction: the Q value itself can be regarded as a quantitative predictor of long-term cognitive improvement of patients; decision interpretation: showing medical staff the expected benefits of the current option and enhancing clinical interpretability.

[0050] The cognitive assessment-based training program determination method for cognitive impairment provided in Embodiment 1 of this invention overcomes the limitations of traditional single-scale assessments in the cognitive assessment dimension. By integrating multimodal data from eye-tracking, scene interaction, and scale assessment, it constructs an attention-based fusion model. This model strictly follows the temporal pattern of "physical stimulation preceding cognitive processing" in visual information processing, sequentially simulating the direct stimulation of eye movements by physical visual features and the feedback regulation of eye movements by cognitive features. This achieves precise quantification of micro-cognitive processes such as patient attention allocation and visual scanning patterns. Simultaneously, it introduces feature enhancement methods, extracting low-frequency stable patterns related to cognitive states through one-dimensional channel attention and discrete Fourier transform, significantly improving the early identification sensitivity for mild cognitive impairment. Regarding treatment program generation and dynamic optimization, this invention constructs an "assessment-modeling-training-stimulation-optimization-reflection" model. This invention employs a closed-loop diagnostic and treatment chain that breaks down long-term treatment goals into actionable, phased objectives, addressing the issues of sparse rewards and long-term credit allocation. A two-layer network architecture ensures precise execution of the underlying strategy, and value function decomposition overcomes monotonicity constraints, allowing local disadvantages to be traded for global long-term optimality. This enables dynamic adaptive adjustment of training difficulty and stimulus parameters. In terms of collaboration, the invention aims to maximize global cumulative rewards, achieving collaborative optimization scheduling across multiple patients, tasks, and stimulus parameters. It can dynamically allocate limited resources based on each patient's cognitive impairment and intervention potential, forming a globally optimal configuration in resource allocation optimization, task priority ranking, and stimulus parameter linkage. It supports deployment in multiple scenarios, including hospitals, communities, and homes. The terminal interaction module adopts an age-friendly design, balancing precision diagnosis and treatment with grassroots screening needs, demonstrating good scalability and clinical applicability.

[0051] Example 2 like Figure 2 As shown, in Embodiment 2 of the present invention, a system for determining a cognitive impairment training program based on cognitive assessment is provided. The system includes: The acquisition module 1 is used to acquire multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data and eye movement assessment data. Enhancement module 2 is used to enhance the features of the eye-tracking assessment data to obtain enhanced features; Extraction module 3 is used to extract physical and cognitive features based on the scene evaluation data to obtain physical and cognitive features; Classification module 4 is used to classify cognitive states based on the enhanced features, the physical features, the cognitive features, and the scale assessment data to obtain a cognitive domain score; The screening module 5 is used to determine an initial cognitive impairment training program library based on the cognitive domain score. The initial cognitive impairment training program library includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. The optimal program is screened from the initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program.

[0052] The enhancement module 2 is used for: The eye movement assessment data is preprocessed and features are extracted to obtain eye movement feature data; The eye-tracking feature data were processed using global average pooling. The data from each channel is aggregated to obtain a global channel feature vector. : ; In the formula, This is a global average pooling process. Number of channels; Weights are generated from the global channel feature vector using a fully connected layer to obtain adjusted weights. : ; ; In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; Based on the adjusted weight Determine the adjustment of eye movement feature data : ; In the formula, This is a channel-by-channel dot product operation; The adjusted eye-tracking feature data is subjected to a discrete Fourier transform to obtain a frequency domain transform result. The low-frequency component in the frequency domain transform result is extracted, and the low-frequency component is subjected to an inverse Fourier transform to obtain the enhanced feature.

[0053] The extraction module 3 is used for: Extract cognitive assessment scene image frames from the scene assessment data, and calculate the image difference between two adjacent frames. : ; In the formula, The color partition index for the histogram. The first , The color histogram vector of the frame; Two adjacent frames with image differences less than a preset difference value are merged to obtain several initial keyframes; Calculate image similarity between initial keyframes : ; In the formula, These are the initial keyframes. The image mean, These are the initial keyframes. variance For the initial keyframe Covariance between These are the first and second constants, respectively. Initial keyframes with image similarity greater than a preset similarity are merged to obtain the final keyframes; Extract the physical features of the final keyframe, including edge direction features, texture features, shape features, and global features; The final keyframe is input into a pre-trained large language model and prompt words are set to obtain semantic description text; The semantic description text is segmented and stop word removed to obtain processed text. A pre-trained text feature extraction model is then used to extract features from the processed text to obtain text feature vectors. The text feature vectors are then subjected to dimensionality reduction to obtain cognitive features.

[0054] The classification module 4 is used for: Calculate the physical characteristics For the enhanced features Influence characteristics : ; ; In the formula, This is a vector concatenation operation. To output the projection matrix, For the first One point of attention, For attention mechanisms, The first A person's attention As a query vector As a key vector, As the projection matrix corresponding to the value vector This is a multi-head attention mechanism; Based on the aforementioned influence characteristics Determine the characteristics of physical stimuli : ; Calculate the cognitive features in relation to the cognitive response features. : ; Based on the cognitive response characteristics Determine fusion features : ; The cognitive domain score is determined based on the fusion features and the scale assessment data.

[0055] The classification module 4 is further used for: The fused features are then subjected to residual connections and fused output to obtain the fused features. : ; In the formula, For layer normalization operation, It is a feedforward neural network consisting of two fully connected layers; The fused features are input into the classification layer for classification to output an initial cognitive score. : ; In the formula, These are the weight matrix and bias term of the classification layer, respectively; The initial cognitive score is weighted and fused with the scale assessment data to obtain the cognitive domain score.

[0056] The filtering module 5 is used for: Acquire training data, and extract state space data, observation space data, and action space data from the training data; Define the first strategy The first strategy is executed every certain number of time steps to achieve the long-term goal. and the global state in the state space data. Input into the first strategy to obtain the sub-target. : ; Determine the first value function of the first strategy, and update the first strategy by maximizing the first value function: ; ; In the formula, For mathematical expectation, For the parameters of the first strategy Find the gradient. Let the objective function of the first strategy be... For the first value function, As a discount factor, For a moment The value corresponding to the first value function; Define the second strategy With the first network The second strategy is executed in the first network, and the second strategy is executed at each time step, taking local observations in the observation space. and the sub-target The input to the second strategy is used to obtain specific actions in the action space data. : ; Determine the second value function of the second strategy, and update the second strategy by maximizing the second value function: ; ; In the formula, For the parameters of the second strategy Find the gradient. Let the objective function of the second strategy be... For the second value function, The consistency constraint coefficient is... In order to actually achieve the sub-goals, For a moment The corresponding total training reward value; Calculate the total training reward value and total Q value : ; ; In the formula, To dynamically adjust the weights, For sub-target values, The cognitive domain scores are before and after training, respectively. As the reward coefficient, , They are time points Task completion rate, fatigue level This is an adjustment factor for the Q value; Define the second network Through the second network Decompose the total Q value into a reward function. With superior and inferior functions : ; ; Based on the aforementioned return function With the aforementioned superiority / inferiority function Determine the mixed Q value : ; ; In the formula, For weight calculation networks; The loss functions of the first network and the second network are determined based on the hybrid Q-value, and the first network and the second network are updated based on the loss functions: ; ; In the formula, This indicates that samples are collected from the experience pool. The mathematical expectation, for The next state, For the ideal mixed Q value of the second network output, This indicates that samples are collected from the experience pool. The mathematical expectation, The loss functions for the first network and the second network are respectively. These are the parameters of the second network; Based on the enhanced features, physical features, cognitive features, scale assessment data, cognitive domain scores, and the initial cognitive impairment training scheme library, the target global state and target observation data are determined. Based on the updated first strategy, second strategy, first network, second network, target global state, and target observation data, the optimal cognitive impairment training scheme is determined.

[0057] In other embodiments of the present invention, the present invention provides the following technical solution: a computer, including a memory 102, a processor 101, and a computer program stored in the memory 102 and executable on the processor 101, wherein the processor 101 executes the computer program to implement the cognitive impairment training program determination method based on cognitive assessment as described above.

[0058] Specifically, the processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0059] The memory 102 may include a large-capacity memory for data or instructions. For example, and not limitingly, the memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 102 may include removable or non-removable (or fixed) media. Where appropriate, the memory 102 may be internal or external to a data processing device. In a particular embodiment, the memory 102 is non-volatile memory. In a particular embodiment, the memory 102 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0060] The memory 102 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 101.

[0061] The processor 101 reads and executes the computer program instructions stored in the memory 102 to implement the above-described method for determining a cognitive impairment training program based on cognitive assessment.

[0062] In some embodiments, the computer may further include a communication interface 103 and a bus 100. For example, Figure 3 As shown, the processor 101, memory 102, and communication interface 103 are connected through bus 100 and complete communication with each other.

[0063] The communication interface 103 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 103 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0064] Bus 100 includes hardware, software, or both, that couples components of a computer device together. Bus 100 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 100 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 100 may include one or more buses. Although specific buses are described and illustrated in the embodiments of the present invention, the present invention is contemplated by any suitable bus or interconnect.

[0065] The computer can execute the cognitive impairment training program determination method based on cognitive assessment of the present invention based on the cognitive assessment of the system for determining cognitive impairment training programs, thereby realizing the determination of cognitive impairment training programs based on cognitive assessment.

[0066] In some further embodiments of the present invention, in conjunction with the above-described method for determining a cognitive impairment training program based on cognitive assessment, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described method for determining a cognitive impairment training program based on cognitive assessment.

[0067] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0068] More specific examples of readable media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0069] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0071] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for determining a training program for cognitive impairment based on cognitive assessment, characterized in that, include: Collect multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data, and eye-tracking assessment data; The eye-tracking assessment data is augmented to obtain enhanced features; Physical and cognitive features are extracted based on the scene evaluation data to obtain physical and cognitive features; Cognitive state classification is performed based on the enhanced features, physical features, cognitive features, and scale assessment data to obtain a cognitive domain score; An initial cognitive impairment training program library is determined based on the cognitive domain score. The initial cognitive impairment training program library includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. The optimal program is selected from the initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program.

2. The method for determining a cognitive impairment training program based on cognitive assessment according to claim 1, characterized in that, The step of performing feature enhancement on the eye-tracking assessment data to obtain enhanced features specifically includes: The eye movement assessment data is preprocessed and features are extracted to obtain eye movement feature data; The eye-tracking feature data were processed using global average pooling. The data from each channel is aggregated to obtain a global channel feature vector. : ； In the formula, This is a global average pooling process. Number of channels; Weights are generated from the global channel feature vector using a fully connected layer to obtain adjusted weights. : ； ； In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; Based on the adjusted weight Determine the adjustment of eye movement feature data : ； In the formula, This is a channel-by-channel dot product operation; The adjusted eye-tracking feature data is subjected to a discrete Fourier transform to obtain a frequency domain transform result. The low-frequency component in the frequency domain transform result is extracted, and the low-frequency component is subjected to an inverse Fourier transform to obtain the enhanced feature.

3. The method for determining a cognitive impairment training program based on cognitive assessment according to claim 1, characterized in that, The steps of extracting physical and cognitive features based on the scene evaluation data to obtain physical and cognitive features include: Extract cognitive assessment scene image frames from the scene assessment data, and calculate the image difference between two adjacent frames. : ； In the formula, The color partition index for the histogram. The first , The color histogram vector of the frame; Two adjacent frames with image differences less than a preset difference value are merged to obtain several initial keyframes; Calculate image similarity between initial keyframes : ； In the formula, These are the initial keyframes. The image mean, These are the initial keyframes. variance For the initial keyframe Covariance between These are the first and second constants, respectively. Initial keyframes with image similarity greater than a preset similarity are merged to obtain the final keyframes; Extract the physical features of the final keyframe, including edge direction features, texture features, shape features, and global features; The final keyframe is input into a pre-trained large language model and prompt words are set to obtain semantic description text; The semantic description text is segmented and stop word removed to obtain processed text. A pre-trained text feature extraction model is then used to extract features from the processed text to obtain text feature vectors. The text feature vectors are then subjected to dimensionality reduction to obtain cognitive features.

4. The method for determining a cognitive impairment training program based on cognitive assessment according to claim 1, characterized in that, The step of classifying cognitive states based on the enhanced features, the physical features, the cognitive features, and the scale assessment data to obtain a cognitive domain score includes: Calculate the physical characteristics For the enhanced features Influence characteristics : ； ； In the formula, This is a vector concatenation operation. To output the projection matrix, For the first One point of attention, For attention mechanisms, The first A person's attention As a query vector As a key vector, As the projection matrix corresponding to the value vector This is a multi-head attention mechanism; Based on the aforementioned influence characteristics Determine the characteristics of physical stimuli : ； Calculate the cognitive features in relation to the cognitive response features. : ； Based on the cognitive response characteristics Determine fusion features : ； The cognitive domain score is determined based on the fusion features and the scale assessment data.

5. The method for determining a cognitive impairment training program based on cognitive assessment according to claim 4, characterized in that, The step of determining the cognitive domain score based on the fusion features and the scale assessment data includes: The fused features are then subjected to residual connections and fused output to obtain the fused features. : ； In the formula, For layer normalization operation, It is a feedforward neural network consisting of two fully connected layers; The fused features are input into the classification layer for classification to output an initial cognitive score. : ； In the formula, These are the weight matrix and bias term of the classification layer, respectively; The initial cognitive score is weighted and fused with the scale assessment data to obtain the cognitive domain score.

6. The method for determining a cognitive impairment training program based on cognitive assessment according to claim 1, characterized in that, The step of selecting the optimal training scheme from the initial cognitive impairment training scheme library based on the improved reinforcement learning algorithm to output the optimal cognitive impairment training scheme includes: Acquire training data, and extract state space data, observation space data, and action space data from the training data; Define the first strategy The first strategy is executed every certain number of time steps to achieve the long-term goal. and the global state in the state space data. Input into the first strategy to obtain the sub-target. : ； Determine the first value function of the first strategy, and update the first strategy by maximizing the first value function: ； ； In the formula, For mathematical expectation, For the parameters of the first strategy Find the gradient. Let the objective function of the first strategy be... For the first value function, As a discount factor, For a moment The value corresponding to the first value function; Define the second strategy With the first network The second strategy is executed in the first network, and the second strategy is executed at each time step, taking local observations in the observation space. and the sub-target The input to the second strategy is used to obtain specific actions in the action space data. : ； Determine the second value function of the second strategy, and update the second strategy by maximizing the second value function: ； ； In the formula, For the parameters of the second strategy Find the gradient. Let the objective function of the second strategy be... For the second value function, The consistency constraint coefficient is... In order to actually achieve the sub-goals, For a moment The corresponding total training reward value; Calculate the total training reward value and total Q value : ； ； In the formula, To dynamically adjust the weights, For sub-target values, The cognitive domain scores are before and after training, respectively. As the reward coefficient, , They are time points Task completion rate, fatigue level This is an adjustment factor for the Q value; Define the second network Through the second network Decompose the total Q value into a reward function. With superior and inferior functions : ； ； Based on the aforementioned return function With the aforementioned superiority / inferiority function Determine the mixed Q value : ； ； In the formula, For weight calculation networks; The loss functions of the first network and the second network are determined based on the hybrid Q-value, and the first network and the second network are updated based on the loss functions: ； ； In the formula, This indicates that samples are collected from the experience pool. The mathematical expectation, for The next state, For the ideal mixed Q value of the second network output, This indicates that samples are collected from the experience pool. The mathematical expectation, The loss functions for the first network and the second network are respectively. These are the parameters of the second network; Based on the enhanced features, physical features, cognitive features, scale assessment data, cognitive domain scores, and the initial cognitive impairment training scheme library, the target global state and target observation data are determined. Based on the updated first strategy, second strategy, first network, second network, target global state, and target observation data, the optimal cognitive impairment training scheme is determined.

7. A system for determining a training program for cognitive impairment based on cognitive assessment, characterized in that, The system includes: The acquisition module is used to acquire multimodal cognitive assessment data of the target, including scale assessment data, scene assessment data, and eye-tracking assessment data. An enhancement module is used to enhance the features of the eye-tracking assessment data to obtain enhanced features; The extraction module is used to extract physical and cognitive features based on the scene evaluation data to obtain physical and cognitive features; The classification module is used to classify cognitive states based on the enhanced features, the physical features, the cognitive features, and the scale assessment data to obtain a cognitive domain score. The filtering module is used to determine an initial cognitive impairment training program library based on the cognitive domain score. The initial cognitive impairment training program library includes cognitive training programs, VR training programs, eye-tracking training programs, and transcranial stimulation programs. The optimal program is selected from the initial cognitive impairment training program library based on an improved reinforcement learning algorithm to output the optimal cognitive impairment training program.

8. The cognitive impairment training program determination system based on cognitive assessment according to claim 7, characterized in that, The enhancement module is used for: The eye movement assessment data is preprocessed and features are extracted to obtain eye movement feature data; The eye-tracking feature data were processed using global average pooling. The data from each channel is aggregated to obtain a global channel feature vector. : ； In the formula, This is a global average pooling process. Number of channels; Weights are generated from the global channel feature vector using a fully connected layer to obtain adjusted weights. : ； ； In the formula, , These are the weight matrix and bias term corresponding to the first fully connected layer, respectively. , These are the weight matrix and bias term corresponding to the second fully connected layer, respectively. This is the intermediate feature vector; Based on the adjusted weight Determine the adjustment of eye movement feature data : ； In the formula, This is a channel-by-channel dot product operation; The adjusted eye-tracking feature data is subjected to a discrete Fourier transform to obtain a frequency domain transform result. The low-frequency component in the frequency domain transform result is extracted, and the low-frequency component is subjected to an inverse Fourier transform to obtain the enhanced feature.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for determining a cognitive impairment training program based on cognitive assessment as described in any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method for determining a cognitive impairment training program based on cognitive assessment as described in any one of claims 1 to 6.

Images