Learning disorder assessment and adaptive training system and method based on multi-dimensional behaviors

Abstract

The invention relates to the technical field of artificial intelligence and big data, and discloses a learning disorder assessment and adaptive training system and method based on multi-dimensional behaviors, and the system comprises a multi-modal assessment module, a path generation module, a nerve regulation and control module and an adaptive engine module. By fusing multi-modal data features of language behaviors, electroneurographic signals and task response behaviors and combining a cross-modal attention mechanism to dynamically construct a cognitive state recognition model, learning disorder typing accuracy is improved, multi-dimensional ability defects such as attention, memory, language understanding and execution control are quantified based on a three-dimensional cognitive map, and learning disorder typing accuracy is improved. It is ensured that the assessment result comprehensively covers the core dimension of cognitive impairment; transcranial direct current stimulation is synchronously triggered in the cognitive task starting stage, space-time precise matching of task state nerve regulation and brain region activation is achieved, stimulation target point coordinates and current intensity are dynamically adjusted according to real-time nerve feedback, and the neuroplasticity reconstruction efficiency is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and big data technology, specifically to a learning disability assessment and adaptive training system and method based on multi-dimensional behavior. Background Technology

[0002] Learning disabilities are a heterogeneous group of disorders characterized by an individual's inability to effectively acquire, retrieve, and use information. They are caused by congenital and acquired abnormalities in brain structure and function, have multiple etiologies, and are difficult to distinguish from poor academic performance caused by environmental factors.

[0003] In the current technological context, learning disability assessment and adaptive training systems and methods based on multi-dimensional behavior have shortcomings, mainly due to the limitations of existing technologies: traditional assessment methods rely on single-dimensional behavioral observation and static cognitive testing, which cannot integrate the dynamic interactive features of multimodal data such as language behavior, neural electrical signals and task response in real time, resulting in insufficient accuracy in cognitive state recognition; neural modulation methods are mostly applied to the resting state, which is disconnected from the task execution process, making it difficult to achieve accurate activation in a time-space synchronization, thus weakening the enhancement effect of neural plasticity.

[0004] Therefore, a learning disability assessment and adaptive training system and method based on multi-dimensional behavior is proposed to solve the above problems. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a learning disability assessment and adaptive training system and method based on multi-dimensional behavior, which solves the problems mentioned in the background.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a learning disability assessment and adaptive training system and method based on multi-dimensional behavior, wherein the method includes the following steps: S1. Collect users' bioelectrical signal data, voice interaction data, and cognitive task response data through wearable devices; S2. Perform multimodal feature fusion processing on the bioelectric signal data, voice interaction data, and cognitive task response data to generate cognitive state recognition data; S3. Based on the cognitive state identification data, perform learning impairment classification processing to generate target impairment type feature data; S4. Match a preset rehabilitation stage model with the target obstacle type feature data to generate personalized rehabilitation path data. The rehabilitation stage model includes a sensory activation stage, a cognitive reconstruction stage, and a strategy transfer stage. S5. During the task execution process, the neuromodulation device is triggered synchronously to perform task-state neuromodulation based on the personalized rehabilitation path data, generating neuroplasticity enhancement data; S6. Real-time acquisition of dynamic feedback indicator set during the training process, the dynamic feedback indicator set including multi-dimensional indicators of language interaction, neural activity and behavioral performance, and generation of dynamic feedback indicator set; S7. Adaptively adjust the task difficulty parameters and neural stimulation intensity based on the dynamic feedback index set to generate optimized rehabilitation path data.

[0007] Preferably, step S1 includes the following steps: S11. Collect electrophysiological signal data, including the power spectral density of the theta and gamma bands, through wearable devices, with a sampling frequency higher than 256Hz; S12. Collect speech behavior feature data through the voice interaction terminal, including speech flow breakpoint frequency, semantic coherence score and reaction time parameter. S13. Collect behavioral response data through the cognitive task terminal, including the completion accuracy and response latency of the N-back working memory task and the Stroop inhibition control task.

[0008] Preferably, step S2 includes the following steps: S21. A cross-modal attention mechanism is used to extract the correlation matrix between speech features and electrophysiological signal features, wherein the speech features are derived from speech interaction data, and the electrophysiological signal features include neural signals in the theta and gamma bands. S22. Input the correlation matrix into the XGBoost classifier to perform cognitive state classification and output the quantitative probability distribution of attention deficit, working memory impairment and executive function impairment. The following feature fusion formula is applied in S21: ; in This represents the attention weight of the i-th language feature vector to the j-th neural signal feature vector, and its range is 0-1. For query vector, For key vectors, Let be a similarity function and denote the vector dot product operation. The total number of eigenvectors; S23. Generate a three-dimensional cognitive radar map based on the quantized probability distribution. The radar map includes score values ​​for four dimensions: attention, memory, language comprehension, and executive control.

[0009] Preferably, step S4 includes the following steps: S41. When the target obstacle type feature data is attention deficit, configure visual tracking task and auditory filtering task in the perception activation phase. S42. When the target impairment type feature data is working memory impairment, configure image sequence recall task and digital spatial memory task in the cognitive reconstruction stage; S43. When the target obstacle type feature data is an executive function obstacle, configure a multi-step planning task and a social situation simulation task in the strategy migration stage.

[0010] Preferably, step S5 includes the following steps: S51. When the task is started, the transcranial direct current stimulation device (tDCS) is activated simultaneously, and the stimulation target is dynamically located to the left prefrontal cortex and parietal cortex according to the task type. S52. A millisecond-level response mechanism is used to apply gradient current pulses at key task nodes, with the stimulation duration synchronized with the task execution cycle. The current intensity adjustment formula is as follows: ; in This is the adjustment amount for tDCS stimulation intensity. The learning rate parameter is in the range of 0.1-0.5. The target task accuracy threshold. This is the real-time value of the current task's accuracy. S53. Based on real-time monitoring of hemoglobin concentration changes using fNIRS, the position of the tDCS electrode is dynamically adjusted to ensure that the stimulation localization error is ≤ ±3 mm.

[0011] Preferably, step S7 includes the following steps: S71. When the language fluency index drops below the threshold, reduce the lexical complexity level of the semantic association task. S72. When the coherence index of the theta band is lower than the baseline level, increase the nerve stimulation intensity by 0.2 mA and prolong the stimulation duration by 20%. S73. When the accuracy of the N-back task is >90% for three consecutive times, automatically increase the working memory load to level N+1.

[0012] Preferably, the method further includes: S8. Construct a three-tiered collaborative platform integrating hospitals, communities, and families, including: The hospital-side configuration path deduction engine generates an initial rehabilitation path based on a cognitive radar map; Deploy task relay terminals on the community side to synchronize training data in real time and monitor device status; The home terminal integrates a voiceprint activation module, embedding family members' voices into task prompts to improve compliance.

[0013] Preferably, the three-level collaborative platform in S8 achieves secure data interaction through blockchain technology, specifically including: S81. The original neural electrical signal data is encrypted using IPFS distributed storage; S82. Controlling the voiceprint access permissions of family member terminals through smart contracts, with a voiceprint feature extraction error ≤0.5%; S83. Enable zero-knowledge proof verification of data integrity during cross-agency data transfer.

[0014] Preferably, the system includes: The multimodal assessment module integrates a wearable EEG device, a voice sensor, and a task response terminal to collect language behavior data, neural electrical signal data, and task response behavior data. The path generation module includes a CrossAttention-XGBoost fusion analysis unit and a 3D cognitive map construction unit; The neuromodulation module is equipped with a task-triggered tDCS device and an fNIRS real-time positioning feedback unit; The adaptive engine module enables closed-loop adjustment of task difficulty parameters and neural stimulation intensity based on a dynamic feedback indicator set.

[0015] Preferably, the system further includes a four-terminal collaborative platform: The patient terminal is equipped with a VR task interface and biofeedback indicator lights, which display changes in the cognitive radar graph in real time. The doctor management platform supports remote adjustment of tDCS target coordinates and stimulation parameters; The family interaction terminal provides a voiceprint recording interface and a visual dashboard for task progress. The institution's management backend deploys a federated learning framework to achieve joint optimization of models across hospitals.

[0016] (III) Beneficial Effects Compared with existing technologies, this invention provides a learning disability assessment and adaptive training system and method based on multi-dimensional behavior, which has the following beneficial effects: 1. In this invention, when implementing learning disability assessment and training, multimodal data features of language behavior, neural electrical signals and task response behavior are integrated, and a cognitive state recognition model is dynamically constructed by combining cross-modal attention mechanisms. This improves the accuracy of learning disability classification, quantifies multidimensional ability deficiencies such as attention, memory, language comprehension and executive control based on a three-dimensional cognitive map, and ensures that the assessment results fully cover the core dimensions of cognitive dysfunction. This avoids the risk of misjudgment caused by traditional single-dimensional assessment and provides a scientific basis for personalized rehabilitation pathways.

[0017] 2. In this invention, during the execution of neuromodulation intervention, transcranial direct current stimulation is triggered synchronously during the cognitive task initiation phase to achieve precise spatiotemporal matching between task-state neuromodulation and brain region activation. The coordinates of the stimulation target point and the current intensity are dynamically adjusted based on real-time neural feedback to enhance the efficiency of neural plasticity remodeling. Combined with a millisecond-level response mechanism, gradient current pulses are applied at key task nodes to overcome the technical limitations of traditional resting-state stimulation being disconnected from cognitive processing, reduce the spatiotemporal error of neuromodulation, and ensure the dynamic adaptability of intervention intensity to individual neural state.

[0018] 3. In this invention, a three-stage progressive model of sensory activation, cognitive reconstruction, and strategy transfer is adopted when constructing the rehabilitation pathway. Based on a dynamic feedback index set, the task difficulty parameters and neural stimulation intensity are optimized in real time to form a closed-loop regulation mechanism of "assessment-intervention-reassessment". At the same time, rehabilitation resources are integrated through a three-level collaborative platform of hospital-community-family, patient compliance is improved by using a voiceprint incentive module, and the secure interaction and joint optimization of multimodal data are ensured by blockchain technology and federated learning framework. Ultimately, the bottlenecks of rigidity, resource fragmentation, and insufficient compliance in traditional rehabilitation pathways are broken, and full-process adaptive rehabilitation management is achieved. Attached Figure Description

[0019] Figure 1 The flowchart shows the learning disability assessment and adaptive training method based on multi-dimensional behavior of this invention. Figure 2 This is a diagram illustrating the architecture of the learning disability assessment and adaptive training system based on multi-dimensional behavior according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Specific embodiment: A learning disability assessment and adaptive training system and method based on multi-dimensional behavior, the method including the following steps: S1. Collect users' bioelectrical signal data, voice interaction data, and cognitive task response data through wearable devices; S2. Perform multimodal feature fusion processing on language behavior data, neural electrical signal data, and task response behavior data to generate cognitive state recognition data; S3. Based on cognitive state recognition data, perform learning disability classification processing to generate target disability type feature data; S4. Match the preset rehabilitation stage model with the target obstacle type characteristic data to generate personalized rehabilitation path data. The rehabilitation stage model includes the sensory activation stage, cognitive reconstruction stage and strategy transfer stage. S5. During task execution, the neuromodulation device is triggered synchronously to perform task-state neuromodulation based on personalized rehabilitation path data, generating neuroplasticity enhancement data; S6. Real-time acquisition of dynamic feedback indicator set during the training process. The dynamic feedback indicator set includes multi-dimensional indicators of language interaction, neural activity and behavioral performance, and generates dynamic feedback indicator set. S7. Adaptively adjust the task difficulty parameters and neural stimulation intensity based on the dynamic feedback indicator set to generate optimized rehabilitation path data.

[0022] S1 includes the following steps: S11. Collect electrophysiological signal data, including the power spectral density of the theta and gamma bands, through wearable devices, with a sampling frequency higher than 256Hz; S12. Collect speech behavior feature data through the voice interaction terminal, including speech flow breakpoint frequency, semantic coherence score and reaction time parameter. S13. Collect behavioral response data through the cognitive task terminal, including the completion accuracy and response latency of the N-back working memory task and the Stroop inhibition control task.

[0023] S2 includes the following steps: S21. A cross-modal attention mechanism is used to extract the correlation matrix between speech features and electrophysiological signal features. The speech features are derived from speech interaction data, and the electrophysiological signal features include neural signals in the theta and gamma bands. S22. Input the correlation matrix into the XGBoost classifier to perform cognitive state classification and output the quantitative probability distribution of attention deficit, working memory impairment and executive function impairment. In S21, the following feature fusion formula is applied: ; in This represents the attention weight of the i-th language feature vector to the j-th neural signal feature vector, and its range is 0-1. For query vector, For key vectors, Let be a similarity function and denote the vector dot product operation. The total number of eigenvectors; S23. Generate a three-dimensional cognitive radar chart based on the quantized probability distribution. The radar chart includes score values ​​for four dimensions: attention, memory, language comprehension, and executive control.

[0024] S4 includes the following steps: S41. When the target obstacle type feature data is attention deficit, configure visual tracking task and auditory filtering task in the perception activation phase. S42. When the target impairment type characteristic data is working memory impairment, configure image sequence recall task and digital spatial memory task in the cognitive reconstruction stage; S43. When the target obstacle type feature data is executive function impairment, configure multi-step planning tasks and social situation simulation tasks in the strategy transfer phase. The formula for allocating the difficulty of each stage of the task is as follows: ; in This represents the difficulty level of the task during the current rehabilitation phase. Basic difficulty This is a difficulty adjustment factor with a range of 0.1. 0.3, This is the cumulative score for the current stage. This is the maximum score threshold for the current stage.

[0025] S5 includes the following steps: S51. When the task is started, the transcranial direct current stimulation device (tDCS) is activated simultaneously, and the stimulation target is dynamically located to the left prefrontal cortex and parietal cortex according to the task type. S52. A millisecond-level response mechanism is used to apply gradient current pulses at key task nodes, with the stimulation duration synchronized with the task execution cycle. The current intensity adjustment formula is as follows: ; in This is the adjustment amount for tDCS stimulation intensity. The learning rate parameter is in the range of 0.1-0.5. The target task accuracy threshold. This is the real-time value of the current task's accuracy. S53. Based on real-time monitoring of hemoglobin concentration changes using fNIRS, the position of the tDCS electrode is dynamically adjusted to ensure that the stimulation localization error is ≤ ±3 mm.

[0026] S7 includes the following steps: S71. When the language fluency index drops below the threshold, reduce the lexical complexity level of the semantic association task. S72. When the coherence index of the theta band is lower than the baseline level, increase the nerve stimulation intensity by 0.2 mA and prolong the stimulation duration by 20%. S73. When the accuracy of the N-back task is >90% for three consecutive times, the working memory load is automatically increased to level N+1. The adaptive difficulty adjustment formula is: ; in Adjust the amount of time to increase the difficulty of the task. The learning rate parameter has a range of 0.05. 0.2, For the target accuracy, This represents the actual accuracy rate.

[0027] The method also includes: S8. Construct a three-tiered collaborative platform integrating hospitals, communities, and families, including: The hospital-side configuration path deduction engine generates an initial rehabilitation path based on a cognitive radar map; Deploy task relay terminals on the community side to synchronize training data in real time and monitor device status; The home terminal integrates a voiceprint activation module, embedding family members' voices into task prompts to improve compliance; The path deduction and decision formula is as follows: ; in Let P be the set of possible rehabilitation paths, representing the optimal rehabilitation pathway. Score the expected effect of the path. For path expectation compliance scoring, These are the weighting coefficients. ,default .

[0028] The S8 three-tier collaborative platform uses blockchain technology to achieve secure data interaction, specifically including: S81. The original neural electrical signal data is encrypted using IPFS distributed storage; S82. Controlling the voiceprint access permissions of family member terminals through smart contracts, with a voiceprint feature extraction error ≤0.5%; S83. Enable zero-knowledge proofs to verify data integrity during cross-agency data transfers; The fingerprint security verification formula is: ; in For voiceprint feature difference degree, The input is the family member's voiceprint feature vector. As a reference voiceprint feature vector, when It is then determined to be a legitimate call.

[0029] The system includes: The multimodal assessment module integrates a wearable EEG device, a voice sensor, and a task response terminal to collect language behavior data, neural electrical signal data, and task response behavior data. The path generation module includes a CrossAttention-XGBoost fusion analysis unit and a 3D cognitive map construction unit; The neuromodulation module is equipped with a task-triggered tDCS device and an fNIRS real-time positioning feedback unit; The adaptive engine module enables closed-loop adjustment of task difficulty parameters and neural stimulation intensity based on a dynamic feedback indicator set; The federated learning update formula is: ; in These are the parameters after the global model update. These are the local model parameters for the k-th hospital node. The federal learning rate is in the range of 0.01. 0.1, Let the gradient of the loss function at the k-th node be denoted as . The number of hospital nodes participating in federal learning.

[0030] The system also includes a four-terminal collaborative platform: The patient terminal is equipped with a VR task interface and biofeedback indicator lights, which display changes in the cognitive radar graph in real time. The doctor management platform supports remote adjustment of tDCS target coordinates and stimulation parameters; The family interaction terminal provides a voiceprint recording interface and a visual dashboard for task progress. The institution's management backend deploys a federated learning framework to achieve joint optimization of models across hospitals.

[0031] The operation steps of this system and method are as follows: Step 1: Dynamic Acquisition of Multimodal Behavioral Data By capturing the power spectrum features of theta / γ band neural electrical signals in real time using wearable EEG devices, and simultaneously collecting language behavior parameters such as speech breakpoint frequency and semantic coherence from a voice interaction terminal, and linking with a task response terminal to obtain completion accuracy and response latency data for N-back working memory tasks and Stroop inhibition control tasks, this step completes the spatiotemporal alignment of three types of heterogeneous data within a millisecond-level time window, constructing a three-dimensional dynamic monitoring matrix covering neural activity, language expression, and behavioral response, providing a holographic input source for cognitive state recognition.

[0032] Step 2: Cross-modal cognitive state fusion assessment A cross-modal attention mechanism is used to analyze the deep correlation between language spectral features and neural signals, and a cross-domain correlation matrix is ​​generated through adaptive weighted fusion. Furthermore, a machine learning classifier is used to probabilistically classify attention deficit, working memory impairment, and executive function impairment, and transform them into a three-dimensional cognitive radar map containing dimensions of attention, memory, language comprehension, and executive control, so as to achieve quantitative and visual localization of cognitive dysfunction.

[0033] Step 3: Dynamic Deduction of Staged Rehabilitation Pathways Based on the cognitive radar map mapping results, visual tracking and auditory filtering tasks are configured in the perception activation stage to awaken basic perception; image sequence recall and digital spatial memory tasks are deployed in the cognitive reconstruction stage to strengthen the core ability network; and multi-step planning and social situation simulation tasks are designed in the strategy transfer stage to promote ability transformation. The system dynamically adjusts the difficulty and load of the stage tasks according to real-time performance, forming a progressive rehabilitation chain of perception → cognition → transfer.

[0034] Step 4: Precise neural modulation in a task-oriented manner The transcranial direct current stimulation device is activated simultaneously at the moment the cognitive task is initiated, and automatically located to the target areas of the left prefrontal and parietal lobes according to the task type. Gradient current pulses are applied at key nodes of the task through a millisecond-level response mechanism, so that the stimulation duration is precisely synchronized with the task execution cycle. At the same time, the electrode position is dynamically calibrated based on real-time feedback from near-infrared spectroscopy to ensure that the spatiotemporal matching accuracy between neural stimulation and brain region activation is controlled within the sub-millimeter error range.

[0035] Step 5: Closed-loop adaptive optimization engine The system monitors dynamic indicators such as language fluency, neural band coherence, and task accuracy in real time. When an abnormal decline in language fluency is detected, the complexity of the semantic association task is automatically reduced. When the theta band coherence deviates from the baseline, the intensity of neural stimulation is increased and the duration of stimulation is prolonged. When the working memory task maintains a high accuracy rate, a load escalation mechanism is triggered. This process forms a closed-loop regulatory pathway of assessment → intervention → reassessment.

[0036] Step Six: Building a Multi-Level Collaborative Ecosystem A three-tiered platform linking hospitals, communities, and families is constructed: At the hospital level, an initial rehabilitation path is generated based on a cognitive graph; at the community level, training data and device status are synchronized through task relay terminals; and at the family level, a voiceprint activation module is embedded to convert relatives' voices into task prompts. The platform uses blockchain technology for encrypted storage of neural data and voiceprint access control, and employs a federated learning framework to achieve cross-institutional model joint optimization, ultimately forming a rehabilitation ecosystem that integrates medical resources, improves family compliance, and ensures secure data exchange.

[0037] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0038] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A learning disability assessment and adaptive training method based on multi-dimensional behavior, characterized by: The method includes the following steps: S1. Collect users' bioelectrical signal data, voice interaction data, and cognitive task response data through wearable devices; S2. Perform multimodal feature fusion processing on the bioelectric signal data, voice interaction data, and cognitive task response data to generate cognitive state recognition data; S3. Based on the cognitive state identification data, perform learning impairment classification processing to generate target impairment type feature data; S4. Match a preset rehabilitation stage model with the target obstacle type feature data to generate personalized rehabilitation path data. The rehabilitation stage model includes a sensory activation stage, a cognitive reconstruction stage, and a strategy transfer stage. S5. During task execution, the neuromodulation device is triggered synchronously to perform task-state neuromodulation based on the personalized rehabilitation path data, generating neuroplasticity enhancement data; S6. Real-time acquisition of dynamic feedback indicator set during the training process, the dynamic feedback indicator set including multi-dimensional indicators of language interaction, neural activity and behavioral performance, and generation of dynamic feedback indicator set; S7. Adaptively adjust the task difficulty parameters and neural stimulation intensity based on the dynamic feedback index set to generate optimized rehabilitation path data.

2. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: S1 includes the following steps: S11. Collect electrophysiological signal data, including the power spectral density of the theta and gamma bands, through wearable devices, with a sampling frequency higher than 256Hz; S12. Collect speech behavior feature data through the voice interaction terminal, including speech flow breakpoint frequency, semantic coherence score and reaction time parameter. S13. Collect behavioral response data through the cognitive task terminal, including the completion accuracy and response latency of the N-back working memory task and the Stroop inhibition control task.

3. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: S2 includes the following steps: S21. A cross-modal attention mechanism is used to extract the correlation matrix between speech features and electrophysiological signal features, wherein the speech features are derived from speech interaction data, and the electrophysiological signal features include neural signals in the theta and gamma bands. S22. Input the correlation matrix into the XGBoost classifier to perform cognitive state classification and output the quantitative probability distribution of attention deficit, working memory impairment and executive function impairment. In S21, the following feature fusion formula is applied: ； in This represents the attention weight of the i-th language feature vector to the j-th neural signal feature vector, and its range is 0-1. For query vector, For key vectors, Let be a similarity function and denote the vector dot product operation. The total number of eigenvectors; S23. Generate a three-dimensional cognitive radar map based on the quantized probability distribution. The radar map includes score values ​​for four dimensions: attention, memory, language comprehension, and executive control.

4. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: S4 includes the following steps: S41. When the target obstacle type feature data is attention deficit, configure visual tracking task and auditory filtering task in the perception activation phase. S42. When the target impairment type feature data is working memory impairment, configure image sequence recall task and digital spatial memory task in the cognitive reconstruction stage; S43. When the target obstacle type feature data is an executive function obstacle, configure a multi-step planning task and a social situation simulation task during the strategy migration phase.

5. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: S5 includes the following steps: S51. When the task is started, the transcranial direct current stimulation device (tDCS) is activated simultaneously, and the stimulation target is dynamically located to the left prefrontal cortex and parietal cortex according to the task type. S52. A millisecond-level response mechanism is used to apply gradient current pulses at key task nodes, with the stimulation duration synchronized with the task execution cycle. The current intensity adjustment formula is as follows: ； in This is the adjustment amount for tDCS stimulation intensity. The learning rate parameter is in the range of 0.1-0.

5. The target task accuracy threshold. This is the real-time value of the current task's accuracy. S53. Based on real-time monitoring of hemoglobin concentration changes using fNIRS, the position of the tDCS electrode is dynamically adjusted to ensure that the stimulation localization error is ≤ ±3 mm.

6. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: S7 includes the following steps: S71. When the language fluency index drops below the threshold, reduce the lexical complexity level of the semantic association task. S72. When the coherence index of the theta band is lower than the baseline level, increase the nerve stimulation intensity by 0.2 mA and prolong the stimulation duration by 20%. S73. When the accuracy of the N-back task is >90% for three consecutive times, automatically increase the working memory load to level N+1.

7. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 1, characterized in that: The method further includes: S8. Construct a three-tiered collaborative platform integrating hospitals, communities, and families, including: The hospital-side configuration path deduction engine generates an initial rehabilitation path based on a cognitive radar map; Deploy task relay terminals on the community side to synchronize training data in real time and monitor device status; The home terminal integrates a voiceprint activation module, embedding family members' voices into task prompts to improve compliance.

8. The learning disability assessment and adaptive training method based on multi-dimensional behavior according to claim 7, characterized in that: The three-tier collaborative platform in S8 achieves secure data interaction through blockchain technology, specifically including: S81. The original neural electrical signal data is encrypted using IPFS distributed storage; S82. Controlling the voiceprint access permissions of family member terminals through smart contracts, with a voiceprint feature extraction error ≤0.5%; S83. Enable zero-knowledge proof verification of data integrity during cross-agency data transfer.

9. A learning disability assessment and adaptive training system based on multi-dimensional behavior, characterized in that: The system includes: The multimodal assessment module integrates a wearable EEG device, a voice sensor, and a task response terminal to collect language behavior data, neural electrical signal data, and task response behavior data. The path generation module includes a CrossAttention-XGBoost fusion analysis unit and a 3D cognitive map construction unit; The neuromodulation module is equipped with a task-triggered tDCS device and an fNIRS real-time positioning feedback unit; The adaptive engine module enables closed-loop adjustment of task difficulty parameters and neural stimulation intensity based on a dynamic feedback indicator set.

10. The learning disability assessment and adaptive training system based on multi-dimensional behavior according to claim 9, characterized in that: The system also includes a four-terminal collaborative platform: The patient terminal is equipped with a VR task interface and biofeedback indicator lights, which display changes in the cognitive radar graph in real time. The doctor management platform supports remote adjustment of tDCS target coordinates and stimulation parameters; The family interaction terminal provides a voiceprint recording interface and a visual dashboard for task progress. The institution's management backend deploys a federated learning framework to achieve joint optimization of models across hospitals.

Images