Human brain cognitive recovery training scheme analysis system and method for cognitive impairment

Abstract

The present application relates to the technical field of cognitive impairment rehabilitation, in particular to a brain cognitive recovery training scheme analysis system and method for cognitive impairment, comprising five units of basic information acquisition, cognitive test execution, test report generation, norm updating and training scheme analysis, wherein the norm database is modeled in multiple dimensions according to age, educational background and medical history, a regional correction factor is introduced, data freshness is verified by means of blockchain storage, an adaptive sequence is generated for cognitive test, multi-modal data such as correct answer rate, touch screen trajectory and eye movement data are collected, time domain signal separation and denoising are performed, the training scheme analysis unit locates cognitive defects, matches the training module, monitors the brain electrical theta wave and P300 latency through biological feedback, dynamically adjusts the training intensity and frequency, promotes neural circuit reorganization and improves rehabilitation accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cognitive impairment rehabilitation, in particular to a human brain cognitive recovery training scheme analysis system and method for cognitive impairment. BACKGROUND

[0002] Cognitive impairment rehabilitation is an important technology. Under the background of the current population aging and the rising incidence of cognitive impairment, this technology is the key support to break through the limitations of traditional cognitive rehabilitation. It can not only avoid the blindness of relying on experience to develop training schemes by scientifically evaluating and positioning specific cognitive deficits, but also rely on the principle of neural plasticity to design training content to promote the reorganization of damaged neural circuits. At the same time, it provides quantitative data of training effect for medical staff, adapts to the needs of clinical diagnosis and treatment and long-term rehabilitation management, and balances the accuracy of rehabilitation and patient compliance.

[0003] The existing cognitive impairment recovery training scheme analysis technology faces the core problems of insufficient individualization and difficulty in ensuring training effect in actual application, which cannot adapt to the cognitive characteristics and neural plasticity level of different patients. The norm database of traditional technology is mostly divided by age, without fully considering the differences in education background, medical history type and regional medical level, resulting in a large deviation between the cognitive index benchmark value and the actual cognitive level of the patient. For example, if the memory test benchmark of patients with different education backgrounds is uniformly set, it is easy to misjudge the cognitive impairment degree of patients with low education background. Cognitive tests mostly use fixed sequences and do not dynamically adjust the difficulty according to the real-time response of patients. Moreover, only single response data is collected, without synchronously recording the correct rate and eye movement trajectory multi-modal information, and it is also disturbed by environmental noise, making it difficult to accurately locate the defect type and associated impact area. After the training scheme is generated, there is no dynamic adjustment mechanism, and the neural response is not reflected by real-time monitoring of the brain electrical characteristics through biological feedback. Long-term fixed training intensity and frequency may lead to neural adaptation and cannot effectively stimulate neural plasticity. Moreover, the neural circuit reorganization effect is not verified through imaging means, making it difficult to iteratively optimize the scheme. This defect leads to inaccurate benchmark, resulting in distorted test data, and further deviation in defect positioning. The matched training module cannot repair the defect specifically, and lacks dynamic adjustment and effect verification. It may not only cause frustration and reduce compliance due to excessive intensity, but also may not promote neural reorganization due to insufficient intensity. In the long run, not only the rehabilitation effect is poor, but also the best rehabilitation opportunity of the patient may be missed. In order to solve this problem, the present application provides a human brain cognitive recovery training scheme analysis system and method for cognitive impairment. SUMMARY

[0004] The present application aims to provide a human brain cognitive recovery training scheme analysis system and method for cognitive impairment to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, a human brain cognitive recovery training scheme analysis system for cognitive impairment is provided, comprising:

[0006] The basic information acquisition unit is used to acquire the basic information of the test subjects and determine the corresponding norm database;

[0007] The cognitive testing execution unit performs cognitive tests and obtains raw test data based on the test subjects' basic information and corresponding norm data.

[0008] The test report generation unit generates a test report based on the raw test data;

[0009] The norm update unit obtains the test report and updates the norm parameters in the norm database according to the test report;

[0010] The training program analysis unit, based on test reports and norm parameters, analyzes the cognitive impairment characteristics of the test subjects and automatically generates personalized recovery training programs using a personalized recovery training algorithm. This analysis includes identifying deviations in cognitive indicators from the test reports to pinpoint specific deficit types. The personalized recovery training algorithm includes matching preset training module combinations based on specific deficit types and dynamically adjusting training intensity, frequency, and duration. It optimizes training modules to enhance neural plasticity and promotes neural circuit reorganization through repetitive stimulation and progressive challenges, achieving dynamic recovery of cognitive function. This includes real-time monitoring of training feedback and iterative updates to the program to ensure effective recovery.

[0011] The second objective of this invention is to provide a method for implementing an analysis system for cognitive rehabilitation training programs for people with cognitive impairment, including any one of the above-described methods, comprising the following steps:

[0012] S1. Obtain basic information of the test subjects, match the norm database based on the multi-dimensional hierarchical model, verify the freshness of the data through the blockchain distributed storage mechanism, and apply the regional dynamic correction factor to calibrate the benchmark value weight.

[0013] S2. Generate an adaptive test sequence based on the calibrated norm parameters, use a multimodal interactive device to synchronously collect the answer accuracy, touch screen trajectory and eye movement data, use time domain signal separation technology to remove environmental noise interference, and output a pure cognitive response data stream;

[0014] S3. Input the test data into the defect localization decision tree to identify specific defect types, construct a cross-dimensional correlation analysis matrix to generate a defect type topology map; predict the neural reorganization rate based on the synaptic plasticity prediction model, and initialize the training module combination, intensity ladder and cycle plan through the dynamic difficulty adjustment engine.

[0015] S4. During training, the EEG characteristics are monitored in real time through a biofeedback closed-loop system. The training intensity, frequency and duration are dynamically adjusted by the neural response-driven engine. Neural path labeling technology is used to strengthen synaptic labeling. The reorganization effect is verified by diffusion tensor imaging and the training program is iteratively updated.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0017] This invention constructs a norm database with multi-dimensional hierarchical modeling, introduces regional dynamic correction factors, and uses blockchain to verify data freshness. It generates adaptive test sequences and combines multimodal data acquisition and temporal signal separation and denoising technology to accurately locate cognitive deficits. It relies on a biofeedback closed-loop system to monitor EEG characteristics and dynamically adjusts training parameters through a neural response-driven engine. It also uses diffusion tensor imaging to verify the effect of neural circuit remodeling. This achieves the effects of accurate norm adaptation, accurate location of cognitive deficits, dynamic matching of training with the patient's neural plasticity, and quantifiable and verifiable results. It effectively solves the problems of existing cognitive impairment recovery training programs, such as insufficient personalization, difficulty in guaranteeing results, and inability to adapt to different patients' cognitive characteristics and neural plasticity levels. Attached Figure Description

[0018] Figure 1 This is an overall block diagram of the present invention;

[0019] Figure 2 This is the overall flowchart of the present invention.

[0020] The meanings of the labels in the diagram are as follows:

[0021] 1. Basic information acquisition unit; 2. Cognitive test execution unit; 3. Test report generation unit; 4. Norm update unit; 5. Training scheme analysis unit. Detailed Implementation

[0022] 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.

[0023] This invention provides an analysis system for cognitive recovery training programs for people with cognitive impairment. Please refer to [link / reference]. Figure 1 As shown, it includes:

[0024] Basic information acquisition unit 1 is used to acquire basic information of the test subjects and determine the corresponding norm database;

[0025] Cognitive test execution unit 2 executes cognitive tests and obtains raw test data based on the basic information of the test subjects and the corresponding norm data;

[0026] Test report generation unit 3 generates a test report based on the raw test data;

[0027] Norm update unit 4 obtains the test report and updates the norm parameters in the norm database according to the test report;

[0028] The training program analysis unit 5, based on the test report and norm parameters, analyzes the cognitive impairment characteristics of the test subjects and automatically generates personalized recovery training programs using a personalized recovery training algorithm. The analysis of the cognitive impairment characteristics of the test subjects includes identifying deviations in cognitive indicators in the test report to locate specific defect types. The personalized recovery training algorithm includes matching preset training module combinations based on specific defect types and dynamically adjusting the training intensity, frequency, and duration. It optimizes training modules to improve neural plasticity and promotes neural circuit reorganization through repetitive stimulation and progressive challenges to achieve dynamic recovery of cognitive function. This includes real-time monitoring of training feedback and iterative updates to the program to ensure recovery effectiveness.

[0029] The norm database is constructed using a multi-dimensional hierarchical modeling technique, specifically including:

[0030] The database is divided into basic dimensions based on age group, educational background, and medical history. Benchmark values ​​and standard deviations of cognitive indicators for the corresponding population are stored under each dimension. Regional dynamic correction factors are introduced to automatically adjust the weight of the benchmark values ​​based on the medical level data of the region where the test subjects are located. The database achieves cross-institutional data sharing through blockchain distributed storage. Data freshness verification is performed each time the norm parameters are called. If the difference between the current data version and the latest version on the cloud exceeds a set threshold, automatic synchronization is triggered.

[0031] The process of conducting cognitive tests and obtaining raw test data specifically includes:

[0032] An adaptive test sequence is generated based on norm parameters. Test items are dynamically switched according to the real-time response of the test subjects. A multimodal interactive acquisition device is used to synchronously record the answer accuracy, touch screen operation trajectory and eye-tracking heat map. An interference suppression algorithm is deployed. During the test, background noise is injected to simulate real environmental interference. Time domain signal separation technology is used to remove environmental noise components from the original data and extract a pure cognitive response data stream.

[0033] The application of personalized recovery training algorithms includes: establishing a knowledge base mapping defect types to training modules, mapping specific defect types to preset neural function reconstruction targets, initializing training intensity parameters through a dynamic difficulty adjustment engine, which calculates the initial challenge level based on the degree of deviation of each cognitive indicator in the test report, and using a synaptic plasticity prediction model to predict the rate of neural circuit reorganization based on historical EEG data, thereby generating an initial scheme framework that includes training module combinations, intensity ladders, and periodic plans.

[0034] The dynamic operation mechanism of the personalized recovery training algorithm specifically includes:

[0035] During the training execution phase, EEG data is collected in real time through a biofeedback closed-loop system. Wave power spectral density and event-related potential P300 latency, when detected When wave synchronization is enhanced and P300 latency is shortened, the adaptive reinforcement learning module is activated to generate an intensity increment strategy based on historical training effect data. If the expected neural response is not achieved after three consecutive training sessions, a cross-modal alternative training strategy is initiated to automatically convert the visual training task into an auditory channel to overcome the neural compensation bottleneck.

[0036] The specific biases in the cognitive metrics identified in the test report include:

[0037] The raw test data is input into the defect localization decision tree, which contains hierarchical discrimination nodes for memory dimension, executive function dimension and language understanding dimension, and constructs a cross-dimensional correlation analysis matrix. When the working memory index deviation exceeds the first-level threshold, the attention allocation index is automatically associated and detected as abnormal. Finally, a defect type topology map is output, marking the core area of ​​the main defects and the influence area of ​​related defects, providing a spatial localization basis for training module matching.

[0038] Dynamically adjusting training intensity, frequency, and duration is achieved through a neural response-driven engine, specifically including:

[0039] According to real-time brainwave The training intensity level is adjusted by band oscillation intensity. When the oscillation amplitude is consistently below the baseline level, the intensity is reduced and the duration of each oscillation is extended. A fatigue compensation algorithm is designed to calculate the nerve fatigue coefficient by the pupil diameter change rate and blink frequency. If the coefficient exceeds the warning value, a rest period is automatically inserted and the total training time for the day is shortened. All adjustment parameters are updated to the training scheme after being verified by the Monte Carlo optimization model.

[0040] Dynamic adjustments further include:

[0041] By embedding a synaptic efficacy assessment module into the neural response-driven engine, and dynamically adjusting the intensity parameters of the training game, users can overlay high-frequency, short-duration reinforcement training on key training items to break through the neural adaptation threshold.

[0042] Promoting neural circuit reorganization, specifically including:

[0043] Based on the power spectrum of brain waves, the most favorable path for recovery is identified by comparing with a norm database. The neural circuits are then intelligently recombined and trained repeatedly through orderly training programs to achieve the compensatory and recombinant functions of neural circuits.

[0044] Further explanation is needed regarding the specific implementation method of multi-dimensional hierarchical modeling and data management of the norm database. After the basic information acquisition unit 1 acquires the basic information of the test subjects, such as age, educational background, and medical history, it needs to match the corresponding norm database to determine the cognitive indicator benchmark. The accuracy of the norm database directly determines the suitability of subsequent testing and training programs. Therefore, a multi-dimensional hierarchical modeling technique is adopted to construct the database, and regional correction and blockchain technology are used to ensure data reliability. The specific implementation method is as follows:

[0045] First, the basic dimensions are divided according to age group, educational background, and medical history type: age group is divided into intervals of 10 years to cover the high-incidence age group of cognitive impairment; educational background is divided into four categories according to the level of education: primary school and below, junior high school, high school, and junior college and above, because the level of education will affect the cognitive test score; medical history type is divided into Alzheimer's disease type, vascular cognitive impairment type, mild cognitive impairment type, and other types according to the cause of cognitive impairment. The cognitive deficit dimensions caused by different causes are different. Under each dimension, the baseline value and standard deviation of the cognitive indicators for the corresponding population are stored. The baseline value of the cognitive indicators is the average score of the population in the standardized cognitive test, and the standard deviation reflects the normal fluctuation range of the indicators of the population, providing a basis for subsequent judgment of the deviation of the test subjects' indicators.

[0046] To accommodate differences in cognitive levels among different regional populations, a regional dynamic correction factor is introduced. This factor is a parameter that adjusts the baseline weight based on the medical level data of the region where the test subject is located. The medical level data is retrieved from the regional medical database, including the number of tertiary hospitals, the coverage rate of cognitive impairment screening, and the per capita cognitive rehabilitation resources index. The regional medical level score is calculated according to a preset formula, such as medical level score = (number of tertiary hospitals / total population × 0.4) + (screening coverage rate × 0.3) + (number of rehabilitation resources × 0.3). The higher the score, the closer the correction factor is to 1.1 (e.g., the medical level score is high in first-tier cities, so the baseline value is × 1.1), and the lower the score, the closer the correction factor is to 0.9 (e.g., in remote areas, the baseline value is × 0.9). For example, the baseline value for delayed recall of a 50-60-year-old high school education population is 25 points, which is corrected to 25 × 0.9 = 22.5 points in areas with low medical levels, ensuring that the baseline value matches the actual cognitive level of the local population.

[0047] The database achieves cross-institutional data sharing through blockchain distributed storage, specifically as follows:

[0048] Each participating medical institution acts as a blockchain node. When an institution uploads new norm data (such as test data from 100 new patients with mild cognitive impairment), the system generates an immutable hash value, which is synchronized to all nodes. When other institutions access the data, they must verify whether the local data hash matches the node hash. If they match, the data can be retrieved, preventing data tampering or forgery. For example, if Hospital A uploads the performance baseline values ​​for vascular cognitive impairment patients aged 60-70, Hospital B verifies the hash consistency when accessing the data, ensuring its authenticity and reliability, and enabling cross-institutional collaboration. Each time norm parameters are accessed, data freshness verification is performed. This verification checks the version difference between the local norm data and the latest data in the cloud, avoiding the use of outdated data. The specific process is as follows:

[0049] The system reads the last update time and version number of local data and compares it with cloud data. If the time difference exceeds a set threshold (e.g., 7 days) or the version number is more than 3 versions lower than the cloud version, the data freshness is determined to be insufficient, triggering automatic synchronization. The latest norm parameters are downloaded from the cloud to overwrite the local data. For example, if the local data was last updated 30 days ago and the latest cloud data was updated 7 days ago, the time difference of 23 days is greater than 7 days. The system automatically synchronizes the latest baseline value and standard deviation to ensure that the norm parameters always reflect the latest cognitive characteristics of the population.

[0050] The specific implementation method for adaptive test sequence generation and multimodal data acquisition and processing is as follows: After the norm parameters are regionally corrected, the cognitive test execution unit 2 needs to generate a test sequence adapted to the test subject, while collecting multi-dimensional data and eliminating environmental interference to ensure the accuracy of the original test data and provide a reliable basis for subsequent defect localization. The specific implementation method is as follows:

[0051] The process of generating adaptive test sequences based on norm parameters requires dynamic adjustments based on the test subjects' baseline dimensions and real-time performance. First, the initial test difficulty is determined according to the norms. For example, if the test subject is 60-70 years old, has a junior high school education, and suffers from vascular cognitive impairment, the norms indicate that the baseline score for word recall in the memory dimension for this population is 20 points. The initial sequence starts with a 5-word recall test. During the test, if the test subject correctly recalls 4 words on the first question (3 words above the passing standard of 20 points), the difficulty of the next question increases to 6 words. If only 1 word is recalled (below the passing standard), the difficulty decreases to 4 words. Through a feedback loop of difficulty and performance, a sequence adapted to the test subject's cognitive level is generated, avoiding tests that are too difficult or too easy. To mitigate potential testing biases, test items are dynamically switched based on the test subjects' real-time responses. For example, in executive function testing, the Stroop test (requiring the ability to recognize words whose font color and meaning conflict) is performed first. If the test subject's reaction time exceeds the norm baseline of 1.5 seconds × 1.5 = 2.25 seconds, it indicates weak executive function. The test is then immediately switched to a simpler connect-the-dots test, connecting numbers in sequence. If the reaction time is below 1.8 seconds (benchmark 1.5 seconds × 1.2), the test is switched to a more complex Wisconsin card sorting test, sorting cards by color and shape. This ensures the test accurately covers the boundaries of the test subject's cognitive deficiencies, rather than being limited to a single item. A multimodal interactive data acquisition device is used to simultaneously record three core categories. The device integrates a microphone, capacitive touchscreen, and eye tracker. The accuracy rate is obtained by measuring the correctness of the tested questions. The touchscreen operation trajectory records the touch coordinates and swipe paths of the tested finger (e.g., the trajectory and dwell time when swiping from option A to B in a memory test), reflecting motor coordination and decision-making processes. The eye-tracking heatmap records the location and duration of the tested's fixation points, generating a pseudo-color heatmap (red represents areas with fixations longer than 1 second, blue represents <0.5 seconds), reflecting attention allocation (e.g., in a memory test, short fixation times on word areas may indicate attention deficits). To simulate real-world interference and ensure data purity, interference is deployed. Suppression Algorithm: During the test, 50 dB of background noise (such as office conversations or keyboard typing) is injected through the device's built-in speaker to simulate interference scenarios in daily life. Then, time-domain signal separation technology is used to remove the environmental noise component from the original data. In this solution, the role of this technology is to distinguish between cognitive response signals and noise signals in the time dimension. The original data is divided into 100-millisecond time windows, and the signal energy of each window is calculated. The energy of environmental noise fluctuates steadily, while the cognitive response signal has non-stationary characteristics. By setting an energy mutation threshold, the non-stationary cognitive response signal is separated, and finally, a pure cognitive response data stream is extracted to avoid test data distortion caused by noise.

[0052] The specific implementation method for constructing the initial scheme of the personalized recovery training algorithm is as follows: After the test report generation unit 3 outputs the cognitive deficit type of the test subject, the training scheme analysis unit 5 needs to generate an initial scheme through the personalized recovery training algorithm. The core is to match the appropriate training module and intensity to ensure the training is targeted and feasible. The specific implementation method is as follows:

[0053] First, a knowledge base mapping cognitive deficit types to training modules is established. This construction process requires combining clinical data and neuroscience research: collecting training effect data from 1000 patients with different cognitive deficits (e.g., the effectiveness rate of word association training for memory deficit patients is 82%, and the effectiveness rate of scene memory recall training is 78%). Deficit types are categorized (memory deficit, executive function deficit, language comprehension deficit). Each deficit type is associated with 2-3 effective training modules. For example, memory deficits are associated with word association training (strengthening memory encoding through semantic association between words) and delayed recall training (gradually extending the recall interval to improve memory retention); executive function deficits are associated with Stroop training (improving inhibitory control) and working memory refresh training (enhancing executive function by updating working memory content); and language comprehension deficits are associated with sentence completion training (completing incomplete sentences to improve language comprehension) and word naming training (recognizing images and naming them to enhance language extraction). The knowledge base is regularly updated with new clinical research findings (e.g., the newly added virtual reality scene memory training has an 85% effectiveness rate for memory deficits) to ensure the scientific rigor of module matching. Its function is to quickly locate and adapt training modules for specific types of cognitive deficits, avoiding blind selection. It maps specific deficit types to preset neural function reconstruction targets, which are the brain regions whose neural functions require repair. The preset process references neuroimaging research: for example, memory deficits are mainly associated with weakened neural connections in the hippocampus, and the preset target is to improve the theta wave synchronicity in the hippocampus, increasing delayed recall scores by 20%; executive function deficits are associated with insufficient activation of the prefrontal cortex, and the preset target is to shorten the prefrontal cortex-related P300 latency by 15% and the Stroop test reaction time by 10%; language comprehension deficits are associated with weakened function in the temporal lobe language area, and the preset target is to shorten word naming reaction time by 20% and increase sentence comprehension accuracy by 25%. These preset targets need to be quantifiable and measurable to provide a basis for subsequent training effect evaluation. The training intensity parameters are initialized through a dynamic difficulty adjustment engine, which automatically calculates the initial training difficulty based on the degree of deviation in cognitive indicators. The core is to ensure that the training difficulty is neither too high, leading to frustration, nor too low, failing to stimulate neural plasticity. The specific process is as follows:

[0054] Extract the degree of deviation for each cognitive indicator from the test report (e.g., a delayed recall score deviation of -4 points in the memory dimension, and a Stroop reaction time deviation of +0.6 seconds in the executive function). Establish a correspondence between the absolute value of the deviation and the initial challenge level (absolute deviation ≤ 1 points is Level 3, 1-3 points is Level 2, > 3 points is Level 1, with Level 1 being the lowest and Level 3 the highest). For example, a delayed recall deviation of -4 points (> 3 points) corresponds to Level 1 of the memory training module (e.g., initial word association training with 5 words), and a Stroop reaction time deviation of +0.6 seconds (1-3 points) corresponds to the executive function... The training module is designed to be at level 2 (e.g., the initial Stroop test uses 10 conflicting words). The initial intensity parameters for each training module are determined by considering the age and physical condition of the participants (e.g., the level is lowered by one level for participants over 70 years old). A synaptic plasticity prediction model is used to estimate the rate of neural circuit remodeling. This model is an algorithm based on historical EEG data (e.g., theta wave power spectral density, alpha wave inhibition rate) to predict the speed of neural pathway reconstruction. In this scheme, its role is to determine the training cycle based on the participants' neuroplasticity level, avoiding cycles that are too short or too long. The specific prediction process is as follows:

[0055] Inputting the subject's historical EEG data (e.g., pre-training theta wave power spectral density 12 μV² / Hz, alpha wave inhibition rate 25%), the model outputs a neural circuit reorganization rate value (e.g., 0.7 / week, representing a weekly increase in neural connectivity strength of 0.7 units). A high rate value (>0.8 / week) indicates good neural plasticity, estimating a training cycle of 4 weeks; a low rate value (<0.5 / week) estimates 6-8 weeks. Based on this, an initial framework for the training program is generated, including training module combinations, intensity tiers, and a cycle plan. Module combinations are prioritized according to the type of deficit (e.g., if memory deficit is primary and executive function is secondary, the combination would be word association training + Str...). OOP training, 15 minutes each day), intensity steps are set according to the recombination rate (rate 0.7 / week, level 1 in weeks 1-2, level 2 in weeks 3-4, level 3 in weeks 5-6). The cycle plan combines rate and goal (e.g., rate 0.7 / week, goal 20% improvement, cycle 6 weeks, 5 days a week, 30 minutes a day). For example, a subject tested with memory deficit and recombination rate of 0.6 / week, the initial plan is word association training (level 1, 5 words) + delayed recall training (level 1, 1 minute interval), 5 days a week, 15 minutes each day, in week 3, increase to level 2 (6 words, 2 minute interval) to ensure that the training is progressive and stimulates neuroplasticity.

[0056] The specific implementation of the dynamic operation mechanism of the personalized recovery training algorithm is as follows: After the initial training scheme is started, the training strategy needs to be adjusted through real-time neural response to avoid training failure due to neural adaptation. Therefore, the training scheme analysis unit 5 relies on the biofeedback closed-loop system to achieve dynamic optimization. The specific implementation is as follows:

[0057] During the training execution phase, the theta wave power spectral density and event-related potential (P300) latency were collected in real time using a biofeedback closed-loop system. Theta wave power spectral density refers to the energy distribution of EEG signals in the 4-8 Hz frequency band, reflecting neural synchronicity in this protocol. In memory training, increased theta wave synchronicity (power spectral density increasing from 12 μV² / Hz to 15 μV² / Hz) indicates more coordinated neural activation in the hippocampus memory-related brain regions, a marker of effective training. The P300 latency refers to the positive EEG potential approximately 300 milliseconds after the stimulus (such as word presentation in a memory test). The latency is the time from stimulus to potential appearance, reflecting cognitive processing speed in this protocol. In executive function training, a shortened P300 latency (from 420 milliseconds to 380 milliseconds) indicates improved cognitive processing efficiency in the prefrontal cortex, demonstrating significant training effectiveness. When increased theta wave synchronicity and P300 latency are detected... When the latency shortens, the criteria for enhancement and shortening must first be clarified: enhancement is defined as the theta wave power spectral density being higher than the previous round for three consecutive training rounds (each round lasting 5 minutes), with a total increase exceeding 15% (e.g., from 12→13→14→14.5μV² / Hz, an increase of 20.8%); shortening is defined as the P300 latency being lower than the previous round for three consecutive rounds, with a total shortening exceeding 8% (e.g., from 420→400→390→380 milliseconds, a shortening of 9.5%). After meeting the above conditions, the adaptive reinforcement learning module is activated, and an intensity increment strategy is generated based on historical training effect data: the system retrieves the intensity-effect correlation data of the previous three training sessions of the tested subject (e.g., the theta wave increase is 10% in level 1 training and 18% in level 2 training), and calculates the intensity of the next level by effect improvement × 0.6 + current intensity × 0.4. For example, if the effect improvement is 18% in the current level 1, the intensity of the next level = 18% × 0.6 + 1 × 0.4 ≈ 1.48, rounded up to level 2, specifically, the number of words in memory training is increased from 5 to 6, and the number of words in the Stroop function training is increased from 10 to 12. This ensures that the intensity increase matches the neural response and avoids overstimulation. If the expected neural response is not achieved after three consecutive training sessions (i.e., theta wave synchronicity does not increase or even decreases, and P300 latency does not shorten or even lengthens), it indicates that the current training modality may exceed the tested neural compensatory capacity, and a cross-modal alternative training strategy needs to be initiated. For example, after the original visual training task (such as recalling words from pictures) is performed three times, the theta wave power spectral density decreases from 12 to 11 μV² / Hz, and P30... The latency increased from 420 to 430 milliseconds. The system automatically switched the visual modality to the auditory modality, requiring the user to recall recorded words. Simultaneously, training parameters were adjusted (e.g., the auditory word playback speed was 20% slower than the visual presentation). The core of cross-modal substitution is utilizing the compensatory capabilities of different neural pathways (visual relies on the occipital lobe, auditory relies on the temporal lobe) to overcome the neural bottlenecks of a single modality. For example, if impaired occipital lobe function renders visual memory training ineffective, auditory memory training may achieve memory encoding through the temporal lobe pathway, reactivating neural plasticity. After two consecutive auditory training sessions achieving the expected response, visual training can be gradually resumed, forming a multimodal collaborative training approach.

[0058] The specific implementation method for cognitive deficit localization and dynamic adjustment of training intensity is as follows: The original test data needs to accurately locate the core and associated areas of cognitive deficits. Simultaneously, the training intensity needs to be adjusted based on neural response and fatigue levels to ensure safe and effective training. The specific implementation method is as follows: The original test data is input into a deficit localization decision tree. This decision tree contains hierarchical discrimination nodes for memory, executive function, and language comprehension dimensions: The memory dimension node first judges the deviation between delayed recall score and immediate recall score (e.g., delayed recall deviation -4 points, immediate recall deviation -1 point, prioritizing the marking of memory deficit). The executive function node judges the deviation between Stroop reaction time and the completion time of the connection test (e.g., Stroop deviation +0.6 seconds, marking weak executive function). The language comprehension node judges the deviation between sentence cloze accuracy and word naming accuracy (e.g., a -15% deviation in sentence cloze accuracy marks a language comprehension deficiency). Simultaneously, a cross-dimensional correlation analysis matrix is ​​constructed. The construction process uses the three core indicators (memory: delayed recall score, executive function: Stroop reaction time, and language comprehension: sentence cloze accuracy) as rows and columns of the matrix. The Pearson correlation coefficient between any two indicators is calculated (e.g., a correlation coefficient of 0.65 between delayed recall score and Stroop reaction time indicates that memory deficiencies are often accompanied by executive function abnormalities). Indicators with a correlation coefficient > 0.5 are marked as strongly correlated, and those < 0.3 are marked as weakly correlated. This matrix can identify the cross-dimensional impact of deficiencies. When the working memory indicator deviation exceeds the first-level threshold, the system automatically correlates and detects an abnormality in the attention allocation indicator. Working memory indicator deviation refers to the difference between the working memory test score (such as the n-back test, which requires remembering stimuli up to the nth time) and the norm baseline. The first-level threshold is set at -20% of the baseline value (e.g., if the baseline score is 30, the threshold is 24; a deviation < 24 is considered exceeding the limit). For example, if the student's n-back test score is 22 (deviation -8, exceeding the limit), the system automatically retrieves the attention allocation indicator (e.g., the error rate of the attention cancellation test, with a baseline error rate of 5%). If the error rate reaches 12% (deviation +7%), the system determines that the working memory deficit is associated with an abnormal attention allocation, indicating a deficiency. It does not exist in isolation; attention and working memory need to be trained simultaneously. The final output is a topology map of defect types. The output process is as follows: the system marks the main defects determined by the decision tree (such as memory defects) as the main defect core area (displayed as a red circle in the center of the topology map, with the defect dimension and bias value marked, such as memory defect: delayed recall bias -4 points). It marks the defects with strong correlation in the correlation analysis matrix (such as executive function defects and attention defects) as the correlation defect influence area (marked as an orange ring around the core area; the higher the correlation, the closer the influence area is to the core area; for example, the ring with an executive function correlation of 0.65 is closer to the core area than the ring with an attention correlation of 0.5). Weakly correlated defects (such as language comprehension correlation of 0.65) are marked as the influence area.2) The edges are marked with small blue circles; the topology map also marks the percentage deviation of indicators in each region (e.g., core area deviation -25%, executive function influence area deviation +20%), providing spatial positioning basis for training module matching (e.g., core area prioritizes matching memory training modules, influence area matches executive function training modules). Regarding training intensity adjustment, the training intensity level is adjusted according to the real-time EEG gamma band (30-80Hz) oscillation intensity. The gamma band oscillation intensity reflects the level of neural excitation. The baseline level is the average intensity collected 5 minutes before training (e.g., 15μV² / Hz). If the oscillation amplitude remains below 80% of the baseline level for 5 minutes (e.g., <12μV² / Hz), it indicates insufficient neural excitation, and the training intensity is reduced (e.g., the number of words in memory training is reduced from 6 to 5, and the number of Stroop words in executive function training is reduced from 12 to 10). Simultaneously, the duration of a single training session is extended (from 15 minutes to 18 minutes) to avoid ineffective training due to excessively low intensity. If the oscillation amplitude remains above 120% of the baseline (>18μV² / Hz), the intensity can be appropriately increased. To address this issue, a fatigue compensation algorithm was designed. In this scheme, the algorithm monitors neural fatigue in real-time using physiological indicators and adjusts the training pace accordingly. This involves collecting pupil diameter change rate (a pupil diameter shrinking more than 10% from its initial value during training is considered a fatigue signal) and blink frequency (more than 20 blinks per minute is considered fatigue) using an eye tracker. The neural fatigue coefficient is then calculated as follows: Coefficient = (Pupil diameter shrinkage rate × 0.5) + ((Blink frequency - Baseline frequency) / Baseline frequency × 0.5), where the baseline frequency is the frequency at which the initial 5 minutes of training began. The average blink rate per minute (e.g., 15 times / minute). For example, if the pupil constricts by 12% and the blink rate is 22 times / minute, the coefficient = (12% × 0.5) + ((22 - 15) / 15 × 0.5) ≈ 0.06 + 0.23 ≈ 0.29. A warning value of 0.4 is set. If the coefficient exceeds the warning value (e.g., 0.45), a 5-minute rest period is automatically inserted (playing soothing music and prompting to close eyes and rest), and the total training time for the day is shortened (from 30 minutes to 25 minutes) to avoid cognitive overload due to nerve fatigue.

[0059] All adjusted parameters must be validated by the Monte Carlo optimization model. This model simulates the training effects of different intensity and duration combinations 1000 times, selects the parameter combination that maximizes the training effect score minus the fatigue risk score (e.g., the combination of intensity level 2 and 16 minutes per session scores the highest), and updates it to the training plan to ensure that the adjustments are scientific and reasonable. A synaptic efficacy assessment module is embedded in the neural response-driven engine. By dynamically adjusting the intensity parameters of the training game, users can superimpose high-frequency short-term reinforcement training on key training items to break through the neural adaptation threshold. Based on the EEG power spectrum, the most favorable path for recovery is identified in the norm database, and the neural circuits are intelligently reorganized and recompensated through orderly repeated training of the training items to complete the compensation and reorganization functions of the neural circuits.

[0060] Further explanation is needed regarding the specific implementation of the synaptic efficacy assessment module and neural circuit reorganization in the neural response-driven engine. Based on the engine's existing methods of adjusting training intensity using EEG gamma-band oscillation intensity and controlling training duration through fatigue compensation algorithms, relying solely on these two adjustments over a long period can easily lead to a neural adaptation dilemma. This means that neural circuits gradually develop habitual responses to training stimuli of fixed intensity and patterns, and synapses cease to exhibit new plasticity changes, such as synaptic connection strength no longer increasing and neurotransmitter release efficiency stagnating, making it difficult to break through training bottlenecks. Therefore, a synaptic efficacy assessment module needs to be embedded in the engine. This module dynamically optimizes training stimuli by quantifying synaptic plasticity levels and uses a norm database to locate the optimal recovery path. Ultimately, through ordered training, neural circuit compensation and reorganization are achieved. The specific implementation is as follows:

[0061] The core function of the synaptic efficacy assessment module is to quantify the transmission efficiency and plasticity potential of neural synapses in real time. Its assessment is not based on a single indicator, but rather integrates multi-dimensional EEG data and training response data output from the biofeedback closed-loop system. On one hand, the module continuously retrieves EEG characteristic parameters during training, including the rate of synchronous change of theta waves (4-8Hz, closely related to memory-related synaptic activation) and the magnitude of latency shortening of event-related potentials (P300), reflecting the improvement in synaptic transmission speed. On the other hand, the module combines the rate of improvement in the test subject's answer accuracy and touchscreen operation. The smoothness of the trajectory changes is analyzed by a pre-set synaptic efficacy calculation model. This model integrates EEG and reaction indicators with a 4:6 weighting to generate an efficacy value of 0-1, where 0 represents no plasticity and 1 represents extremely high plasticity. The current synaptic efficacy level is output in real time. For example, if a subject improves theta wave synchronicity by 8%, shortens P300 latency by 7%, and improves accuracy by 12% during memory training, the module calculates a synaptic efficacy value of 0.65. If this value does not increase for two consecutive training days and remains at 0.65, it is determined that the current training intensity is close to the neural adaptation threshold, and a dynamic adjustment strategy needs to be activated.

[0062] In cases of synaptic efficacy stagnation, the module precisely adjusts the intensity parameters of the training game. This training game must match the subject's cognitive deficit type. For example, memory deficits correspond to word matching games, requiring the memorization of the correspondence between words and images; executive function deficits correspond to Stroop color and word conflict games, requiring the identification of font color while ignoring word meaning. Intensity parameters encompass difficulty level, stimulus presentation frequency, and feedback interval. The adjustment logic follows efficacy orientation. If the synaptic efficacy value is below 0.5, indicating weak plasticity, the stimulus presentation frequency is slightly increased to avoid a sudden increase in difficulty that could lead to frustration. If the efficacy value is between 0.5 and 0.7, indicating moderate plasticity, then… Simultaneously increasing the difficulty level and the timeliness of feedback, the module activates synapses through dual stimulation of difficulty and feedback. Based on the adjustment of intensity parameters, the module superimposes high-frequency, short-duration reinforcement training on key training items to break through the neural adaptation threshold. Key training items refer to those directly related to the core areas of the test subject's deficiencies, including delayed recall training for memory deficiencies and sentence completion training for language comprehension deficiencies. High frequency is reflected in the stimulation frequency being 2-3 times higher than conventional training, while short duration is controlled to 2-3 minutes per reinforcement session to avoid neural over-excitation and fatigue. Furthermore, reinforcement training and conventional training alternate at 5-minute intervals and 2-minute reinforcement intervals. The core logic of this mode is:

[0063] High-frequency stimulation can rapidly activate the presynaptic membrane and promote the release of neurotransmitters. Short-duration design can avoid the decline in synaptic efficacy caused by neural fatigue, thereby breaking the state of neural adaptation. That is, synapses that were originally unresponsive to conventional stimuli can undergo plasticity changes again under high-frequency, short-duration stimulation. For example, a subject who experienced efficacy stagnation during "delayed recall training" showed that after high-frequency, short-duration reinforcement, their theta wave synchronicity increased from 65% to 72% within 2 minutes, indicating that the synapse has re-entered a plastic state. After completing the synaptic efficacy enhancement and threshold breakthrough, the module compares the optimal recovery path with a norm database based on the real-time acquired EEG power spectrum. The power spectrum analysis requires extracting energy distribution characteristics of key frequency bands during training, such as the energy proportion of theta waves (4-8Hz) in the hippocampus and the activation intensity of gamma waves (30-80Hz) in the prefrontal cortex. These characteristics directly reflect the activation state of neural circuits. The norm database stores power spectrum templates for patients with different types of cognitive deficits and different recovery stages. During the comparison process, the module calculates the similarity between the current measured power spectrum and the norm template, selecting the recovery path corresponding to the template with the highest similarity. For example, if the measured current theta wave energy proportion in the hippocampus increases by 12%, it is similar to the norm template of "2 weeks of memory deficit recovery". With a similarity of 0.85, the recovery path corresponding to this template is "word association training, high-frequency short-delayed recall, and scene memory reproduction training." The module uses this as its core reference. Based on the optimal recovery path, the module intelligently reorganizes the order and combination of training items, and achieves neural circuit compensation and reorganization through orderly and repeated training. The reorganization follows the logic of first activating basic pathways, then strengthening core synapses, and finally consolidating new connections. For example, referring to the above path, 10 minutes of "word association training" is first arranged to activate basic neural pathways around the hippocampus, laying the groundwork for synaptic strengthening, followed by 5 minutes of "high-frequency short-delayed recall training," specifically targeting... Strengthening the connection strength of memory-related synapses, the final 10-minute "scene memory reproduction training" is conducted to integrate the synapses activated by the previous training into a complete neural circuit. The orderly repetition is reflected in two rounds of training according to this combination every day. The training time and intensity of each item in each round are gradually adjusted according to the synaptic efficacy assessment results. For example, the "scene memory reproduction" in the first week is a simple indoor scene, and the second week is upgraded to a complex outdoor scene. During this process, if a core neural pathway being tested is damaged, repeated stimulation will activate the alternative pathway. Neural compensation is achieved by improving the synaptic efficacy of the alternative pathway, ensuring that subsequent training always adapts to the changes in the neuroplasticity of the test subjects.

[0064] In this invention, a norm database is hierarchically modeled according to multiple dimensions such as age, educational background, and medical history. A regionalized correction factor is introduced, and blockchain is used to store and verify the freshness of the data. Cognitive tests generate adaptive sequences, and multimodal data collection includes answer accuracy, touch screen trajectory, and eye movement data. Temporal signal separation and noise reduction are performed, and a training program analysis unit locates cognitive deficits. Training modules are matched, and biofeedback is used to monitor the EEG theta waves and P300 latency. The training intensity and frequency are dynamically adjusted to promote neural circuit reorganization and improve the accuracy of rehabilitation.

[0065] The second objective of this invention is to provide a method for implementing an analysis system for cognitive rehabilitation training programs for people with cognitive impairment, including any of the above-mentioned features, comprising the following steps:

[0066] S1. Obtain basic information of the test subjects, match the norm database based on the multi-dimensional hierarchical model, verify the freshness of the data through the blockchain distributed storage mechanism, and apply the regional dynamic correction factor to calibrate the benchmark value weight.

[0067] S2. Generate an adaptive test sequence based on the calibrated norm parameters, use a multimodal interactive device to synchronously collect the answer accuracy, touch screen trajectory and eye movement data, use time domain signal separation technology to remove environmental noise interference, and output a pure cognitive response data stream;

[0068] S3. Input the test data into the defect localization decision tree to identify specific defect types, construct a cross-dimensional correlation analysis matrix to generate a defect type topology map, predict the neural reorganization rate based on the synaptic plasticity prediction model, and initialize the training module combination, intensity ladder and cycle plan through the dynamic difficulty adjustment engine.

[0069] S4. During training, the EEG characteristics are monitored in real time through a biofeedback closed-loop system. The training intensity, frequency and duration are dynamically adjusted by the neural response-driven engine. Neural path labeling technology is used to strengthen synaptic labeling. The reorganization effect is verified by diffusion tensor imaging and the training program is iteratively updated.

[0070] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A human brain cognitive recovery training program analysis system for cognitive impairment, characterized by, Comprise: The basic information acquisition unit (1) is used for acquiring the basic information of the measured personnel and determining the corresponding norm database; The cognitive test execution unit (2) executes cognitive test according to the basic information of the measured personnel and the corresponding norm data and acquires test raw data; The test report generation unit (3) generates test report according to test raw data; The norm updating unit (4) acquires test report and updates norm parameters in norm database according to test report; The training scheme analysis unit (5) generates personalized recovery training scheme automatically based on test report and norm parameters by analyzing cognitive impairment characteristics of the measured personnel, wherein analyzing cognitive impairment characteristics of the measured personnel includes identifying deviation of cognitive indicators in test report to locate specific defect type, the personalized recovery training algorithm includes matching preset training module combination based on specific defect type and dynamically adjusting training intensity, frequency and duration, optimizing training module to improve neural plasticity, promoting neural circuit reorganization through repeated stimulation and progressive challenge to realize dynamic recovery of cognitive function, including real-time monitoring of training feedback and iterative updating of scheme to ensure recovery effect; The dynamic operation mechanism of the personalized recovery training algorithm, specifically includes: In the training execution phase, the brain waves power spectrum density and the event-related potential P300 latency are collected in real time through the biofeedback closed-loop system When the wave synchronization is enhanced and the P300 latency is shortened, the adaptive reinforcement learning module is activated, the intensity increment strategy is generated based on the historical training effect data, if the expected neural response is not reached for three consecutive times, the cross-modal substitution training strategy is started, and the visual training task is automatically converted into the auditory channel for execution to break through the neural compensation bottleneck. When the wave synchronization is enhanced and the P300 latency is shortened, the adaptive reinforcement learning module is activated, the intensity increment strategy is generated based on the historical training effect data, if the expected neural response is not reached for three consecutive times, the cross-modal substitution training strategy is started, and the visual training task is automatically converted into the auditory channel for execution to break through the neural compensation bottleneck. The dynamic adjustment of training intensity, frequency and duration is realized by neural response driving engine, specifically including: According to real-time electroencephalogram The oscillation intensity of the brain wave band is adjusted to adjust the training intensity level, the intensity is reduced and the single duration is prolonged when the oscillation amplitude continuously falls below the baseline level, a fatigue compensation algorithm is designed, a neural fatigue coefficient is calculated through the change rate of pupil diameter and the blink frequency, if the coefficient exceeds the warning value, a rest period is automatically inserted and the total training time of the day is shortened, and all adjustment parameters are updated to the training plan after being verified by a Monte Carlo optimization model.

2. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 1, characterized in that: The norm database is constructed by using multi-dimensional hierarchical modeling technology, specifically including: According to the medical level data of the region where the measured personnel is located, the baseline value weight is automatically adjusted, and the database is stored by block chain distributed storage to realize cross-institution data sharing. Each time the norm parameters are called, data freshness verification is performed. If the difference between the current data version and the latest version in the cloud exceeds the set threshold, automatic synchronization is triggered.

3. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 2, characterized in that, The process of executing cognitive test and acquiring test raw data, specifically includes: Based on norm parameters, generate adaptive test sequence, dynamically switch test items according to real-time reaction of measured personnel, use multi-modal interactive collection device to record answer accuracy, touch screen operation trajectory and eye tracking hotspot map, and deploy interference suppression algorithm, simulate real environment interference by injecting background noise during test, and use time domain signal separation technology to separate environmental noise component from original data to extract pure cognitive reaction data stream.

4. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 3, characterized in that: The application of personalized recovery training algorithm includes: establishing defect type-training module mapping knowledge base, mapping specific defect type to preset neural function reconstruction target, initializing training intensity parameter through dynamic difficulty adjustment engine, the engine calculates initial challenge level based on the deviation degree of each cognitive indicator in test report, and uses synaptic plasticity prediction model to estimate neural circuit reorganization rate according to historical electroencephalogram data, and generates initial scheme framework including training module combination, intensity ladder and cycle plan.

5. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 3, characterized in that: The deviation of cognitive indicators in test report specifically includes: The test raw data is input into a defect positioning decision tree, the decision tree includes hierarchical discrimination nodes of memory dimension, execution function dimension and language understanding dimension, and a cross-dimension correlation analysis matrix is constructed, when the working memory index deviation exceeds the first threshold value, the attention allocation index abnormality is automatically associated and detected, and finally a defect type topology graph is output, a main defect core area and a correlation defect influence area are marked, and spatial positioning basis is provided for training module matching.

6. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 5, characterized in that: The dynamic adjustment further includes: A synaptic efficacy evaluation module is embedded in the neural response driving engine, the intensity parameters of the training game are dynamically adjusted, high-frequency short-time intensive training is superimposed on the user on the key training items, and the neural adaptation threshold is broken through.

7. The human brain cognitive recovery training scheme analysis system for cognitive impairment according to claim 6, characterized in that: The promoting neural circuit reorganization specifically includes: Based on the power spectrum of brain waves, the most beneficial path for recovery is compared in the norm database, the neural circuit compensation and reorganization function is completed through intelligent reorganization and ordered repeated training of training items.

8. A method for implementing a system for analyzing a cognitive recovery training program for the human brain for cognitive impairment comprising the system according to any one of claims 1 to 7, characterized in that: The method comprises the following steps: S1, obtaining the basic information of the measured person, matching the norm database based on the multi-dimensional hierarchical model, verifying the data freshness through the block chain distributed storage mechanism, and applying the regional dynamic correction factor to calibrate the benchmark value weight; S2, generating an adaptive test sequence according to the calibrated norm parameters, synchronously collecting the answer correctness, touch screen trajectory and eye movement data by using a multi-modal interactive device, stripping environmental noise interference by using time domain signal separation technology, and outputting pure cognitive reaction data stream; S3, inputting the test data into a defect positioning decision tree to identify specific defect types, constructing a cross-dimension correlation analysis matrix to generate a defect type topology graph; based on a synaptic plasticity prediction model, the neural reorganization rate is predicted, and the training module combination, intensity ladder and cycle plan are initialized by using a dynamic difficulty adjustment engine; S4, in the training, the brain electrical characteristics are monitored in real time by using a biological feedback closed loop system, the training intensity, frequency and duration are dynamically adjusted by using a neural response driving engine, synaptic markers are strengthened by using neural path marking technology, and the reorganization effect is verified by diffusion tensor imaging to update the training scheme iteratively.

Images