Intelligent question and answer method and system for child brain health fusing multi-modal data

Abstract

The application provides a kind of fusion multimodal data's child brain health intelligent question and answer method and system, it is related to child brain health question and answer technical field, by receiving the question and answer request containing the description of brain health problem to be consulted, child identity mark and multi-scene service demand mark, corresponding multimodal data set is obtained and preliminary association mark is called by calling multimodal database.The data after marking and problem description are mapped across modal semantics, and a unified semantic space is constructed to mine potential associated data and generate intervention basis.According to the intervention basis and multi-scene mark, the question and answer response content and individualized intervention scheme are generated, and are sent to multiple ends simultaneously and feedback is collected, so that the child brain health problem can be comprehensively and accurately understood, the individualized intervention scheme is provided, and the consultation and intervention efficiency and quality are improved.

Description

A Smart Question-Answering Method and System for Children's Brain Health that Integrates Multimodal Data Technical Field

[0001] This invention relates to the field of children's brain health question-and-answer technology, and more specifically, to a children's brain health intelligent question-and-answer method and system that integrates multimodal data. Background Technology

[0002] In the field of children's brain health, as society pays increasing attention to children's physical and mental development, the demand for consultation and intervention on related issues is growing. Currently, there are several main approaches to children's brain health consultation. Firstly, traditional manual consultation relies on the experience of professional doctors or experts; however, the number of professionals is limited, failing to meet the large demand for children's brain health consultations, and the consultation efficiency is low, making timely responses difficult. Secondly, most existing intelligent question-and-answer systems are based on only a single type of data, such as text data, performing simple keyword matching and responses to children's brain health questions, lacking a comprehensive and in-depth understanding of children's brain health. Children's brain health involves multiple aspects, including brain structure, language behavior, and genes; a single data source cannot accurately grasp the true state of children's brain health. Furthermore, existing systems do not fully consider the differences and specificities of children's brain health issues in different scenarios, failing to provide individualized intervention plans for different scenarios such as families, schools, and medical facilities, resulting in poor intervention outcomes. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for intelligent question-and-answering about children's brain health that integrates multimodal data, the method comprising:

[0004] Receive a multi-scenario question and answer request for children's brain health initiated by a user. The multi-scenario question and answer request for children's brain health includes a description of the brain health problem to be consulted, the associated child's identity identifier, and a multi-scenario service requirement identifier. The multi-scenario service requirement identifier is used to indicate the family care scenario, school adaptation scenario, or medical consultation scenario corresponding to the requirement.

[0005] Based on the child's identity identifier, a preset multimodal database of children's brain health is invoked to obtain a multimodal data set corresponding to the child's identity identifier. The multimodal data set is initially associated with the multi-scenario service demand identifier. The multimodal data set includes children's brain structure imaging data, language behavior record data, gene association data, and phased questionnaire feedback data.

[0006] Cross-modal semantic mapping is performed on the multimodal dataset with preliminary association labels and the brain health problem description to construct a unified semantic space for multimodal data. Potential association data related to the brain health problem description are mined in the unified semantic space for multimodal data to generate a basis for intervention of brain health problems.

[0007] Based on the intervention criteria for brain health issues and the service demand identifiers for multiple scenarios, a question-and-answer response content and an individualized brain health intervention plan that conform to the clinical knowledge base for children's brain health are generated. The question-and-answer response content includes an analysis of the causes of the problems and key points for scientific responses. The individualized brain health intervention plan includes detailed implementation rules for different scenarios.

[0008] The question-and-answer response content and the personalized brain health intervention plan are simultaneously sent to user terminals, teacher terminals of associated schools, and doctor terminals of cooperating medical institutions, triggering a multi-terminal collaborative feedback process to collect feedback information on the use of the question-and-answer response content and the personalized brain health intervention plan from each terminal.

[0009] Furthermore, embodiments of the present invention also provide a children's brain health intelligent question-and-answer system that integrates multimodal data, including:

[0010] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described intelligent question-and-answer method for children's brain health that integrates multimodal data by executing the machine-executable instructions.

[0011] Based on the above, by receiving question-and-answer requests containing descriptions of brain health issues to be consulted, child identification, and multi-scenario service demand identifiers, user demand information can be obtained. Based on the child's identification, a pre-defined multimodal database of children's brain health is accessed to obtain a multimodal data set, which is then preliminarily associated and labeled. This fully utilizes multi-source information such as children's brain structure imaging data, language behavior records, gene association data, and phased questionnaire feedback data to characterize children's brain health status from multiple dimensions, improving the comprehensiveness and accuracy of understanding children's brain health issues. Cross-modal semantic mapping processing is performed on the multimodal data set with preliminarily associated labels and the brain health issue descriptions to construct a unified semantic space for multimodal data. This allows for the discovery of potential related data related to the brain health issue descriptions. Based on the intervention criteria for brain health issues and the multi-scenario service demand identifiers, question-and-answer response content and individualized brain health intervention plans are generated. These plans include analysis of the causes of the problems and key points of scientific response, as well as detailed implementation rules for different scenarios, achieving accurate diagnosis and personalized intervention for children's brain health issues. The question-and-answer response content and individualized brain health intervention plan are simultaneously sent to user terminals, teachers' terminals of associated schools, and doctors' terminals of partner medical institutions, triggering a multi-terminal collaborative feedback process. This allows for the collection of usage feedback information from each terminal, further optimizing the question-and-answer response content and intervention plan, and significantly improving the efficiency and quality of children's brain health consultation and intervention. Attached Figure Description

[0012] Figure 1 is a schematic diagram of the execution flow of the intelligent question-and-answer method for children's brain health that integrates multimodal data provided in an embodiment of the present invention.

[0013] Figure 2 is a schematic diagram of exemplary hardware and software components of the children's brain health intelligent question-and-answer system that integrates multimodal data provided in an embodiment of the present invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 is a flowchart illustrating a method for intelligent question-and-answering about children's brain health that integrates multimodal data, according to an embodiment of the present invention. The method for intelligent question-and-answering about children's brain health that integrates multimodal data will be described in detail below.

[0015] Step S110: Receive a multi-scenario question and answer request for children's brain health initiated by the user. The multi-scenario question and answer request for children's brain health includes a description of the brain health problem to be consulted, the associated child's identity identifier, and a multi-scenario service demand identifier. The multi-scenario service demand identifier is used to indicate the family care scenario, school adaptation scenario, or medical consultation scenario corresponding to the demand.

[0016] In this embodiment, taking the consultation initiated by the parent of an 8-year-old male child (hereinafter referred to as the "target child") through a mobile application as an example, the question and answer request is transmitted to the backend processing module via HTTPS encryption protocol. The description of the brain health problem is submitted through the natural language input box, which reads: "In the past three months, the teacher has repeatedly reported that the target child cannot maintain a sitting posture for more than 15 minutes in class, frequently leaves his seat without permission, has a math homework completion accuracy rate of only 60%, and exhibits obvious peer interaction avoidance behavior." The child's identity identifier uses a system-generated 18-bit unique code, including the region code, institution code, and check digit, which is obtained by calling the device's local secure storage module. The multi-scenario service requirement identifier is determined by single selection through the scenario selection list. The parent selects "school-adapted scenario," and this identifier is embedded in the request header field in binary bit (01) form.

[0017] Step S120: Based on the child's identity identifier, call the preset child brain health multimodal database to obtain the multimodal data set corresponding to the child's identity identifier, and initially associate the multimodal data set with the multi-scenario service demand identifier. The multimodal data set includes the child's brain structure imaging data, language behavior record data, gene association data, and phased questionnaire feedback data.

[0018] Upon receiving the aforementioned question-and-answer request, the child's identity was first anonymized. A data index value was then generated using the SHA-256 algorithm, and the index value was used as the key to query the distributed database cluster. This multimodal database employs a hierarchical storage architecture, with hot data (data from the past 6 months) stored on an SSD array and cold data stored in a tape library. The acquired multimodal data set of the target children includes: 3.0T MRI brain structure imaging data (DICOM format, 0.5mm slice thickness, containing 300 slices each of T1-weighted and T2-weighted images) collected in March 2024; language behavior records from January to June 2024 (including 50 hours of classroom audio clips and 2000 sentences of daily conversation text); gene association data from October 2023 (whole-exome sequencing results, including detection data for 3 SNP loci related to neurodevelopment); and interim questionnaire feedback data from April 2024 (including raw response data and scale scores for 5 standardized scales, including the Conners Child Behavior Scale and the Child Anxiety Scale). After obtaining the above data, the initial association marking is completed by adding the "Scene Identifier: School Adaptation" tag to the metadata field of each data file. The marking process is implemented by calling the database API through a Python script, and the marking results are written to the data version control module in real time.

[0019] In this embodiment, a data collection and usage authorization agreement needs to be displayed to the user through the application interface. This agreement clearly informs the user of the types of data to be collected (including but not limited to children's identification, descriptions of brain health problems, multimodal datasets, etc.), their purpose (for generating Q&A response content, individualized intervention plans, and multi-terminal collaborative feedback optimization), storage period (until the child reaches adulthood or the user actively requests deletion), and the scope of third-party data sharing (only necessary data is shared with service providers such as affiliated schools and cooperating medical institutions). The user needs to complete the authorization by checking a confirmation box and electronically confirming with a handwritten signature. The entire authorization process is recorded and archived for verification. For the collection of privacy-sensitive data, such as gene-related data and brain structure imaging data, the system will pop up a separate authorization window, detailing the data encryption and storage method (using the AES-256 encryption algorithm), access control (only authorizing doctors' terminals and system administrators to access through two-factor authentication), and the de-identification process (removing direct identifiers such as names and ID numbers from the original data and using virtual identity identifiers to associate the data). All data collection activities comply with legal and regulatory requirements. Users can view or revoke authorization at any time through the "Privacy Settings" module of the application. After authorization is revoked, the system will immediately stop using the data and complete the data deletion operation within 30 days. The deletion record will be uploaded to the blockchain evidence storage system to ensure that it cannot be tampered with.

[0020] Step S130: Perform cross-modal semantic mapping processing on the multimodal data set with preliminary association labels and the brain health problem description to construct a unified semantic space for multimodal data. In the unified semantic space for multimodal data, mine potential associated data related to the brain health problem description to generate a basis for intervention of brain health problems.

[0021] Step S131: Extract features from various types of data in the multimodal dataset, including brain region morphological features from brain structure imaging data, expression coherence features from language behavior recording data, genetic locus association features from gene association data, and behavioral tendency features from phased questionnaire feedback data.

[0022] After acquiring and initially associating the multimodal dataset, the feature extraction process is initiated. For brain structural imaging data, image files are first read using the DICOM parsing library. A deep learning-based automatic brain region segmentation algorithm is then used to process the T1-weighted images. This algorithm includes convolutional layers, pooling layers, and upsampling layers. The input is 3D image data. The convolutional layers extract local features, the pooling layers perform downsampling, and the upsampling layers restore spatial resolution, ultimately outputting segmentation masks for 12 key brain regions (including the prefrontal cortex, hippocampus, and amygdala). Based on the segmentation results, morphological parameters such as volume, surface area, and cortical thickness of each brain region are calculated. Fifteen morphological parameters are extracted from each brain region, forming a 180-dimensional brain region morphological feature vector. For language behavior recording data, natural language processing (NLP) techniques were employed. First, speech segments were converted to text to obtain text data. Then, the text data underwent word segmentation, part-of-speech tagging, and syntactic analysis to extract features such as sentence length, average word length, number of pauses, and grammatical error rate. Simultaneously, language fluency indicators, such as the number of effective words per unit time and the proportion of repeated words, were calculated. These features were integrated into a 60-dimensional expression coherence feature vector. For gene association data, gene loci related to neurodevelopment were screened from whole-exome sequencing results. Based on a pre-defined gene database, the risk allele frequency, genotype combination, and association strength with brain functional phenotypes for each locus were determined. This information was quantified into a 40-dimensional genetic locus association feature vector. For interim questionnaire feedback data, the raw responses were converted into standardized scores according to the scoring rules of different scales. The correlation between items on each scale was analyzed, and behavioral tendency dimensions such as attention deficit factor, hyperactivity / impulsivity factor, and anxiety factor were extracted, forming a 50-dimensional behavioral tendency feature vector.

[0023] Step S132: Input the extracted brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features into a pre-trained cross-modal semantic mapping model. The cross-modal semantic mapping model includes a feature alignment layer, a semantic transformation layer, and a spatial fusion layer.

[0024] The four extracted feature vectors are concatenated in a preset order to form a multimodal feature matrix, which serves as the input to the cross-modal semantic mapping model. This cross-modal semantic mapping model is built on the Transformer architecture and includes a feature alignment layer, a semantic transformation layer, and a spatial fusion layer. The feature alignment layer consists of two fully connected sub-layers. The first sub-layer performs a linear transformation on the input multimodal feature matrix, and the second sub-layer uses batch normalization to reduce the distribution differences between different modal features. The semantic transformation layer contains six Transformer encoder blocks, each consisting of a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism is used to capture long-distance dependencies between different modal features, and the feedforward neural network performs a non-linear transformation on the attention output. The spatial fusion layer uses a convolutional neural network structure, containing three convolutional layers and two pooling layers, to map the semantically transformed features to a unified semantic space.

[0025] Step S133: Through the feature alignment layer of the cross-modal semantic mapping model, the different types of features are processed to unify the dimensions, and the brain region morphological features, the expression coherence features, the genetic locus association features and the behavioral tendency features are converted into feature vectors of the same dimension.

[0026] In the feature alignment layer, the dimensions of brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features are first adjusted independently. For the brain region morphological feature vector (180 dimensions), it is mapped to a 256-dimensional intermediate vector through a fully connected layer; for the expression coherence feature vector (60 dimensions), it is also mapped to 256 dimensions through a fully connected layer; the genetic locus association feature vector (40 dimensions) and the behavioral tendency feature vector (50 dimensions) are also mapped to 256 dimensions through their respective fully connected layers. During the mapping process, the weight parameters of each fully connected layer are obtained through pre-training and non-linearly transformed using the ReLU activation function. After the dimension mapping is completed, the four 256-dimensional intermediate vectors are added element-wise to obtain a fused 256-dimensional feature vector, which serves as the output of the feature alignment layer.

[0027] Step S1331: Obtain the original dimensional parameters of the brain region morphological features. These original dimensional parameters are determined by the acquisition resolution of the brain structure image data and the number of brain regions extracted. In this embodiment, the acquisition resolution of the brain structure image data determines the number of voxels in the original image. Combined with the number of brain regions obtained through automatic segmentation, after feature extraction, the original dimensional parameters of the brain region morphological features are represented by a specific vector length. This length is related to the number of morphological parameters extracted from each brain region and the total number of brain regions. For example, if A morphological parameters are extracted from each brain region, and a total of B brain regions are segmented, then the original dimensional parameters are the product of A and B.

[0028] Step S1332: Obtain the original dimensional parameters of the expressive coherence feature. These original dimensional parameters are determined by the collection duration of the language behavior recording data and the number of extracted language feature indicators. The collection duration of the language behavior recording data affects the amount of analyzable speech segments and text. The extracted expressive coherence features cover multiple language feature indicators such as sentence structure, vocabulary usage, and pause patterns. Assuming C language feature indicators are extracted, each containing several sub-parameters, the original dimensional parameters of the integrated expressive coherence feature are the length of the vector formed by the sum of these indicators and sub-parameters.

[0029] Step S1333: Obtain the original dimensional parameters of the genetic locus association features. These original dimensional parameters are determined by the number of detection loci in the gene association data and the number of association analysis indicators. In the gene association data, the number of detection loci directly affects the feature dimension. Each detection locus corresponds to multiple association analysis indicators, such as allele frequency and genotype combination score. If D related gene loci are detected, and E association analysis indicators are calculated for each locus, then the original dimensional parameters of the genetic locus association features are the vector length formed by the product of D and E.

[0030] Step S1334: Obtain the original dimension parameters of the behavioral tendency feature. These original dimension parameters are determined by the number of questions in the phased questionnaire feedback data and the number of statistical behavioral dimensions. The phased questionnaire feedback data contains multiple questions, each corresponding to a different behavioral expression. Through statistical analysis of the question answers, multiple behavioral tendency dimensions are extracted. For example, if the questionnaire contains F questions, factor analysis yields G behavioral tendency dimensions, each with its corresponding statistical value. Therefore, the original dimension parameters of the behavioral tendency feature are the length of the vector composed of the statistical values ​​of the G dimensions.

[0031] Step S1335: Input the original dimensional parameters of the brain region morphological features, the expression coherence features, the genetic locus association features, and the behavioral tendency features into the feature dimension calculation module. Calculate the target dimensional parameter that meets all feature transformation requirements. The target dimensional parameter is greater than or equal to the maximum value among the original dimensional parameters of each type of feature and is compatible with dimension transformation algorithms for each type of feature. The feature dimension calculation module first compares the four input original dimensional parameters to determine the maximum value. Then, considering the possible information loss during the transformation process of each type of feature and the requirements of the subsequent semantic transformation layer on the feature vector dimension, select a dimensional value greater than or equal to this maximum value that can be effectively processed by dimension transformation algorithms for each type of feature (such as fully connected layer mapping) as the target dimensional parameter. The selection of this target dimensional parameter must ensure that the transformed feature vector retains the key information of the original features, while facilitating cross-modal semantic alignment and fusion.

[0032] Step S1336: For the morphological features of the brain regions, a feature interpolation algorithm is used to convert their original dimensions to the target dimensions, preserving key structural information in the brain region morphological features during the conversion process. The feature interpolation algorithm extends the length of the feature vector to the target dimension by inserting intermediate values ​​calculated based on neighboring feature values ​​between the dimensions of the original feature vector. During the interpolation process, special attention is paid to parameters that are representative of the morphological structure of the brain regions, such as the volume of key brain regions and cortical thickness, to ensure that this key structural information is not distorted after the dimension conversion. For example, for parameters representing the volume of the prefrontal cortex in the original dimension, reasonable calculations are performed based on their relationship with the volume parameters of adjacent brain regions during interpolation to maintain the structural correlation between brain regions.

[0033] Step S1337: For the expressive coherence features, a feature expansion algorithm is used to transform their original dimensions into the target dimension. Dimension expansion is achieved by supplementing the relevant feature values ​​related to the language expression logic. The feature expansion algorithm does not simply add redundant data, but rather derives relevant related features from existing expressive coherence features based on the inherent logic of language expression. For example, auxiliary features for semantic coherence can be derived based on the grammatical structure features of a sentence; language style-related features can be expanded based on vocabulary selection preferences. These supplemented related feature values ​​enable the expressive coherence feature vector to reach the target dimension, while simultaneously enriching the linguistic logic information contained within the features.

[0034] Step S1338: For the genetic locus association features, their original dimensions are converted into target dimensions. The original genetic locus features are grouped and recombined, with each group corresponding to a new dimension, maintaining the genetic association. The original genetic locus features may be relatively scattered. Multiple biologically functionally related genetic locus features are grouped together, with each group serving as a new dimension in the target dimension. For example, multiple locus features related to neurotransmitter metabolism can be combined into a new dimension. This increases the number of dimensions while maintaining the intrinsic association between genetic loci, making the converted feature vector better reflect the synergistic effects at the gene level.

[0035] Step S1339: For the behavioral tendency features, a feature mapping algorithm is used to transform their original dimensions into target dimensions, establishing a mapping relationship between the original behavioral features and the target dimension features to maintain the accuracy of the behavioral tendency. The feature mapping algorithm first analyzes the importance and interrelationships of each dimension in the original behavioral tendency features, and then constructs a mapping function from the original dimension space to the target dimension space. Through this mapping function, the original behavioral tendency feature vector is projected into a higher-dimensional space, while ensuring that the mapped features can still accurately reflect the child's behavioral tendencies, such as attention tendency and emotional stability tendency.

[0036] Step S13310 involves normalizing the brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features after conversion to the target dimension, forming a feature vector with a unified dimension. The normalization process uses the Min-Max standardization method, transforming the values ​​of each element in each feature vector to the range [0, 1]. Specifically, for each element in each feature vector, the minimum value in the feature vector is subtracted from the element value, and then divided by the difference between the maximum and minimum values ​​in the feature vector. This process unifies the numerical scale of different types of feature vectors, facilitating further processing and fusion by the subsequent cross-modal semantic mapping model.

[0037] Step S134: Through the semantic transformation layer of the cross-modal semantic mapping model, each feature vector after unification is mapped to the preset children's brain health semantic dictionary space, and a semantic label corresponding to each feature vector is generated. The semantic label includes brain health indicator category and data association scenario identifier.

[0038] The semantic transformation layer takes the 256-dimensional feature vector output from the feature alignment layer as input, and first feeds this feature vector into the first Transformer encoder block. In the multi-head self-attention mechanism, the input vector is linearly transformed into a query vector, a key vector, and a value vector, respectively. The attention weight of each head is calculated by the dot product of the query vector and the key vector, normalized by the softmax function, and then multiplied by the value vector to obtain the output of that head. The outputs of multiple heads are concatenated and linearly transformed to obtain the output of the multi-head self-attention mechanism. This output is then residually connected to the input vector, normalized by layers, and then fed into a feedforward neural network. The feedforward neural network contains two linear transformation layers with a ReLU activation function in between. The output, after residual connection and layer normalization, becomes the output of the first encoder block. This process is repeated in six encoder blocks, with the output of each encoder block serving as the input to the next. After processing by six encoder blocks, the semantically transformed feature vector is obtained. The feature vector is compared with a pre-defined semantic dictionary space for children's brain health. This semantic dictionary space contains 500 core semantic concepts, each corresponding to a semantic label. The cosine similarity between the feature vector and each semantic concept is calculated, and the top 10 semantic concepts with the highest similarity are selected as the semantic labels corresponding to the feature vector. The format of the semantic labels is "brain health indicator category - data association scene identifier", such as "attention function - school scene" and "emotion regulation - family scene".

[0039] Step S135: Through the spatial fusion layer of the cross-modal semantic mapping model, semantic associations between different feature vectors are established based on the semantic labels corresponding to each feature vector, forming an initial semantic network.

[0040] The spatial fusion layer receives the feature vectors and corresponding semantic labels output by the semantic transformation layer. First, it constructs a directed graph structure where nodes represent semantic labels and edges represent the relationships between them. For each feature vector and its corresponding semantic label, it calculates the co-occurrence frequency and semantic similarity. The co-occurrence frequency is obtained by counting the number of times semantic label pairs appear simultaneously in different feature vectors, and the semantic similarity is calculated using the concept distance in the semantic dictionary. The weighted sum of the co-occurrence frequency and semantic similarity is used as the association strength between semantic label pairs, with the weights determined through cross-validation. Based on the association strength, when the association strength exceeds a preset threshold, a directed edge is established between the two semantic label nodes, with the edge weight being the association strength value. In this way, all semantic label nodes corresponding to feature vectors are connected to form the initial semantic network. The initial semantic network contains nodes and directed edges; the number of nodes is equal to the total number of semantic labels, and the number of edges is determined based on the association strength threshold.

[0041] Step S136: Perform semantic parsing on the brain health problem description and extract the core semantic elements in the brain health problem description. The core semantic elements include the brain health field involved in the problem, the child development stage of concern, and the expected answer dimension.

[0042] Natural Language Processing (NLP) techniques were employed to semantically analyze descriptions of brain health issues. First, the problem description text was segmented into word sequences using a word segmentation tool, followed by part-of-speech tagging to identify nouns, verbs, adjectives, etc. Next, a named entity recognition model was used to identify key entities in the text, such as "classroom," "attention," "fidgeting," "learning efficiency," and "Attention Deficit Hyperactivity Disorder" (ADHD). Dependency parsing was used to analyze the grammatical relationships between words, identifying core verbs and noun phrases. Based on the above processing results, core semantic elements were extracted. The brain health domain involved in the problem was determined based on the identified disease names and symptom keywords; for example, "Attention Deficit Hyperactivity Disorder" corresponds to the "Attention Disorder domain." The developmental stage of the child was determined as "school age" based on the child's age. The expected dimensions of the solution were determined based on the interrogative words and requests in the problem, including dimensions such as "cause analysis," "intervention suggestions," and "school adaptation strategies." The extracted core semantic elements were represented as structured data, with each element containing an element type and an element value.

[0043] Step S137: Integrate the extracted core semantic elements into the initial semantic network, adjust the weights of each semantic relationship in the initial semantic network, strengthen the semantically overlapping relationships with the core semantic elements, and weaken the semantically non-overlapping relationships with the core semantic elements.

[0044] Step S1371: Weights are assigned to the core semantic elements. Weight values ​​are set based on the semantic overlap between the core semantic elements and the description of the brain health problem. Core semantic elements with a semantic overlap reaching a first preset threshold are assigned a first weight value, while core semantic elements with a semantic overlap below a second preset threshold are assigned a second weight value. The first weight value is greater than the second preset threshold, and the second weight value is less than the first preset threshold. In this embodiment, core semantic elements include "attention function - school scenario," "learning efficiency - classroom performance," and "hyperactivity - teacher feedback," etc. Semantic overlap is obtained by calculating the number of matches and semantic similarity between each core semantic element and keywords in the brain health problem description text. For example, "attention function - school scenario" is highly correlated with "short attention span in class" in the problem description, and its semantic overlap reaches the first preset threshold, therefore it is assigned a higher first weight value; while "emotion regulation - family scenario" has a weaker semantic association with the current problem description, and its semantic overlap is below the second preset threshold, therefore it is assigned a lower second weight value.

[0045] Step S1372: Traverse all semantic relationships in the initial semantic network and identify the semantic label corresponding to the feature vector involved in each semantic relationship. The initial semantic network is a graph structure composed of multiple nodes (semantic labels) and edges (semantic relationships). The traversal process starts from any node in the network and visits all nodes sequentially along the connection direction of the edges. For each semantic relationship represented by an edge, record the semantic labels of the two nodes it connects. For example, a semantic relationship connects two semantic label nodes: "prefrontal morphology - attention" and "frequency of classroom fidgeting - behavioral performance".

[0046] Step S1373 involves matching the semantic tags involved in each semantic association with the core semantic elements. A semantic similarity algorithm is used to calculate the matching degree between the semantic tags and the core semantic elements. The higher the semantic overlap between the semantic tags and the core semantic elements, the larger the matching degree. The semantic similarity algorithm is based on a pre-trained word vector model, converting semantic tags and core semantic elements into vector representations, and then calculating the cosine similarity between the two vectors as the matching degree. For example, the semantic tag "prefrontal cortex morphology - attention" has a high vector cosine similarity with the core semantic element "attention function - school scene," resulting in a large matching degree; while the semantic tag "hippocampal volume - memory function" has a smaller matching degree with the core semantic element "attention function - school scene."

[0047] Step S1374: Calculate the weight adjustment coefficient for each semantic association based on the weight values ​​and matching degrees of the core semantic elements. The weight adjustment coefficient is the product of the weight value and matching degree of the core semantic element. When a semantic association involves multiple core semantic elements, the average of the multiple weight adjustment coefficients is taken as the final adjustment coefficient. For each semantic association, if its connected semantic tag matches multiple core semantic elements, calculate the product of the weight value and the corresponding matching degree of each core semantic element to obtain multiple weight adjustment coefficients. Then, average the above coefficients to obtain the final weight adjustment coefficient for the semantic association. For example, if a semantic association matches both "attention function - school scene" (weight value W1, matching degree M1) and "learning efficiency - classroom performance" (weight value W2, matching degree M2), the final adjustment coefficient is (W1*M1 + W2*M2) / 2.

[0048] Step S1375: Multiply the original weights of each semantic association in the initial semantic network by the corresponding weight adjustment coefficient to obtain the adjusted semantic association weights. The larger the product of the original weight and the adjustment coefficient, the higher the adjusted weight; the smaller the product, the lower the adjusted weight. The initial weights are set based on the co-occurrence frequency between semantic tags and prior knowledge when constructing the initial semantic network. For example, if the original weight of a semantic association is W0 and its corresponding final adjustment coefficient is C, then the adjusted weight is W0*C. If C is greater than 1, the adjusted weight is higher than the original weight, indicating that the association is more important in the current problem context; if C is less than 1, the adjusted weight is lower.

[0049] Step S1376 involves normalizing the adjusted semantic relationship weights to maintain the sum of all semantic relationship weights at the initial set value. Normalization is achieved by dividing each adjusted weight by the sum of all adjusted weights, and then multiplying by the initial weight sum. Assuming the adjusted sum of all semantic relationship weights is S, and the initial weight sum is S0, then each normalized weight Wi' = Wi * (S0 / S). This ensures that the relative magnitudes of weights in the semantic network remain unchanged, while avoiding the impact of excessively high or low overall weight values ​​on subsequent network analysis and data mining.

[0050] Step S1377: Identify semantic relationships whose adjusted weights are lower than a preset weight threshold, determine them as irrelevant relationships, and remove them from the semantic network. The preset weight threshold is set based on experience and experimental results, and is used to filter out semantic relationships with extremely low relevance to the description of current brain health issues. For example, after adjustment and normalization, if the weight of a certain semantic relationship is 0.02, while the preset weight threshold is 0.05, then this relationship is determined to be an irrelevant relationship and removed from the semantic network to simplify the network structure and highlight important relationships.

[0051] Step S1378: Identify semantic associations whose adjusted weights exceed a preset high-weight threshold. The logic verification module checks their logical consistency; if a logical contradiction exists, the weight is reduced to a preset reasonable range. The preset high-weight threshold is used to filter out strong associations that may have logical conflicts. The logic verification module checks whether there are biological or psychological contradictions between the semantic tags connected to these high-weight associations. For example, there is usually a positive correlation between "dopamine receptor gene - hyperactivity tendency" and "low prefrontal metabolism - attention deficit." However, if an association shows a negative correlation with extremely high weights, a logical contradiction may exist. In this case, the logic verification module will reduce its weight to a preset reasonable range, such as reducing the weight from 0.8 to 0.5.

[0052] Step S1379 involves performing an overall matching degree check between the weighted semantic network and the core semantic elements. The matching degree calculation module outputs the matching score between the overall semantic network and the core semantic elements. If the matching score reaches a preset threshold, the weight adjustment is completed. If the matching score does not reach the preset threshold, the weights of the core semantic elements are readjusted, and the above weight adjustment steps are repeated until the matching score reaches the preset threshold. The matching degree calculation module comprehensively considers factors such as the coverage of the core semantic elements in the adjusted semantic network, the matching degree between high-weight associations and core elements, and calculates a comprehensive matching score. For example, if all core semantic elements have corresponding high-weight association nodes in the network, and these associations are logically consistent, the matching score is high. If the matching score does not reach the preset threshold, it indicates that the current weight adjustment has not sufficiently highlighted the importance of the core semantic elements. It is necessary to return to step S1371 to readjust the weights of the core semantic elements, possibly increasing the weight values ​​of some key elements, and then repeat the subsequent weight adjustment steps until the matching score meets the requirements.

[0053] Step S138: Optimize the structure of the adjusted semantic network by deleting redundant semantic association paths and supplementing missing key association nodes to form a unified semantic space for multimodal data.

[0054] After adjusting the semantic network weights, structural optimization is performed. First, redundant semantic association paths are removed. Redundancy is determined by calculating the redundancy of a path, defined as the degree of overlap between the semantic information conveyed by that path and other paths. Paths with redundancy exceeding a preset threshold are deleted. Next, missing key association nodes are added. By analyzing the connectivity and betweenness centrality of nodes in the semantic network, nodes with low connectivity but high betweenness centrality are identified; these are likely key association nodes. For missing key association nodes, corresponding semantic label nodes and association edges are added based on the semantic relationships in the semantic dictionary space. For example, if the semantic network contains nodes for "attention function" and "learning efficiency," but there is no direct association between them, and according to the semantic dictionary, "attention function" and "learning efficiency" have a close semantic relationship, then an association edge from "attention function" to "learning efficiency" is added, and corresponding weights are set. After structural optimization, a unified semantic space for multimodal data is formed. This space contains the optimized semantic network structure and the attribute information of each node and edge.

[0055] Step S139: Verify the rationality of data association relationships in the unified semantic space of multimodal data. By comparing the data association rules in the clinical knowledge base of children's brain health, remove semantic association pairs that do not conform to the data association rules, and retain semantic association pairs that conform to the data association rules to update the unified semantic space of multimodal data.

[0056] The data associations in the unified semantic space of multimodal data are compared with the data association rules in the clinical knowledge base for children's brain health. The clinical knowledge base for children's brain health contains 200 data association rules based on evidence-based medicine. Each rule defines the reasonable association direction and strength range between two brain health indicators. For each semantic association pair (i.e., the directed edge between two semantic label nodes) in the semantic network, its association direction and weight value are extracted and compared with the corresponding association rule in the clinical knowledge base. If the association direction is consistent and the weight value is within the strength range specified by the rule, the semantic association pair is considered to conform to the rule and is retained; if the association direction is opposite or the weight value exceeds the strength range, it is considered to not conform to the rule and is removed. For important data association rules that exist in the clinical knowledge base but are missing in the semantic network, corresponding semantic association pairs are added to the semantic network, and weight values ​​are set according to the rules. After reasonableness verification, the unified semantic space of multimodal data is updated to obtain the final unified semantic space.

[0057] Step S1310: Mine potential related data related to the description of brain health problems in the unified semantic space of the multimodal data, and generate a basis for intervention of brain health problems.

[0058] Based on the unified semantic space of the final multimodal data, graph mining algorithms are used to mine potential related data concerning brain health problem descriptions. First, the core semantic elements of brain health problem descriptions are used as query nodes. A breadth-first search is performed in the semantic network with a search depth of 3 layers to obtain all semantic label nodes directly or indirectly related to the query nodes. Then, the association strength between these nodes and the query nodes is calculated, represented by the sum of the weights of the edges along the path. The top 20 nodes with the highest association strength are selected as key association nodes, and their corresponding raw multimodal data are collected, including morphological parameters of specific brain regions from brain structural imaging data, specific language features from language behavior records, relevant locus information from gene association data, and scale scores from questionnaire feedback data. This data is integrated and analyzed, combined with clinical evidence from the pediatric brain health clinical knowledge base, to generate intervention criteria for brain health problems. The intervention criteria include possible causal analysis of the problem, assessment of relevant risk factors, and supporting research evidence, presented in structured text format.

[0059] Step S140: Based on the intervention criteria for brain health problems and the multi-scenario service demand identifiers, generate question-and-answer response content and individualized brain health intervention plans that conform to the clinical knowledge base for children's brain health. The question-and-answer response content includes analysis of the causes of problems and key points for scientific coping. The individualized brain health intervention plan includes detailed implementation rules for different scenarios.

[0060] First, a Q&A response is generated based on the intervention criteria for brain health issues. The problem cause analysis section, based on the causal analysis in the intervention criteria and combined with the characteristics of children's multimodal data, elaborates on biological factors (such as abnormal brain region morphology), psychological factors (such as behavioral tendencies), and environmental factors (such as school environment adaptation). The scientific response key points section summarizes general principles and methods for addressing this problem, referencing recommendations from the children's brain health clinical knowledge base to ensure the scientific rigor and authority of the content. Then, based on the multi-scenario service needs identification of "school-appropriate scenarios," individualized brain health intervention plans are generated. The scenario-specific implementation details of the individualized brain health intervention plan are designed for school scenarios, including specific content such as classroom management strategies, suggestions for adjusting learning tasks, and key points for teacher training. Each detail includes implementation steps, expected goals, and precautions.

[0061] Step S141: Decompose the basis for intervention in the brain health problem, and separate the direct basis data that is directly related to the description of the brain health problem and the indirect basis data used for supplementary explanation.

[0062] Before generating Q&A responses and individualized brain health intervention plans, the basis for intervention in brain health issues is first broken down. Direct evidence data refers to data directly related to the core symptoms described in the brain health problem description, such as prefrontal cortex thickness measurements of the target child, hyperactivity-impulsivity factor scores on the Conners Child Behavior Scale, and the frequency of attention deficit in classroom language behavior records. Indirect evidence data refers to data used to supplement the problem background, influencing factors, or as intervention references, such as reference values ​​for brain region development in children of the same age, intervention effect data from similar cases, and research literature on the impact of the school environment on attention. This breakdown process is implemented using a rule-based text classification algorithm, which categorizes sentences in the intervention evidence text into direct or indirect evidence data based on preset keywords and semantic patterns.

[0063] Step S142: Arrange the direct evidence data according to the logical hierarchy of the question and answer. The first layer arranges the evidence data related to the cause of the question, the second layer arranges the evidence data related to the impact of the question, and the third layer arranges the evidence data related to the response to the question. Based on the arrangement of each logical hierarchy, the main framework of the question and answer response content is generated.

[0064] The data will be arranged according to the logical hierarchy of the question-and-answer format. The first layer contains data related to the causes of the problem, including data on biological factors (such as abnormal brain region morphology), psychological factors (such as behavioral tendency data), and environmental factors (such as school environment adaptation data). The second layer contains data related to the impact of the problem, including data on its impact on academic performance and peer relationships. The third layer contains data related to problem coping, including data on the effectiveness of existing interventions and data on recommended coping methods. Based on these three logical layers, a main framework for the question-and-answer response content will be generated. This framework includes a title and subtitle for each layer, as well as corresponding data filling positions.

[0065] Step S143: Based on the pre-configured clinical knowledge base for children's brain health, supplement the data in each layer of the main framework, add clinical research conclusions and scientific explanations corresponding to the data, improve the main framework of the question and answer response content, and form a complete question and answer response content.

[0066] For each layer of data within the main framework, clinical research conclusions and scientific explanations related to that data were retrieved by consulting the Children's Brain Health Clinical Knowledge Base. For example, regarding the data on abnormal prefrontal cortex thickness in the first layer, relevant research was retrieved from the clinical knowledge base showing that the prefrontal cortex is closely related to attention control function and that abnormal cortical thickness may lead to attention deficit symptoms. Corresponding scientific explanations were then added to illustrate the neural mechanisms of the prefrontal cortex in attention regulation. These supplementary explanations were then added to the corresponding positions within the main framework to complete the main framework of the question-and-answer response, forming a complete response.

[0067] Step S144: Determine the target service scenario based on the multi-scenario service demand identifier. When the multi-scenario service demand identifier indicates a home care scenario, focus on intervention directions related to daily home care; when the multi-scenario service demand identifier indicates a school adaptation scenario, focus on intervention directions related to adjusting the campus learning environment; when the multi-scenario service demand identifier indicates a medical consultation scenario, focus on intervention directions related to professional medical examination and treatment advice.

[0068] Based on the "School-Adapted Scenario" identifier for multi-scenario service needs, the target service scenario is determined to be the school-adapted scenario. Within this scenario, intervention focuses on aspects related to adjusting the campus learning environment, including classroom seating arrangements, adapting teaching methods, designing learning tasks, and guiding peer interaction.

[0069] Step S145: For the determined target service scenario, extract intervention reference data related to the target service scenario from the indirect basis data. The intervention reference data includes children's intervention case data in the same scenario and the effect data of intervention methods adapted to the scenario.

[0070] Step S1451 involves tagging the indirect evidence data with scenario labels. Based on the data source and the scenario information described, scenario labels are added to each piece of indirect evidence data. Scenario labels include family scenario labels, school scenario labels, and medical scenario labels. Indirect evidence data may come from various literature sources, case reports, research papers, etc. For each piece of data, the environmental background of its source and the scenario information involved in the description are analyzed. For example, a case report describing the improvement of children's attention through behavioral training in the family has data from the family environment and describes daily family activities, therefore it is tagged with "family scenario label"; a research paper on the impact of classroom teaching strategies on children with attention deficit disorder is tagged with "school scenario label"; and clinical trial data on drug treatment for attention deficit hyperactivity disorder is tagged with "medical scenario label". The tagging process can combine natural language processing technology to automatically identify scenario keywords in the text for initial tagging, followed by review and confirmation by professionals.

[0071] Step S1452: Based on the determined target service scenario, filter indirect evidence data with corresponding scenario tags. When the target service scenario is a home care scenario, filter indirect evidence data with a home scenario tag; when the target service scenario is a school adaptation scenario, filter indirect evidence data with a school scenario tag; when the target service scenario is a medical consultation scenario, filter indirect evidence data with a medical scenario tag. In this embodiment, the target service scenario is a "school adaptation scenario," so the system will traverse all indirect evidence data and only retain those data labeled with "school scenario tag." For example, filter out case data and research data containing school scenario-related content such as "classroom management," "teacher intervention," "peer interaction," and "learning task adjustment."

[0072] Step S1453 involves classifying the selected indirect evidence data into three categories: successful intervention case data, partially effective case data, and ineffective case data. Successful intervention case data represents data showing improvement in children's brain health after intervention; partially effective case data represents data showing partial improvement in children's brain health after intervention, but not reaching the expected level; and ineffective case data represents data showing no improvement in children's brain health after intervention. For the selected indirect evidence data from school scenarios, classification is based on the intervention results described in the data. For example, if a case shows that through a specific classroom attention training program, a child's classroom attention duration increased from A minutes to B minutes after 3 months (B is significantly greater than A), and academic performance improved significantly, this is classified as successful intervention case data. In another case, the frequency of fidgeting decreased after intervention, but the improvement in attention duration was not significant, failing to meet the preset improvement target; this is classified as partially effective case data. In some cases, after implementing a certain intervention, the child's classroom performance and attention indicators showed no statistically significant difference compared to before the intervention; these are classified as ineffective case data. Classification criteria can be based on preset quantitative indicator thresholds, such as the percentage increase in attention duration or the reduction in the incidence of behavioral problems.

[0073] Step S1454: Extract intervention details from successful intervention case data. These details include the specific implementation methods, frequency, duration, and children's reaction data during the intervention. These intervention details serve as core intervention reference data. Successful intervention case data is a crucial basis for generating effective intervention plans. From these cases, various aspects of the intervention methods are extracted in detail. For example, the "classroom task decomposition method" described in a successful case involves breaking down the learning tasks of each lesson into several smaller tasks, each with clear goals and completion deadlines; the implementation frequency is once each for math and language arts classes daily; the implementation duration is one semester; and children's reaction data during the intervention includes a gradual increase in task completion rate and an increase in the number of times they actively ask questions. These details are accurately extracted and recorded as core intervention reference data.

[0074] Step S1455 involves extracting factors influencing the intervention effect from a subset of valid case data. These factors include family environment factors, school cooperation factors, and individual differences among children. These factors serve as supplementary intervention reference data, highlighting potential influencing conditions for the intervention effect. While some valid cases were not entirely successful, they reveal key factors affecting the intervention effect. For example, one valid case mentioned that the intervention was effective at school, but the family failed to cooperate in providing consolidation training, resulting in the overall effect falling short of expectations. This led to the extraction of the influencing factor "family environment factor - lack of consolidation training." Another case revealed that due to the large number of students in the class, teachers struggled to address the individualized intervention needs of each child, leading to the extraction of "school cooperation factor - insufficient teacher resources." Still other cases showed differences in the acceptance and response speed of different children to the same intervention, leading to the extraction of "individual differences among children - learning style preferences." This supplementary intervention reference data helps to consider various potential influencing conditions when developing intervention plans and to prepare accordingly in advance.

[0075] Step S1456 involves extracting the reasons for intervention failure from the invalid case data. These reasons include specific manifestations of mismatch between the intervention measures and the child's condition, oversights in the intervention implementation process, and interference factors from the external environment. These reasons serve as risk warning reference data to avoid similar problems in the intervention plan generation process. Invalid case data can provide valuable lessons to help avoid repeating mistakes. For example, in one invalid case, the intervention measures used were suitable for younger children, but the target child was older, and their cognitive level exceeded the applicable range of the measures, resulting in "mismatch between intervention measures and child's condition - age and cognitive level incompatibility." In another case, after the intervention plan was formulated, the teacher failed to strictly follow the plan, frequently omitting certain key steps, which falls under "oversights in the intervention implementation process - non-standard implementation." In yet another case, the noisy environment around the classroom frequently interfered with the implementation of the intervention measures, which is "interference factors from the external environment - environmental noise." The aforementioned risk warning reference data will be used to check the applicability, feasibility, and adaptability to the external environment when generating individualized intervention plans, thereby reducing the risk of intervention failure.

[0076] Step S1457: The core intervention reference data, supplementary intervention reference data, and risk warning reference data are sorted according to their importance. The core intervention reference data has a higher priority than the supplementary intervention reference data, which in turn has a higher priority than the risk warning reference data. Since the core intervention reference data directly provides successful intervention methods and is most crucial for generating effective intervention plans, it has the highest priority. Supplementary intervention reference data helps optimize intervention plans and considers influencing factors, so its priority is lower. Risk warning reference data is mainly used to mitigate risks, so its priority is relatively lower. In actual sorting, each type of data can be further subdivided. For example, the core intervention reference data can be further sorted according to the significance of the intervention effect, the number of cases, etc., to ensure that the most important data is considered first.

[0077] Step S1458 involves integrating the sorted reference data, removing duplicate reference information, and merging semantically similar reference content to form an intervention reference data set. During the extraction and sorting process, different case data may contain duplicate or semantically similar details of intervention methods, influencing factors, or reasons for failure. For example, multiple successful cases may mention the intervention method of "increasing classroom interaction frequency," requiring the removal of such duplicate information, retaining only the most comprehensive and representative description. Semantically similar content, such as "positive teacher feedback" and "encouraging teacher evaluation," is merged to form more general and accurate reference content. The integrated intervention reference data set is a deduplicated and merged structured data set, facilitating efficient retrieval and reference when generating intervention plans.

[0078] Step S1459 involves matching and verifying the intervention reference data set with the intervention needs of the target service scenario to determine whether each intervention need has corresponding intervention reference data support. If a need lacks supporting reference data, the system returns to the indirect basis data set for re-filtering and supplementing relevant data. The intervention needs of the target service scenario are derived from the description of brain health problems and analysis of multimodal datasets. For example, for the problem of "lack of concentration in class," intervention needs in a school scenario might include "increasing classroom focus time," "reducing fidgeting in class," and "improving the efficiency of completing learning tasks." Each piece of data in the intervention reference data set is matched with these needs to check whether each need has at least one supporting reference data. If a need, such as "improving the efficiency of completing learning tasks," lacks corresponding intervention reference data, the system returns to the initial indirect basis data set, expands the filtering scope or adjusts the filtering conditions, and re-searches for school scenario data related to that need until all intervention needs are supported by reference data, ensuring that the generated intervention plan has sufficient basis.

[0079] Step S146: Based on the intervention reference data and the intervention principles in the clinical knowledge base for children's brain health, generate a preliminary intervention plan module for the target service scenario. The preliminary intervention plan module for the home care scenario includes detailed rules for guiding daily behavior and suggestions for adjusting the home environment; the preliminary intervention plan module for the school adaptation scenario includes classroom interaction adaptation strategies and learning task adjustment plans; the preliminary intervention plan module for the medical consultation scenario includes a list of recommended examination items and suggestions for phased treatment.

[0080] Based on intervention reference data and intervention principles from the clinical knowledge base of children's brain health, a preliminary intervention plan module for school-adapted scenarios was generated. Classroom interaction adaptation strategies include specific strategies such as increasing the frequency of teacher attention to target children, adopting multi-sensory teaching methods, and setting up classroom interaction reward mechanisms; learning task adjustment plans include specific plans such as breaking down learning tasks into smaller steps, extending task completion time, and providing visual cues. Each strategy and plan clearly defines the implementing entity, implementation steps, and expected results.

[0081] Step S147: Adapt the preliminary intervention plan module to the child's multimodal data set. Combine the brain region development in the child's brain structure imaging data, the ability level in the language behavior recording data, and the genetic risk indications in the gene association data to modify the intervention clauses in the preliminary intervention plan module that do not match the child's multimodal data and strengthen the intervention focus in the preliminary intervention plan module that matches the child's multimodal data.

[0082] The initial intervention program modules were adapted to the target children's multimodal data set. For example, based on brain structural imaging data showing a thinner prefrontal cortex, and given that the prefrontal cortex is associated with executive function, the classroom interaction adaptation strategy emphasized training content aimed at improving executive function. Based on the good fluency of the target children's language expression in the language behavior recording data, the proportion of oral reporting tasks was increased in the learning task adjustment plan. Based on the absence of significant genetic risk indications in the gene association data, the intensity requirements of certain intervention measures were appropriately relaxed. For intervention clauses in the initial intervention program modules that did not match the children's multimodal data, such as requiring high-intensity memory training when the children's hippocampal volume and memory ability assessment were normal, these clauses were modified to reduce the training intensity or replace them with other more suitable training methods.

[0083] Step S148: Prioritize the adjusted intervention program modules, set sorting rules according to the urgency, implementation difficulty, and expected effect of the intervention measures, and determine the execution order of each intervention measure according to the sorting rules to form a structured individualized brain health intervention program.

[0084] The intervention measures in the adjusted intervention program module were prioritized. First, the prioritization rules were determined: intervention measures with high urgency, low implementation difficulty, and good expected results had higher priority. Urgency was assessed based on the severity and development trend of the problem; implementation difficulty was assessed based on the school's resources and teachers' execution capabilities; and expected results were assessed based on the effect data in the intervention reference data. Then, each intervention measure was scored for its urgency, implementation difficulty, and expected results. A weighted summation method was used to calculate the overall score, with weights determined based on multi-scenario service needs and the child's specific situation. The intervention measures were then ranked from highest to lowest based on the overall score, forming a structured, individualized brain health intervention program. Each intervention measure in the program was arranged in priority order, and each measure included information such as number, name, implementation steps, priority, and responsible personnel.

[0085] Step S149: The personalized brain health intervention plan is associated with the question and answer response content. The association tag is used to establish a one-to-one correspondence between the key points of the question and answer response content and the intervention measures in the personalized brain health intervention plan, so as to meet the user's reading and execution needs.

[0086] By adding the same association markers to both the scientific response points in the Q&A response content and the intervention measures in the individualized brain health intervention plan, a one-to-one correspondence is established. For example, the marker "G1" is added next to the item "Using diverse teaching methods to improve children's attention" in the scientific response points, and the marker "G1" is also added next to the intervention measure "Classroom interaction adaptation strategies - using multi-sensory teaching methods" in the individualized brain health intervention plan. In this way, users can quickly find the corresponding intervention measures by using the association markers when reading the Q&A response content, which facilitates understanding and implementation.

[0087] Step S150: The question-and-answer response content and the personalized brain health intervention plan are simultaneously sent to the user terminal, the teacher terminal of the associated school, and the doctor terminal of the cooperating medical institution, triggering a multi-terminal collaborative feedback process to collect feedback information on the use of the question-and-answer response content and the personalized brain health intervention plan from each terminal.

[0088] The message push service simultaneously sends the Q&A response content and personalized brain health intervention plan to user terminals (the mobile application of the target child's parents), teachers' terminals in associated schools (the office computer terminal of the teacher in the target child's class), and doctors' terminals in partner medical institutions (the hospital information system terminal of the target child's attending physician). The sent content is in encrypted PDF format to ensure information security. At the same time, a multi-terminal collaborative feedback process is triggered, displaying feedback entry points on each terminal and prompting users, teachers, and doctors to provide usage feedback information.

[0089] Step S151: When sending the question-and-answer response content and the individualized brain health intervention plan to each terminal, a feedback information collection template is generated simultaneously. The feedback information collection template includes feedback items on the comprehension of the question-and-answer response content, feedback items on the feasibility of the individualized brain health intervention plan, and feedback items on supplementary needs.

[0090] While sending the Q&A response content and the personalized brain health intervention plan, a feedback information collection template is generated for each terminal. The comprehension feedback item includes a rating of the level of comprehension for each part of the Q&A response content (e.g., "fully understood", "partially understood", "not understood") and a text input box for content that is not understood; the feasibility feedback item includes a feasibility rating of each intervention measure in the personalized brain health intervention plan (e.g., "fully feasible", "partially feasible", "not feasible") and a text input box for reasons for infeasibility; the supplementary needs feedback item is an open text input box used to collect other needs and suggestions from users, teachers, and doctors.

[0091] Step S152: Set a feedback submission time limit for the feedback information collection template. The time limit length is set according to the urgency of the child's brain health problem. The feedback submission time limit for urgent problems is shorter than the feedback submission time limit for regular problems.

[0092] The feedback submission deadline is set according to the urgency of the brain health issue. For attention deficit problems in target children, which are assessed as routine issues, a 7-day feedback submission deadline is set. The submission deadline and remaining time are clearly displayed in the feedback information collection template to remind end users to submit feedback in a timely manner.

[0093] Step S153: The feedback information collection template is bound to the question and answer response content and the individualized brain health intervention plan and sent to the user terminal, teacher terminal and doctor terminal to establish the association between the feedback template received by each terminal and the corresponding content.

[0094] By using file association technology, feedback information collection templates are linked to Q&A response content and personalized brain health intervention plans. This allows the feedback information collection templates to automatically pop up or be displayed in a designated location when a user opens the Q&A response content or intervention plan. Simultaneously, the system backend establishes a relationship between the feedback templates received by each terminal and their corresponding content, using terminal and content identifiers to ensure that feedback information accurately corresponds to the relevant Q&A response content and intervention plan.

[0095] Step S154: Within the feedback submission time limit, receive feedback information submitted by each terminal in real time, classify and store the received feedback information, and divide it into user feedback data, teacher feedback data and doctor feedback data according to terminal type.

[0096] Within the feedback submission deadline, the system backend monitors feedback submission requests from each terminal in real time and receives feedback information. The received feedback information is first validated to ensure data integrity and correct format. After successful validation, it is categorized and stored according to terminal type. User feedback data is stored in the user feedback database table, teacher feedback data in the teacher feedback database table, and doctor feedback data in the doctor feedback database table. Each database table contains fields such as feedback ID, terminal identifier, content identifier, feedback time, and feedback content.

[0097] Step S155: Analyze the categorized feedback data to extract the core feedback points from each data set. The core feedback points for user feedback data include content related to comprehension barriers and records related to execution barriers. The core feedback points for teacher feedback data include records of adaptation conflicts in campus scenarios and records of conflicts with teaching arrangements. The core feedback points for doctor feedback data include supplementary content for medical advice and suggestions for adjusting intervention plans.

[0098] Content analysis is performed on the categorized feedback data. For user feedback data, natural language processing (NLP) techniques are used to extract content sections corresponding to comprehension barriers, such as specific paragraphs in the explanation of the causes of problems marked as "not understood" in the comprehension feedback item; records corresponding to implementation barriers are also extracted, such as intervention measures marked as "infeasible" by users in the feasibility feedback item and their reasons. For teacher feedback data, records of adaptation conflicts in campus scenarios are extracted, such as teachers reporting conflicts between a certain intervention measure and existing classroom management regulations; records of conflicts with teaching arrangements are also extracted, such as an intervention measure requiring additional teaching time, conflicting with the course schedule. For doctor feedback data, supplementary content for medical recommendations is extracted, such as doctors suggesting adding a certain neuropsychological assessment; suggestions for adjusting intervention plans are also extracted, such as doctors suggesting changing the implementation frequency of a certain intervention measure.

[0099] Step S156: Associate the extracted core feedback points with the unified semantic space of the multimodal data, update the association weights of related data in the unified semantic space of the multimodal data, and strengthen the association weights of data that have semantic association with the feedback points.

[0100] The core feedback points are associated with a unified semantic space of multimodal data. The semantic similarity between the core feedback points and each semantic label node in the semantic space is calculated, and nodes with high similarity are identified as relevant data nodes. Then, the association weights between these relevant data nodes are updated by multiplying the association weights by a reinforcement coefficient greater than 1. The magnitude of the reinforcement coefficient is positively correlated with the importance of the core feedback points. For example, if the core feedback point in the teacher feedback data is "conflict between classroom interaction adaptation strategies and existing teaching progress," and it has a high correlation with the semantic label node "teaching progress - school scenario" in the semantic space, then the association weight of this node with other relevant nodes is strengthened.

[0101] Step S157: Based on the adjustment suggestions in the core feedback points, iteratively optimize the question-and-answer response content and the individualized brain health intervention plan, modify the statements in the question-and-answer response content that have comprehension difficulties, and adjust the intervention measures in the individualized brain health intervention plan that have execution difficulties.

[0102] Step S1571 involves categorizing the adjustment suggestions in the core feedback points into two categories based on their target audience: suggestions for adjusting the question-and-answer response content and suggestions for adjusting the individualized brain health intervention plan. The core feedback points originate from user terminals, teacher terminals, and doctor terminals. For example, user feedback such as "the term 'executive function deficit' in the question-and-answer response content is too technical and difficult to understand" falls under the category of adjustment suggestions for the question-and-answer response content; teacher feedback such as "the provision of 'three attention training sessions per lesson' in the individualized intervention plan is difficult to implement in actual teaching" falls under the category of adjustment suggestions for the individualized brain health intervention plan. By conducting thematic analysis of the text content of the core feedback points, the specific target audience of each adjustment suggestion is identified, thus completing the classification.

[0103] Step S1572: Based on the suggestions for adjusting the Q&A response content, extract the content sections corresponding to the comprehension obstacles, identify the triggering conditions for these obstacles, and determine if the triggering condition is that the proportion of technical terms in the Q&A response content exceeds a preset threshold. In this case, replace the technical terms with colloquial expressions and add simple explanations. If the triggering condition is that the overlap of logical levels in the Q&A response content exceeds a preset threshold, readjust the logical structure of the Q&A response content and establish connections between different parts by adding transitional expressions. From the Q&A response content adjustment suggestions, locate the specific paragraphs or sentences that users or teachers find difficult to understand. For example, one adjustment suggestion might point out that "the description of 'the relationship between prefrontal cortex development and attention network' in the third paragraph of the problem cause analysis section is too abstract." Further analysis of the triggering conditions for comprehension difficulties is conducted. The proportion of technical terms (such as "prefrontal cortex," "attention network," and "synaptic plasticity") in this section is statistically analyzed. If the proportion exceeds a preset threshold (e.g., A technical terms per 100 words), these technical terms are replaced with more colloquial expressions, such as "the area in the front of the brain responsible for focusing attention," "the neural system that manages attention," or "the ability of brain cells to change connections," with a brief explanation in parentheses added upon their first appearance. If comprehension difficulties are due to logical inconsistencies, such as multiple points of analysis intersecting and lacking clear hierarchy, the logical overlap between paragraphs (e.g., topic similarity) is calculated. When the overlap exceeds a preset threshold, the content is reorganized, merging content on the same topic and elaborating on different topics separately. Transitional phrases such as "firstly," "secondly," "in addition," and "therefore" are added between paragraphs to make the logical structure clearer.

[0104] Step S1573: Modify the question-and-answer response content according to the comprehension obstacle triggering conditions. After modification, test the response content. By simulating the reading habits of the target user, output a judgment result on whether the content meets the readability requirements. If the judgment result is that it does not meet the readability requirements, continue to adjust until the judgment result meets the requirements. After modification, conduct a readability test. Simulate the reading habits of the target user (such as children's parents, general teachers), such as reading speed, mastery of professional knowledge, etc. Use a readability assessment tool to evaluate the modified question-and-answer response content. The evaluation indicators include average sentence length, vocabulary difficulty level, paragraph structure clarity, etc. If the evaluation result shows that the readability score of the content does not meet the preset standard (e.g., the Flesch-Kincaid grade level is lower than the preset grade), it needs to be adjusted again to further simplify complex sentences, reduce vocabulary difficulty, and optimize paragraph structure until the readability test judgment result meets the requirements.

[0105] Step S1574: For the personalized brain health intervention program adjustment suggestions, extract the intervention measures corresponding to the execution obstacles indicated, identify the triggering conditions for these obstacles, and replace the intervention measure with an alternative intervention measure that matches the family conditions below a preset threshold when the triggering condition is: If the intervention measure's matching degree with family conditions is below a preset threshold when the triggering condition is: If the intervention measure's matching degree with school resources is below a preset threshold when the triggering condition is: Coordinate and supplement auxiliary intervention measures that match the school resources below a preset threshold when the triggering condition is: If the intervention measure's matching degree with medical resources is below a preset threshold when the triggering condition is: Recommend alternative medical solutions that match the nearest medical resources below a preset threshold when the triggering condition is: From the personalized brain health intervention program adjustment suggestions, identify the specific intervention measures that are indicated to have execution difficulties. For example, if a teacher reports that "the intervention program requires the use of specific attention training software, but the school's computer equipment is outdated and cannot run the software," this indicates an execution obstacle for the intervention measure of "using attention training software." Analyze the triggering conditions for execution obstacles by calculating the matching degree between the resource requirements of the intervention measure and the actual situation of family, school, or medical resources. If the match between family circumstances (such as economic situation and parents' time and energy) and a certain intervention measure (such as hiring private counselors) is below a preset threshold, it needs to be replaced with an alternative measure that the family can afford, such as "using free online educational resources for home counseling." If the match between school resources (such as teachers, equipment, and venues) and the intervention measure (such as small-class teaching and counseling) is insufficient, supplementary measures should be coordinated, such as "the school psychologist providing relevant counseling skills training to the homeroom teacher, and then the homeroom teacher implementing group counseling within the class." In cases of low matching with medical resources, such as when a recommended medical examination cannot be performed at a local hospital, similar alternative examinations that can be provided by a nearby hospital should be recommended.

[0106] Step S1575: Replace or modify the intervention measures in the individualized brain health intervention plan according to the execution obstacle triggering conditions. After modification, perform a suitability check on the adjusted intervention measures with the child's multimodal data set, and output the matching degree between the adjusted intervention measures and the child's multimodal data. If the matching degree meets the preset threshold, it is confirmed that there is no obvious fit conflict. After replacing or modifying the intervention measures, it is necessary to ensure that the new measures are still suitable for the specific situation of the target child. Compare and analyze the adjusted intervention measures with the child's multimodal data set, such as brain structure imaging data, language behavior record data, gene association data, questionnaire feedback data, etc. For example, if the adjusted intervention measure is "increase outdoor physical activity time to improve attention", it is necessary to check whether there is relevant information such as limited motor ability and preference for outdoor activities in the child's multimodal data, and calculate the matching degree between the measure and these data. If the matching degree (such as the matching degree between activity ability and activity intensity requirements, and the matching degree between interests and activity types) meets the preset threshold, it is considered that the intervention measure has no obvious fit conflict with the child's individual situation and can be adopted.

[0107] Step S1576 involves checking the correlation between the adjusted Q&A response content and the individualized brain health intervention plan. The corresponding relationships between the key points in the Q&A response content and the adjusted intervention measures are output. If a break in the correspondence is found, the key points in the Q&A response content are modified to ensure the correspondence meets the preset requirements. The scientific key points in the Q&A response content should correspond one-to-one with the specific intervention measures in the individualized brain health intervention plan, guiding users to understand why the measures are taken and how they relate to the causes of the problem. The correlation check is achieved by comparing the keywords and logical relationships between the two. For example, if the key point in the Q&A response content mentions "improving attention through structured task scheduling," and the adjusted intervention plan includes the measure of "daily classroom task list management," a correspondence should exist between the two. If the check finds that a key point lacks a corresponding intervention measure, or that an intervention measure lacks theoretical support in the key points, the correspondence is considered broken. The key points in the Q&A response content need to be modified, supplementing relevant explanations or adjusting the wording to ensure the correspondence is complete, accurate, and meets the preset correspondence requirements (such as a clear one-to-one or one-to-many correspondence).

[0108] Step S1577: Submit the adjusted Q&A response and individualized brain health intervention plan to the pediatric brain health expert review module. This module references the latest pediatric brain health clinical knowledge base and outputs a review result indicating whether the content conforms to clinical standards. The pediatric brain health expert review module contains a built-in clinical knowledge base comprised of the latest pediatric brain health clinical guidelines, expert consensus, and research literature. After the adjusted content is submitted, the module automatically compares the problem's cause analysis and key scientific responses in the Q&A response with the theoretical basis in the clinical knowledge base to check if they conform to currently accepted medical viewpoints. It also compares the intervention measures in the individualized intervention plan with the recommended treatment plans and intervention principles in the clinical knowledge base to check their safety, effectiveness, and standardization. For example, it reviews whether the dosage and frequency of the intervention measures are within safe limits and whether there are any contraindications. After the review is completed, it outputs a review result of "conforming to clinical standards" or "not conforming to clinical standards." For non-conforming content, it points out the specific discrepancies and provides suggestions for modification.

[0109] Step S1578: If the review result is in compliance with clinical guidelines, the optimized final content is determined. If the review result is not in compliance with clinical guidelines, adjustments are made again based on the modification suggestions in the review result until the review result is in compliance with clinical guidelines, forming the final optimized content. If the expert review module deems the adjusted content compliant with clinical guidelines, the content is determined as the final optimized result and can be sent to all end users. If the review result is not in compliance, for example, indicating that the recommended strength of a certain intervention has been reduced in the latest clinical guidelines, or that the expression of a scientific response point contradicts the latest research conclusions, then it is necessary to return to the corresponding adjustment step (such as modifying the Q&A response content in step S1572, or modifying the intervention in step S1574) for further adjustment based on the specific modification suggestions in the review result. After adjustment, the content is submitted for review again, and this cycle continues until the expert review module outputs a review result of "compliant with clinical guidelines," ultimately forming optimized content that meets both user needs and clinical professional standards.

[0110] Step S158: The optimized question-and-answer response content and the individualized brain health intervention plan are synchronized to each terminal again to complete a multi-terminal collaborative feedback loop. When each terminal submits feedback on the optimized content, the above feedback collection and optimization steps are repeated.

[0111] The iteratively optimized question-and-answer response content and personalized brain health intervention plan are then sent back to all terminals via push notification service, prompting users, teachers, and doctors to view the updated content. Users on each terminal can submit feedback on the optimized content again, and the system repeatedly performs steps such as feedback information collection, content parsing, semantic space association, and content optimization to continuously improve the quality of the question-and-answer response content and personalized brain health intervention plan.

[0112] Furthermore, the method may also include: step S210, updating the children's brain health multimodal database based on multi-terminal collaborative feedback information, which includes: extracting data update prompts from the collected multi-terminal collaborative feedback information, wherein the data update prompts include key data types missing in the multimodal data set and prompts indicating insufficient timeliness of existing data.

[0113] Data update prompts are extracted from multi-platform collaborative feedback. For example, teacher feedback mentioning "a lack of detailed observation records of the target child's recent classroom behavior" indicates a missing key data type—classroom behavior observation data—in the multimodal dataset; parent feedback mentioning "genetic association data is from a year ago, and we hope to update it with the latest data" suggests that the existing genetic association data is not timely enough. Data update prompts are automatically extracted from feedback information using text mining algorithms. These algorithms, based on keyword matching and semantic analysis, identify expressions related to missing or outdated data.

[0114] Step S211: Determine the data type to be supplemented based on the data update prompt. If the data update prompt indicates that the latest brain structure imaging data is missing, the data type to be supplemented is the updated brain structure imaging data. If the data update prompt indicates that the timeliness of language behavior recording data is insufficient, the data type to be supplemented is the newly added language behavior recording data. If the data update prompt indicates that gene-related data needs to be updated, the data type to be supplemented is the updated gene-related data. If the data update prompt indicates that the interim questionnaire feedback data is missing, the data type to be supplemented is the newly added questionnaire feedback data.

[0115] Based on the extracted data update prompts, the specific data types to be supplemented were determined. For the prompt "Missing classroom behavior observation data," the data type to be supplemented was determined to be "classroom behavior observation record data"; for the prompt "Insufficient timeliness of gene-related data," the data type to be supplemented was determined to be "gene-related updated data." The determination of the data types to be supplemented referenced the classification criteria of multimodal datasets to ensure consistency in data types.

[0116] Step S212: Send a data collection request to the associated data source terminal, including home data collection terminal, school data collection terminal, and medical institution data collection terminal. The data collection request includes the data type to be supplemented and the data collection requirements.

[0117] Based on the type of data to be supplemented, a data collection request is sent to the corresponding associated data source terminal. For example, for classroom behavior observation record data, a request is sent to the school's data collection terminal (the student behavior record system of the target child's school). The data collection requirements include the recording period (two consecutive weeks), recording frequency (three times a day, in the morning, afternoon, and during breaks), and recording content (the target child's sitting posture, attention span, number of fidgeting movements, and interaction with classmates, etc.). For gene association update data, a request is sent to the medical institution's data collection terminal (the laboratory system of the hospital where the target child is treated). The data collection requirements include the testing items (whole exome sequencing) and testing standards (compliant with the latest clinical testing guidelines). The data collection request is sent via an encrypted API interface and includes information such as the request ID, child identification, type of data to be supplemented, data collection requirements, and deadline.

[0118] Step S213: Receive the supplementary data returned by each data source terminal, and perform format unification processing on the received supplementary data to ensure that the format of the supplementary data is consistent with the format of the existing data in the Children's Brain Health Multimodal Database.

[0119] The system receives supplementary data from various data source terminals. Classroom behavior observation records returned by school data collection terminals are in CSV format, containing fields such as timestamp, behavior type, and duration. Gene association update data returned by medical institution data collection terminals are in XML format, containing fields such as test items, test results, and reference ranges. The data undergoes format standardization processing, converting the CSV and XML formats to JSON format supported by the database. Field names and data types are adjusted according to the existing data format to ensure consistency with the format of existing data in the Children's Brain Health Multimodal Database. For example, the "behavior type" field in the classroom behavior observation record data is uniformly converted to the same behavior classification code as in the language behavior record data.

[0120] Step S214: Associate the data to be supplemented after standardization with the corresponding child identification mark, and establish the correspondence between the data to be supplemented and the child's historical multimodal data set through the association mark.

[0121] A child identification field is added to the metadata of the data to be supplemented after standardization of the format, associating the data with the unique code of the target child. Simultaneously, an index is created in the database linking the data to be supplemented with the child's historical multimodal data set. This index allows for quick retrieval of all multimodal data for a given child, including historical data and newly added data to be supplemented. The association and index creation are implemented through database transactions to ensure data consistency.

[0122] Step S215: Add the data to be supplemented after association and labeling to the multimodal data set corresponding to the child in the multimodal database of children's brain health, and update the timestamp information of the multimodal data set.

[0123] The data to be supplemented after being associated and tagged is written into the storage directory of the multimodal data set corresponding to the target child in the Children's Brain Health Multimodal Database, and stored in the corresponding subdirectories according to data type (e.g., classroom behavior observation records are stored in the "Behavior Data" subdirectory). At the same time, the timestamp information of the multimodal data set is updated, and the last modified time is updated to the current time for subsequent data management and version control.

[0124] Step S216: Based on the updated multimodal data set, re-execute the cross-modal semantic mapping process, adjust the semantic associations in the unified semantic space of the multimodal data, and make the semantic associations match the updated data.

[0125] Using the updated multimodal dataset, the cross-modal semantic mapping processing flow from steps S131 to S139 is re-executed, including feature extraction, cross-modal semantic mapping model processing, semantic network construction, and optimization. Due to the addition of classroom behavior observation records and updated gene association data, corresponding classroom behavior features and updated genetic locus association features are extracted during the feature extraction stage. These new features affect the output of the cross-modal semantic mapping model, leading to changes in the semantic relationships within the semantic network. Through reprocessing, the weights and structure of semantic relationships in the unified semantic space of the multimodal data are adjusted to ensure that the semantic relationships accurately reflect the updated data characteristics.

[0126] Step S217: Store the updated multimodal data set and the adjusted multimodal data in a unified semantic space.

[0127] The updated multimodal dataset and the adjusted unified semantic space for multimodal data are stored in the database and semantic space storage module, respectively. The multimodal dataset uses distributed storage to ensure data reliability and scalability; the unified semantic space for multimodal data is stored in a graph database, storing semantic labels and relationships in the form of nodes and edges, facilitating efficient query and update operations. During storage, data is encrypted using the AES encryption algorithm to ensure data security.

[0128] For example, the method may further include: step S310, a step of evaluating and adjusting the phased effects of the individualized brain health intervention program, which includes: setting an evaluation cycle according to the execution cycle in the individualized brain health intervention program, wherein the evaluation cycle corresponds to the execution phase of the intervention program, and setting an evaluation to be automatically triggered after each execution phase ends.

[0129] The individualized brain health intervention program has a 3-month implementation cycle, divided into 3 phases, each lasting 1 month. Therefore, the evaluation cycle is set at 1 month, with an automatic effectiveness evaluation triggered at the end of each phase. The evaluation cycle is set through a task scheduling system. When the intervention program is initiated, a periodic evaluation task is created in the task scheduling system, with the task triggered on the last day of each phase.

[0130] Step S311: When the evaluation period arrives, send an intervention effect data collection request to the user terminal, teacher terminal, and doctor terminal. The intervention effect data collection request includes effect indicators to be collected, including records of children's behavioral improvement, language ability changes, and medical examination results.

[0131] When the evaluation period arrives, the task scheduling system triggers the intervention effect data collection process, sending intervention effect data collection requests to user terminals, teacher terminals, and doctor terminals. Effect indicators are determined based on the expected goals of the individualized brain health intervention plan. Records of children's behavioral improvements include indicators such as classroom attention duration, frequency of fidgeting, and number of peer interactions; records of language ability changes include indicators such as fluency of speech and vocabulary growth; and records of medical examination results include scores on neuropsychological assessment scales and electroencephalographic indicators (if applicable). The data collection request includes a list of effect indicators, a data collection form template, and a submission deadline.

[0132] Step S312: Receive intervention effect data returned by each terminal, classify and organize the received intervention effect data, and divide it into home care scenario effect data, school adaptation scenario effect data, and medical consultation scenario effect data.

[0133] The system receives intervention effect data from various terminals. User terminals return effect data for home care scenarios (such as behavioral changes observed at home), teacher terminals return effect data for school-adapted scenarios (such as improvements in classroom behavior and completion of learning tasks), and doctor terminals return effect data for medical consultation scenarios (such as medical examination results). The received data is categorized and organized, stored according to scenario type, and then cleaned and validated to remove data with excessive outliers and missing values.

[0134] Step S313: Compare the categorized and organized intervention effect data with the expected effect indicators in the individualized brain health intervention plan, and output the deviation between the actual effect and the expected effect through the deviation identification module.

[0135] The categorized and organized intervention effect data is compared with the expected effect indicators in the individualized brain health intervention plan. For example, in a school-adapted scenario, the expected effect indicator is "classroom attention duration reaches 25 minutes," while the actual collected data is 20 minutes, resulting in a deviation of 5 minutes. The deviation identification module quantifies the deviation by calculating the percentage difference between the actual effect data and the expected effect indicator. When the absolute value of the percentage difference exceeds a preset threshold, it is determined to be a significant deviation. The deviation report is output, including information such as the effect indicator name, expected value, actual value, deviation value, and deviation percentage.

[0136] Step S314: Based on the updated multimodal dataset of the child, output the reasons for the bias. When the bias does not match the family performance data, the reason is marked as family performance bias; when the bias does not match the school adaptation data, the reason is marked as school adaptation bias; when the bias does not match the changes in the child's individual data, the reason is marked as individual situation change bias.

[0137] By combining updated multimodal datasets of children, the causes of biases are analyzed. For example, if family implementation data shows that parents did not implement family behavior training at the frequency required by the intervention plan, and the effect bias in the family care scenario is significant, the cause is marked as family implementation bias. If school adaptation data shows that classroom interaction adaptation strategies were not strictly implemented (e.g., teachers did not increase attention frequency as required), and the effect bias in the school scenario is significant, it is marked as school adaptation bias. If updated brain structure imaging data shows new changes in the development of a certain brain region of the target child, leading to changes in individual circumstances and thus causing effect bias, it is marked as individual circumstance change bias. Causal analysis is achieved through a combination of rule-based reasoning and machine learning. A series of association rules between the causes of biases and data features are pre-defined, and a trained classification model is used to predict the causes of biases.

[0138] Step S315: Generate an intervention plan and adjust the strategy based on the cause of the deviation. When the cause is marked as family implementation deviation, adjust the description of the intervention measures in the family care scenario and add details of the implementation guidance. When the cause is marked as school adaptation deviation, coordinate and supplement auxiliary intervention methods in the school adaptation scenario. When the cause is marked as individual situation change deviation, regenerate the appropriate intervention measures based on the updated multimodal data.

[0139] Based on the causes of the deviations, corresponding intervention plans and adjustment strategies are generated. For family implementation deviations, the wording of intervention measures in the home care scenario is adjusted to make vague implementation requirements more specific. For example, "daily attention training" is changed to "30 minutes after dinner each day, parents accompany their children in attention card games for 15 minutes each time, using a timer to record the child's attention span," adding more detailed implementation guidance. For school adaptation deviations, supplementary auxiliary intervention methods are coordinated, such as equipping teachers with classroom behavior observation assistants to help record and remind them of the implementation of intervention measures. For deviations due to changes in individual circumstances, based on the updated multimodal dataset, the generation process of individualized brain health intervention plans is re-executed, adjusting the content and intensity of intervention measures to adapt to changes in the child's individual circumstances.

[0140] Step S316: Modify the intervention measures in the individualized brain health intervention plan according to the adjustment strategy, and update the implementation details and expected effect indicators in the individualized brain health intervention plan.

[0141] Based on the intervention plan adjustment strategy, the corresponding intervention measures in the individualized brain health intervention plan were modified. The modifications included detailed implementation steps, responsible personnel, resource requirements, and corresponding expected outcome indicators (such as the adjusted attention duration target). The modified intervention plan was managed through a version control system to retain historical versions for easy traceability and comparison.

[0142] Step S317: The adjusted individualized brain health intervention plan is simultaneously sent to the user terminal, teacher terminal, and doctor terminal, triggering a new round of multi-terminal collaborative feedback process. Feedback information on the adjusted intervention plan is collected through the multi-terminal collaborative feedback process.

[0143] The revised individualized brain health intervention plan will be simultaneously sent to all relevant terminals, replacing the old version. At the same time, a new round of multi-terminal collaborative feedback will be triggered, prompting users, teachers, and doctors to provide feedback on the revised intervention plan, collecting problems and suggestions encountered during use, in order to further optimize the intervention plan.

[0144] This embodiment involves the collection and use of multimodal data on children's brain health, which may include privacy-sensitive data such as genetic data and medical imaging data. To protect this privacy-sensitive data, the following technical measures are adopted: During data transmission, HTTPS encryption protocol and end-to-end encryption technology are used to ensure that data is not leaked during transmission; during data storage, AES-256 encryption algorithm is used to encrypt and store sensitive data, and database access employs strict identity authentication and access control mechanisms; during data use, sensitive data is anonymized, such as removing children's real names, ID numbers, and other identifying information, and using anonymized child identification; a data access audit log is established to record all access and operation behaviors of sensitive data for traceability and auditing. These technical measures ensure the security of privacy-sensitive data and prevent data leakage.

[0145] In an exemplary embodiment, a children's brain health intelligent question-and-answer system integrating multimodal data is provided. This system can be a terminal, server, etc., and its internal structure diagram is shown in Figure 2. The system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a children's brain health intelligent question-and-answer method integrating multimodal data. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or it can be buttons, a trackball, or a touchpad set on the shell of a children's brain health intelligent question and answer system that integrates multimodal data, or it can be an external keyboard, touchpad, or mouse, etc.

[0146] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A children's brain health intelligent question-and-answer method integrating multimodal data, characterized in that, The method includes: receiving a user-initiated multi-scenario Q&A request for children's brain health, wherein the multi-scenario Q&A request for children's brain health includes a description of the brain health problem to be consulted, an associated child identity identifier, and a multi-scenario service demand identifier, wherein the multi-scenario service demand identifier is used to indicate the corresponding home care scenario, school adaptation scenario, or medical consultation scenario; based on the child identity identifier, calling a preset multimodal database for children's brain health, obtaining a multimodal data set corresponding to the child identity identifier, and initially associating the multimodal data set with the multi-scenario service demand identifier, wherein the multimodal data set includes the child's brain structure imaging data, language behavior record data, gene association data, and phased questionnaire feedback data; Cross-modal semantic mapping is performed on the multimodal data set with preliminary association tags and the brain health problem description to construct a unified semantic space for multimodal data. Potential association data related to the brain health problem description is mined within this unified semantic space to generate intervention criteria for brain health problems. Based on the intervention criteria and the multi-scenario service requirement identifiers, question-and-answer response content and individualized brain health intervention plans conforming to the clinical knowledge base for children's brain health are generated. The question-and-answer response content includes an analysis of the problem's causes and key scientific responses, while the individualized brain health intervention plan includes scenario-specific implementation details. The question-and-answer response content and the individualized brain health intervention plan are simultaneously sent to the user terminal and the association learning platform. The school's teacher terminals and the doctor terminals of cooperating medical institutions trigger a multi-terminal collaborative feedback process to collect feedback information from each terminal on the question-and-answer response content and the use of the individualized brain health intervention plan; the cross-modal semantic mapping processing of the multimodal data set with preliminary association tags and the brain health problem description to construct a unified semantic space for multimodal data includes: extracting features from various types of data in the multimodal data set, extracting brain region morphological features from brain structural imaging data, expression coherence features from language behavior recording data, genetic locus association features from gene association data, and behavioral tendency features from phased questionnaire feedback data; and combining the extracted brain region morphological features, expression coherence features, and the description of brain health problems with the description of brain health problems. Genetic locus association features and behavioral tendency features are input into a pre-trained cross-modal semantic mapping model. The cross-modal semantic mapping model includes a feature alignment layer, a semantic transformation layer, and a spatial fusion layer. Through the feature alignment layer of the cross-modal semantic mapping model, different types of features are processed to unify their dimensions, converting the brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features into feature vectors of the same dimension. Through the semantic transformation layer of the cross-modal semantic mapping model, each feature vector after unification of dimensions is mapped to a preset children's brain health semantic dictionary space, generating a semantic label corresponding to each feature vector. The semantic label includes brain health indicator category and data association scenario identifier.Through the spatial fusion layer of the cross-modal semantic mapping model, semantic associations between different feature vectors are established based on the semantic labels corresponding to each feature vector, forming an initial semantic network. Semantic parsing is performed on the brain health problem description to extract core semantic elements, which include the brain health domain involved in the problem, the child development stage of concern, and the expected solution dimension. The extracted core semantic elements are integrated into the initial semantic network, and the weights of each semantic association in the initial semantic network are adjusted, strengthening associations that semantically overlap with the core semantic elements and weakening associations that do not. The adjusted semantic network is structurally optimized by deleting redundant semantic association paths and supplementing missing key association nodes, forming a unified semantic space for multimodal data. The rationality of data associations in the unified semantic space for multimodal data is verified by comparing them with data association rules in the children's brain health clinical knowledge base, eliminating semantic association pairs that do not conform to the data association rules, and retaining semantic association pairs that conform to the data association rules to update the unified semantic space for multimodal data.

2. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 1, characterized in that, The process of generating a question-and-answer response and an individualized brain health intervention plan based on the intervention criteria for brain health problems and the multi-scenario service demand identifiers includes: decomposing the intervention criteria for brain health problems to separate direct evidence data directly related to the description of the brain health problems and indirect evidence data for supplementary explanation; arranging the direct evidence data according to the logical hierarchy of the question and answer, with the first layer arranging evidence data related to the causes of the problems, the second layer arranging evidence data related to the impact of the problems, and the third layer arranging evidence data related to the responses to the problems, and generating the main framework of the question-and-answer response content based on each logical hierarchy; and then, based on the pre-configured clinical knowledge base for children's brain health, refining the main framework... Each layer of the data is supplemented with explanations, adding corresponding clinical research conclusions and scientific explanations to improve the main framework of the Q&A response content and form a complete Q&A response content; the target service scenario is determined based on the multi-scenario service demand identifier. When the multi-scenario service demand identifier indicates a home care scenario, the focus is on intervention directions related to daily home care; when the multi-scenario service demand identifier indicates a school adaptation scenario, the focus is on intervention directions related to adjusting the school learning environment; when the multi-scenario service demand identifier indicates a medical consultation scenario, the focus is on intervention directions related to professional medical examination and treatment recommendations; for the determined target service scenario, intervention reference data related to the target service scenario is extracted from the indirect data. The intervention reference data includes data on intervention cases involving children in the same scenario and data on the effectiveness of intervention methods adapted to the scenario. Based on the intervention reference data and the intervention principles in the clinical knowledge base of children's brain health, a preliminary intervention plan module is generated for the target service scenario. The preliminary intervention plan module for the home care scenario includes detailed rules for guiding daily behavior and suggestions for adjusting the home environment; the preliminary intervention plan module for the school adaptation scenario includes classroom interaction adaptation strategies and learning task adjustment plans; the preliminary intervention plan module for the medical consultation scenario includes a list of recommended examination items and suggestions for phased treatment. The preliminary intervention plan module is then adapted to the children's multimodal data set, combined with the brain region development data in the children's brain structural imaging data and language behavior recording data. Based on the ability level and genetic risk indications in the gene association data, the intervention clauses in the preliminary intervention plan module that do not match the children's multimodal data are modified, and the intervention focus in the preliminary intervention plan module that matches the children's multimodal data is strengthened; the adjusted intervention plan module is prioritized, and sorting rules are set according to the urgency, implementation difficulty, and expected effect of the intervention measures. The execution order of each intervention measure is determined according to the sorting rules to form a structured individualized brain health intervention plan; the individualized brain health intervention plan is associated with the question and answer response content, and a one-to-one correspondence is established between the key points of the question and answer response content and the intervention measures in the individualized brain health intervention plan through the association tags, which adapts to the user's reading and implementation needs.

3. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 1, characterized in that, The multi-terminal collaborative feedback process, which collects feedback information from each terminal regarding the Q&A response content and the individualized brain health intervention plan, includes: simultaneously generating a feedback information collection template when sending the Q&A response content and the individualized brain health intervention plan to each terminal; the feedback information collection template includes feedback items on the comprehension of the Q&A response content, feasibility feedback items on the individualized brain health intervention plan, and supplementary needs feedback items; setting a feedback submission time limit for the feedback information collection template, with the time limit length set according to the urgency of the child's brain health problem, and the feedback submission time limit for urgent problems being shorter than that for regular problems; binding the feedback information collection template with the Q&A response content and the individualized brain health intervention plan and sending it to user terminals, teacher terminals, and doctor terminals, establishing an association between the feedback template received by each terminal and the corresponding content; receiving feedback information submitted by each terminal in real time within the feedback submission time limit, classifying and storing the received feedback information according to terminal type into user feedback data, teacher feedback data, and doctor feedback data; and performing content analysis on the classified feedback data. The core feedback points are extracted from each set of feedback data. User feedback data includes content related to comprehension difficulties and records related to execution difficulties. Teacher feedback data includes records of adaptation conflicts in campus scenarios and conflicts with teaching arrangements. Doctor feedback data includes supplementary medical advice and suggestions for adjusting intervention plans. The extracted core feedback points are then associated with the unified semantic space of the multimodal data, updating the association weights of related data in the unified semantic space and strengthening the association weights of data semantically related to the feedback points. Based on the adjustment suggestions in the core feedback points, the question-and-answer response content and the personalized brain health intervention plan are iteratively optimized. Expressions indicating comprehension difficulties in the question-and-answer response content are modified, and intervention measures indicating execution difficulties in the personalized brain health intervention plan are adjusted. The optimized question-and-answer response content and the personalized brain health intervention plan are then synchronized back to each terminal, completing a multi-terminal collaborative feedback loop. When each terminal submits feedback on the optimized content, the above feedback collection and optimization steps are repeated.

4. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 1, characterized in that, The feature alignment layer of the cross-modal semantic mapping model performs dimensional unification processing on different types of features, converting the brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features into feature vectors of the same dimension. This includes: obtaining the original dimensional parameters of the brain region morphological features, determined by the acquisition resolution of the brain structure imaging data and the number of extracted brain regions; obtaining the original dimensional parameters of the expression coherence features, determined by the acquisition duration of the language behavior record data and the number of extracted language feature indicators; obtaining the original dimensional parameters of the genetic locus association features, determined by the number of detection sites in the gene association data and the number of indicators in the association analysis; obtaining the original dimensional parameters of the behavioral tendency features, determined by the number of questions in the phased questionnaire feedback data and the number of statistical behavioral dimensions; and inputting the original dimensional parameters of the brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features into the feature dimension calculation module to calculate the full dimension. The target dimension parameter satisfies all feature transformation requirements. This target dimension parameter is greater than or equal to the maximum value among the original dimension parameters of various features and is compatible with dimension transformation algorithms for various features. For the brain region morphological features, a feature interpolation algorithm is used to convert their original dimensions to the target dimension, preserving key structural information in the brain region morphological features during the conversion process. For the expression coherence features, a feature expansion algorithm is used to convert their original dimensions to the target dimension, achieving dimension expansion by supplementing the associated feature values ​​of language expression logic. For the genetic locus association features, their original dimensions are converted to the target dimension, and the original genetic locus features are grouped and recombined, with each group of features corresponding to a new dimension, maintaining the genetic association relationship. For the behavioral tendency features, a feature mapping algorithm is used to convert their original dimensions to the target dimension, establishing a mapping relationship between the original behavioral features and the target dimension features, maintaining the accuracy of the behavioral tendency. The brain region morphological features, expression coherence features, genetic locus association features, and behavioral tendency features, after conversion to the target dimension, are normalized to form a feature vector of a unified dimension.

5. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 2, characterized in that, The step of extracting intervention reference data related to the target service scenario from the indirect evidence data for the determined target service scenario includes: tagging the indirect evidence data with scenario tags; adding scenario tags to each piece of indirect evidence data based on the data source and scenario information described in the data; scenario tags include family scenario tags, school scenario tags, and medical scenario tags; filtering indirect evidence data with corresponding scenario tags based on the determined target service scenario; when the target service scenario is a home care scenario, filtering indirect evidence data with family scenario tags; when the target service scenario is a school adaptation scenario, filtering indirect evidence data with school scenario tags; target service scenario When considering medical consultation scenarios, indirect evidence data tagged with "medical scenario" is filtered. This filtered indirect evidence data is then categorized into successful intervention case data, partially effective case data, and ineffective case data. Successful intervention case data represents data showing improvement in children's brain health after intervention; partially effective case data represents data showing partial improvement in children's brain health after intervention, but not meeting expectations; and ineffective case data represents data showing no improvement in children's brain health after intervention. Details of the intervention methods are extracted from the successful intervention case data. These details include the specific implementation methods, frequency, duration, and children's responses during the intervention. Details serve as core intervention reference data; factors influencing intervention effectiveness are extracted from partially effective case data, including family environment factors, school cooperation factors, and individual differences among children. These factors serve as supplementary intervention reference data to highlight potential influencing conditions; reasons for intervention failure are extracted from ineffective case data, including specific manifestations of mismatch between intervention measures and children's conditions, oversights in the intervention implementation process, and external environmental interference factors. These reasons for intervention failure serve as risk warning reference data to avoid similar problems in the intervention plan generation process; the core intervention reference data... The data, the supplementary intervention reference data, and the risk warning reference data are sorted according to their importance, with core intervention reference data having a higher priority than supplementary intervention reference data, and supplementary intervention reference data having a higher priority than risk warning reference data. The sorted reference data is then integrated, duplicate reference information is removed, and semantically similar reference content is merged to form an intervention reference data set. This intervention reference data set is then matched and verified against the intervention requirements of the target service scenario to determine whether each intervention requirement has corresponding intervention reference data support. If a requirement lacks supporting reference data, the data is re-filtered and supplemented based on the indirect reference data set.

6. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 3, characterized in that, The iterative optimization of the Q&A response content and the individualized brain health intervention plan based on the adjustment suggestions in the core feedback points includes: dividing the adjustment suggestions in the core feedback points into Q&A response content adjustment suggestions and individualized brain health intervention plan adjustment suggestions according to the adjustment objects; for the Q&A response content adjustment suggestions, extracting the content corresponding to the comprehension obstacles, identifying the triggering conditions of the comprehension obstacles, and when the comprehension obstacle triggering condition is that the proportion of professional terms in the Q&A response content exceeds a preset threshold, replacing the professional terms with colloquial expressions and adding simple explanations; when the comprehension obstacle triggering condition is that the overlap of the logical levels of the Q&A response content exceeds a preset threshold, readjusting the logic of the Q&A response content. The structure establishes connections between parts by adding transitional statements; the question-and-answer response content is modified based on comprehension obstacle triggering conditions, and then tested. By simulating the reading habits of the target user, the content is judged to meet readability requirements. If the result is that it does not meet readability requirements, adjustments are made until it does; for individualized brain health intervention program adjustment suggestions, intervention measures corresponding to executive disorders are extracted, and the triggering conditions of executive disorders are identified. If the triggering condition of an executive disorder is that the matching degree between the intervention measure and the family conditions is lower than a preset threshold, it is replaced with an alternative intervention measure that matches the family conditions to the preset threshold; executive disorder triggering conditions. When the matching degree between the intervention measures and school resources is lower than a preset threshold, supplementary intervention measures with a matching degree that meets the preset threshold are coordinated and added. When the execution obstacle trigger condition is that the matching degree between the intervention measures and medical resources is lower than a preset threshold, alternative medical solutions with a matching degree that meets the preset threshold with nearby medical resources are recommended. Based on the execution obstacle trigger condition, the intervention measures in the individualized brain health intervention plan are replaced or modified. After modification, the adjusted intervention measures are tested for compatibility with the children's multimodal data set, and the matching degree between the adjusted intervention measures and the children's multimodal data is output. If the matching degree meets the preset threshold, it is confirmed that there is no obvious adaptation conflict. The adjusted question-and-answer response content is compared with the individualized brain health intervention plan. A correlation check is performed, outputting the correspondence between the key points of the Q&A response and the adjusted intervention measures. If the correspondence is broken, the key points of the Q&A response are modified to meet the preset requirements. The adjusted Q&A response and individualized brain health intervention plan are submitted to the pediatric brain health expert review module. The expert review module refers to the latest pediatric brain health clinical knowledge base and outputs the review results to determine whether the content conforms to clinical standards. If the review result is in compliance with clinical standards, it is determined as the optimized final content. If the review result is not in compliance with clinical standards, it is adjusted again according to the modification suggestions in the review result until the review result is in compliance with clinical standards, forming the final optimized content.

7. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 1, characterized in that, The process of integrating the extracted core semantic elements into the initial semantic network and adjusting the weights of each semantic relationship in the initial semantic network includes: assigning weights to the core semantic elements; setting weight values ​​based on the semantic overlap between the core semantic elements and the description of the brain health problem; assigning a first weight value to core semantic elements with a semantic overlap reaching a first preset threshold, and assigning a second weight value to core semantic elements with a semantic overlap below a second preset threshold; wherein the first weight value is greater than the second preset threshold and the second weight value is less than the first preset threshold; traversing all semantic relationships in the initial semantic network and identifying the semantic labels corresponding to the feature vectors involved in each semantic relationship; matching the semantic labels involved in each semantic relationship with the core semantic elements, and calculating the matching degree between the semantic labels and the core semantic elements using a semantic similarity algorithm; the higher the semantic overlap between the semantic labels and the core semantic elements, the larger the matching degree value; calculating the weight adjustment coefficient for each semantic relationship based on the weight value and matching degree of the core semantic elements; the weight adjustment coefficient is the product of the weight value of the core semantic element and the matching degree; when a semantic relationship involves multiple core semantic elements, multiple weights are taken. The average value of the adjustment coefficients is used as the final adjustment coefficients. The original weights of each semantic relationship in the initial semantic network are multiplied by the corresponding weight adjustment coefficients to obtain the adjusted semantic relationship weights. The larger the product of the original weight and the adjustment coefficient, the higher the adjusted weight; the smaller the product, the lower the adjusted weight. The adjusted semantic relationship weights are normalized to maintain the sum of all semantic relationship weights at the initial set value. Semantic relationships with adjusted weights below a preset weight threshold are identified, judged as irrelevant, and deleted from the semantic network. Semantic relationships with adjusted weights above a preset high weight threshold are identified, and their logical consistency is checked through a logic verification module. If a logical contradiction exists, the weight is reduced to a preset reasonable range. The semantic network with adjusted weights is compared with the core semantic elements for overall matching. The matching score between the semantic network and the core semantic elements is output through a matching score calculation module. When the matching score reaches a preset threshold, the weight adjustment is completed. If the matching score does not reach the preset threshold, the weights of the core semantic elements are readjusted, and the above weight adjustment steps are repeated until the matching score reaches the preset threshold.

8. The intelligent question-and-answer method for children's brain health that integrates multimodal data according to claim 1, characterized in that, The method further includes a step of updating the children's brain health multimodal database based on multi-terminal collaborative feedback information. This step includes: extracting data update prompts from the collected multi-terminal collaborative feedback information, wherein the data update prompts include key data types missing in the multimodal data set and prompts indicating insufficient timeliness of existing data; determining the data types to be supplemented according to the data update prompts; when the data update prompt indicates missing latest brain structure imaging data, the data type to be supplemented is updated brain structure imaging data; when the data update prompt indicates insufficient timeliness of language behavior recording data, the data type to be supplemented is newly added language behavior recording data; when the data update prompt indicates that gene-related data needs to be updated, the data type to be supplemented is updated gene-related data; when the data update prompt indicates missing phased questionnaire feedback data, the data type to be supplemented is newly added questionnaire feedback data; and sending data collection requests to associated data source terminals, wherein the data source terminals include home data collection terminals, school data collection terminals, and medical data collection terminals. The institutional data acquisition terminal receives supplementary data from various data source terminals, performs format unification processing on the received supplementary data to ensure its format is consistent with the existing data in the children's brain health multimodal database, associates the format-unified supplementary data with the corresponding child identification mark, establishes a correspondence between the supplementary data and the child's historical multimodal data set through the association mark, adds the associated supplementary data to the multimodal data set corresponding to the child in the children's brain health multimodal database, and updates the timestamp information of the multimodal data set; based on the updated multimodal data set, re-executes cross-modal semantic mapping processing, adjusts the semantic associations in the unified semantic space of the multimodal data to match the updated data, and stores the updated multimodal data set and the adjusted unified semantic space of the multimodal data.

9. A children's brain health intelligent question-and-answer system integrating multimodal data, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the intelligent question-and-answer method for children's brain health that integrates multimodal data according to any one of claims 1 to 8 by executing the machine-executable instructions.

Images