Search

CN-121983246-A - Modeling and intervention simulation method and system for Chinese language reading and writing difficulty children based on digital twin brain technology

CN121983246ACN 121983246 ACN121983246 ACN 121983246ACN-121983246-A

Abstract

The invention discloses a modeling and intervention simulation method and system for a Chinese language reading and writing difficulty child based on a digital twin brain technology, wherein a data acquisition module acquires behavior data and brain imaging data of the Chinese language reading and writing difficulty child and a normal control child, a model construction module is used for constructing a Chinese language reading and writing digital twin base model to simulate the function split and consciousness in the brain reading and writing process to form dynamic states, an individuation modeling module adopts a double constraint optimization method to perform individuation fine tuning on the Chinese language reading and writing digital twin base model to establish an individuation digital twin model, an intervention simulation module is used for verifying and performing mechanism analysis on the individuation digital twin model and simulating different intervention strategies to predict the improvement effect of the individuation reading and writing behaviors and brain representation, and an individuation intervention scheme is recommended and output.

Inventors

  • YANG YANG
  • Zhang Aoxue

Assignees

  • 中国科学院心理研究所

Dates

Publication Date
20260505
Application Date
20260123

Claims (13)

  1. 1. A modeling and intervention simulation method for Chinese language reading and writing difficulty children based on a digital twin brain technology is characterized by comprising the following steps: acquiring behavior data and brain imaging data of children with difficult Chinese reading and writing and normal control children; constructing a Chinese read-write digital twin base model comprising a Siamese dual coding network, a multitask discrimination branch, a Chinese component sensitive module and a circulating feedback mechanism, and simulating functional shunting and consciousness of a brain in the read-write process to form a dynamic state; Performing individual fine adjustment on the Chinese read-write digital twin base model by adopting a dual constraint optimization method of behavioral data alignment and brain imaging data alignment, and establishing an individual digital twin model; verifying and analyzing a mechanism of the personalized digital twin model to reveal a causal mechanism of read-write difficulty; Based on the personalized digital twin model, different intervention strategies are simulated, the improvement effect of the intervention strategies on the individual read-write behaviors and brain characterization is predicted, and a personalized intervention scheme is recommended for the individual.
  2. 2. The modeling and intervention simulation method for the Chinese language reading and writing difficulty children according to claim 1, wherein the behavior data are response time and accuracy obtained by a voice discrimination task, a font discrimination task, a component character discrimination task and a shape similarity discrimination task in a trial-by-trial mode, the brain imaging data are task state and resting state whole brain BOLD signals acquired by using functional magnetic resonance imaging, wherein key interested areas comprise a left shuttle-shaped gyrus vision word shape area, a left temporal-top joint area, frontal lower gyrus, mid-frontal gyrus and primary vision cortex, and the key interested areas are used for alignment constraint of internal characterization of a Chinese language reading and writing digital twin base model.
  3. 3. The modeling and intervention simulation method for the children with difficulty in reading and writing according to claim 1, wherein four key neurophysiologic parameters including global nerve gain G, internal noise sigma, cyclic feedback gain R and neuron excitation threshold T are introduced in the process of establishing a twin base model of the Chinese reading and writing, a Bayesian optimization or genetic algorithm global search method is adopted, the total loss of individuals is taken as an evaluation function, and the optimal (G, sigma and R, T) combination is found in a search parameter space; In the Chinese read-write digital twin base model, the global nerve gain G is used for describing the balance of cortex excitability and inhibitive property and is realized by multiplying coefficients after convolution output of each layer; the internal noise sigma simulates neural activity inherent random fluctuation by adding Gaussian noise after convolution calculation of each layer; the cyclic feedback gain R determines the feedback cyclic intensity by multiplying the cyclic connection weight of each layer by a coefficient; The neuron excitation threshold T controls the minimum input intensity required for the neuron to transition from the resting state to the activated state by adjusting the threshold parameter of the activation function.
  4. 4. The modeling and intervention simulation method for the children with difficulty in reading and writing according to claim 1, wherein the specific method for constructing the twin base model for the Chinese reading and writing is as follows: aiming at the paired stimulation discrimination paradigm, a Siamese dual coding network is constructed; adopting a parallel convolution branch structure shared by two parameters to simulate a human brain paired stimulation parallel processing mechanism, and sharing a visual coding trunk; each branch adopts a V1-V2-V4-IT four-layer circular convolution structure, and hierarchical visual characteristics are extracted; Introducing a fusion layer after the IT layer, and integrating double-branch characteristics by adopting a splicing and differential strategy; After sharing the visual coding trunk, four parallel task discrimination branches are derived, and the neural mechanisms of different information paths are processed in parallel when the brain reads and writes.
  5. 5. The modeling and intervention simulation method for children with difficulty in reading and writing according to claim 4, wherein the four-layer cyclic convolution structure of v1→v2→v4→it corresponds to the neuroanatomy of the primate ventral view path, specifically: the V1 layer corresponds to the primary visual cortex, and basic visual characteristics such as edges, directions and the like are extracted; The V2 layer corresponds to the secondary visual cortex, and integrates simple features to form more complex shape characterization; the V4 layer corresponds to the advanced visual zone, and middle layer characteristics such as shape, texture and the like are extracted; the IT layer corresponds to the temporal lobe cortex, forming a high-level representation of objects and words.
  6. 6. The modeling and intervention simulation method for the Chinese language difficult to read and write children is characterized in that a fusion layer is introduced after an IT layer, and a splicing and difference strategy is adopted to integrate double-branch characteristics; the training adopts supervised multitasking classification, each discrimination task has a definite classification label, and the cross entropy loss is used for supervising output.
  7. 7. The modeling and intervention simulation method for the Chinese language reading and writing difficulty children according to claim 6, wherein a component attention branch is additionally arranged behind the V4 layer, chinese character radicals and high-frequency components are selectively responded through a pre-trained component detector, an attention heat map is generated, and the element product acts on the original feature map, so that the characterization capability of a key component area in the subsequent IT layer processing is enhanced.
  8. 8. The method for modeling and intervening simulation of a difficult-to-read-write chinese child according to claim 4, wherein the four parallel task discrimination branches adopted comprise: A word sound judging branch corresponding to the back side voice channel and judging whether the pronunciation of the two words is the same from the left temporal top joint area to the frontal lower return; A font discrimination branch corresponding to the side view word shape region for judging whether the two stimuli are the same Chinese character; A component character forming branch corresponding to the left middle-return form processing passage and judging whether the component can form a part of the whole character; The shape is similar to the branch, is used for pure visual graph judgment, and is used as a visual base line of children with difficulty in reading and writing.
  9. 9. The modeling and intervention simulation method for the Chinese language reading and writing difficulty children according to claim 4, wherein a loop feedback mechanism of local loop connection is introduced into a four-layer loop convolution structure of V1, V2, V4 and IT, information update of a plurality of loop periods is executed for each input in each layer, and a feedback processing process in the formation of visual consciousness is simulated.
  10. 10. The method for modeling and intervening simulation of a Chinese read-write difficulty child according to claim 1, wherein the behavioral data alignment uses individual actual response as a supervision signal to fit an error mode and response time characteristics of the individual actual response, the brain imaging data alignment matches an internal representation of a model with an individual functional magnetic resonance imaging activation mode, and a corresponding relation between each layer of the Chinese read-write digital twin base model and a brain region is established, wherein an IT layer corresponds to a left shuttle-shaped visual word shape region, a voice branch hidden layer corresponds to a left temporal-top joint region and a frontal lower back, a component module corresponds to a left middle frontal back, and an early convolution layer corresponds to a primary visual cortex.
  11. 11. The modeling and intervention simulation method for the Chinese language reading and writing difficulty children according to claim 10, wherein the verification and mechanism analysis of the personalized digital twin model is carried out, and the method specifically comprises the following steps: verifying the prediction accuracy of the individual behaviors by verifying the individualized digital twin model on the unobserved test time; And comparing the similarity of the internal representation of the personalized digital twin model and the brain region multisubstance activation mode, simulating specific channel damage or parameter adjustment in the personalized digital twin model through module ablation and parameter intervention experiments, deducing the influence of the specific channel damage or parameter adjustment on the behavior, and revealing a causal mechanism.
  12. 12. The modeling and intervention simulation method for children with difficult reading and writing in Chinese according to claim 1, wherein the specific method for recommending personalized intervention schemes for individuals is as follows: Based on the parameters and the characterization characteristics of the individualized digital twin model, subtype classification is carried out on children with difficult reading and writing, wherein the adopted subtype classification comprises pure voice mapping obstacle type, visual attention defect type and mixed defect type; Establishing a 'silicon-based brain' virtual intervention simulation platform, simulating the influence of various imaginary intervention means on reading and writing in a computer through an individuation digital twin model, wherein the intervention strategies comprise voice strengthening training simulation, visual word shape training simulation and attention regulating training simulation; And combining simulation conditions of different intervention strategies, predicting the intervention effect, and recommending an optimal intervention scheme for the child individual with difficult reading and writing.
  13. 13. A digital twin brain technology-based modeling and intervention simulation system for a Chinese language difficult to read and write child, the system comprising: The data acquisition module is used for acquiring multi-mode data of children with difficult Chinese reading and writing and normal contrast children; The model construction module is used for constructing a Chinese read-write digital twin base model, and the Chinese read-write digital twin base model comprises a visual coding unit, a task discriminating unit, a component sensitive unit and a circulating feedback unit; the individualized modeling module is used for performing individualized fine tuning on the Chinese read-write digital twin base model by adopting a dual constraint optimization method of behavior data alignment and brain imaging data alignment, and establishing an individualized digital twin model; the intervention simulation module is used for simulating different intervention strategies and predicting intervention effects; and the output module is used for outputting the personalized diagnosis report and the personalized intervention scheme recommendation.

Description

Modeling and intervention simulation method and system for Chinese language reading and writing difficulty children based on digital twin brain technology Technical Field The invention belongs to the technical field of intersection of cognitive neuroscience and artificial intelligence, and particularly relates to a modeling and intervention simulation method and system for children with difficulty in reading and writing based on a digital twin brain technology, which are used for revealing a neural mechanism with difficulty in reading and writing and simulating and predicting an intervention effect. Background The children with difficult reading and writing are common learning disorders, and 5% -17.5% of school children are affected. The existing research is based on brain imaging data analysis at the group level, individual differences are difficult to be described, and mechanism deduction and intervention prediction cannot be realized. Traditional functional magnetic resonance imaging and behavior association researches can reveal some brain region abnormalities, but cannot dynamically simulate interaction between brains and causal relation between the interaction and behaviors. In addition, the existing method is mainly focused on conscious-plane processing, neglects the role of unconscious processing in read-write automation, and limits the integrity of mechanism interpretation. Over the past three decades, a variety of noninvasive brain imaging techniques such as functional magnetic resonance imaging, magnetoencephalography, electroencephalography, etc. have been widely applied in this field, and functional abnormalities in reading and writing difficulties in a plurality of brain regions from the cortex to the subcortical have been gradually revealed. Studies have shown that read-write difficulties are mainly associated with inadequate activation of the left language network, involving brain areas including superior temporal, medial temporal, temporal-parietal intersections and visual word areas, where abnormal activity affects word recognition and speech decoding, resulting in reduced read-write efficiency. However, how these abnormal activities specifically lead to difficulty in reading and writing, and the core pathogenic mechanism thereof has not been clarified yet. The mechanism of synergy between the different regions of the brain has not been systematically elucidated, subject to the inherent limitations of traditional research approaches. Cross-language research shows that the neural phenotype with difficult reading and writing has both universality and language specificity. Taking Chinese as an example, as an ideographic system, there is a significant difference from alphabetic writing in the correspondence between visual structure and shape and meaning. The research shows that the core brain area with difficult Chinese reading and writing is located in the left forehead middle back, which is different from the common temporal top lobe abnormality in alphabetic writing. Therefore, the mechanism research and intervention system of Chinese read-write difficulty needs to be constructed based on the language characteristics, and the research framework of alphabetic writing cannot be simply adopted. Moreover, due to the limitation of technology and ethics, the mechanism of the brain of the children with read-write disorder can only be indirectly measured, and the working principle of the brain is difficult to be fundamentally understood and the causal relationship is built. Because of lack of guidance of brain mechanism research, the design of dyskinesia is designed from the aspect of phenomenology, the core mechanism is difficult to reach, and the intervention effect is uneven. In recent years, the combination of digital twin technology and deep learning provides a new approach for constructing a 'brain-behavior' computable model at an individual level. Digital twinning techniques can build interpretable brain function models at the individual level, enabling spanning from data description to mechanism simulation. By combining the deep neural network, an individual deep neural network model can be further constructed, and the cognitive behavior and brain activity characteristics of an individual are reproduced in the virtual model by regulating and controlling parameters such as nerve excitability, connection strength and the like, so that 'neural digital twin' is realized. The individualized deep neural network model constructed by the method has successfully revealed the neurophysiologic mechanism of learning disorder in alphabetic language. The model can be used for highly fitting the behavior and brain activity characteristics of the learning disorder children, and simulating the phenomena of reduced learning accuracy, slow learning rate, nerve hyperexcitation, reduced differentiation degree of the neural characterization of the digital problem and the like. However, the existing neural digita