CN-122025026-A - Child autism rehabilitation training strategy generation method and system

CN122025026ACN 122025026 ACN122025026 ACN 122025026ACN-122025026-A

Abstract

The invention discloses a method and a system for generating a child autism rehabilitation training strategy, and relates to the technical field of intelligent rehabilitation. The three core defects of strong subjectivity, strategy templatization and adjustment lag of data in the traditional rehabilitation training are overcome. The technical scheme is characterized in that a time sequence convolution network and a gating cross-modal attention mechanism are fused to achieve accurate association of symptom causes, multi-step reasoning is conducted based on a dynamic knowledge graph to identify hidden risks, neural development logic ordering and dynamic feasibility verification are achieved through a multi-level fusion algorithm, and feedback data is utilized to drive continuous optimization of the knowledge graph. The full-flow intelligent effect from objective evaluation, accurate intervention to continuous optimization is achieved, high-accuracy, strong-prospective and scientific and effective rehabilitation training support is provided for autistic children, and pertinence, predictability and long-term effectiveness of rehabilitation training are remarkably improved.

Inventors

Wang Chengqu
WU ZHIFENG
ZHANG YUPING
LU SIYU
ZHU CHUNXIAO
Fan Qiongli
WEI JUN
MING LI
Su Lisha

Assignees

中国人民解放军陆军军医大学第二附属医院

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. The method for generating the child autism rehabilitation training strategy is characterized by comprising the following steps of: s1, acquiring multi-modal data of a patient, and performing correlation analysis of symptoms and causes through an improved deep learning model to obtain a symptom list and cause assumptions, wherein the multi-modal data comprises physiological data, behavior data and environmental data; S2, constructing a dynamic knowledge graph containing symptoms, causes and strategy mapping relations based on historical data, and carrying out multi-step reasoning and implicit risk prediction on the dynamic knowledge graph by taking a symptom list and cause assumptions as starting points to obtain a candidate strategy set containing explicit association and implicit association; s3, performing strategy coordination and conflict detection on the candidate strategy set by adopting a multi-level optimization fusion algorithm, and sequencing according to progressive behavior logic to obtain a rehabilitation training strategy; and S4, dynamically updating the dynamic knowledge graph based on feedback data to form a closed loop.
2. The method for generating a child autism rehabilitation training strategy according to claim 1, wherein the association analysis incorporates a time-series convolution network and a gating cross-modal attention mechanism, wherein: performing symptom judgment through a multi-branch feature extraction network according to the multi-mode data to generate the symptom list; Extracting local key features of each mode data in a time window before symptom occurrence by using the time sequence convolution network; The gating cross-modal attention mechanism obtains the incentive hypothesis by calculating dynamic association weights among the multi-modal data according to the local key features to identify key incentives causing specific symptoms.
3. The method of claim 2, wherein the multi-branch feature extraction network comprises a time-series convolution branch for processing physiological data, a three-dimensional convolution neural network branch for processing behavioral video data, and a fully-connected network branch for processing environmental data.
4. The method for generating the child autism rehabilitation training strategy according to claim 1, wherein the construction of the dynamic knowledge graph specifically comprises the following steps: Extracting entities, relations and attributes from the historical rehabilitation records and clinical observation texts, and generating a static map containing association relations between symptoms, causes and strategies; generating an embedded representation for nodes and relations in the static map based on a graph attention network, and dynamically adjusting the relation weight among the nodes according to time sequence association in the multi-mode data; When the continuous significant change of the association of the specific entity is monitored, triggering the structural remodeling operation of the knowledge graph, wherein the structural remodeling operation comprises entity node splitting, fusion or semantic label updating so that the graph continuously reflects the real condition evolution of the patient.
5. The method for generating the child autism rehabilitation training strategy according to claim 1, wherein the multi-step reasoning and hidden risk prediction is specifically as follows: The symptom list and the incentive hypothesis are taken as starting nodes, a graph reasoning algorithm is utilized on the dynamic knowledge graph to explore and evaluate a plurality of potential risk evolution paths, and the graph reasoning algorithm can screen out neighborhood information with strong relevance to a target and aggregate information under multiple relations; According to the risk evolution process in the historical data, calculating the risk probability of a potential path, identifying the hidden risk which is not currently shown, and generating a candidate strategy set with both explicit association and implicit association from the map according to the matching of the hidden risk.
6. The method for generating a child autism rehabilitation training strategy according to claim 1, wherein the multi-level optimization fusion algorithm comprises: Performing refined clustering based on the function description, implementation conditions and expected effects of the policies to merge redundant policies, and identifying cooperative combinations and potential conflicts among the policies through association rule mining; mapping the strategy set subjected to merging redundancy to a preset progressive model, generating a training path with progressive difficulty, and sequencing to obtain the rehabilitation training strategy.
7. The method for generating a child autism rehabilitation training strategy according to claim 6, further comprising a dynamic priority and feasibility checking mechanism, wherein the dynamic priority and feasibility checking mechanism is used for dynamically adjusting the starting priority of the strategy and filtering out the strategy which is not feasible under the current condition according to the real-time physiological data and family environment information of the patient.
8. The method for generating the children autism rehabilitation training strategy according to claim 1, wherein the dynamic updating comprises adding or modifying entity relations according to the feedback data, adjusting node characteristics and weights through algorithms such as a graph attention network and the like, and pruning and optimizing low-importance nodes or edges.
9. A child autism rehabilitation training strategy generation system, characterized in that the system is configured to implement a child autism rehabilitation training strategy generation method according to any one of claims 1-8, comprising: The symptom cause analysis module comprises a deep learning unit and a symptom cause analysis module, wherein the deep learning unit is used for acquiring multi-modal data of a patient, and carrying out association analysis on symptoms and causes through an improved deep learning model to obtain a symptom list and cause assumptions, wherein the multi-modal data comprises physiological data, behavior data and environmental data; The strategy matching module comprises a dynamic knowledge graph construction engine, a multi-step reasoning unit and a hidden risk prediction unit, and is used for carrying out multi-step reasoning and hidden risk prediction on the symptom list and the incentive hypothesis to obtain a candidate strategy set containing explicit association and implicit association; The strategy optimization module comprises a cluster analysis engine, an association rule miner and a logic sequencing unit, and is used for carrying out strategy coordination and conflict detection on the candidate strategy set and sequencing according to progressive logic to obtain a rehabilitation training strategy conforming to individual characteristics; the dynamic updating module comprises a knowledge graph updating unit and is used for driving the autonomous evolution and dynamic tuning of the graph structure, the relation weight and the entity node through incremental learning according to the feedback data.
10. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for generating a child autism rehabilitation training strategy according to any one of claims 1-8 when executing the computer program.

Description

Child autism rehabilitation training strategy generation method and system Technical Field The invention relates to the technical field of intelligent rehabilitation, in particular to a method and a system for generating a child autism rehabilitation training strategy. Background Autism Spectrum Disorder (ASD) is a neuro-developmental disorder whose core symptoms manifest as social interaction disorders, notch repetitive behaviors, and stenosis of interest. The incidence rate of autism in the global scope is continuously rising, the demand for rehabilitation resources is rapidly increased, and early scientific intervention is a key for improvement, so that the development of accurate and effective autism scheme generation technology is urgently needed. Most of the existing autism scheme generating systems collect multi-source data of patients through sensors, and match symptoms with solutions according to a static rule base comprising mapping relations between the symptoms and corresponding strategies. However, when matching, only the data fluctuation at the moment of symptom burst is concerned, the local key features contained in the data sequence before symptom occurrence are ignored, deep causes of symptoms cannot be deeply caused, the accuracy and effectiveness of a generated scheme are insufficient only by carrying out mechanical matching according to surface symptoms, a static rule base only responds to the symptoms which are displayed by a patient, whether the patient has the unrepresented hidden symptoms or not cannot be judged, the execution effect of the generated strategy is not expected easily, in addition, in the specific implementation process of the scheme, the scheme list generated in the prior art is only subjected to simple duplicate elimination and fusion, the synergic or behavioral logic conflict among schemes is not considered, and the discomfort of the patient is aggravated, so that the rehabilitation progress is influenced. Therefore, how to research and design a method and a system for generating a rehabilitation training strategy for children autism, which can overcome the defects, is an urgent problem to be solved at present. Disclosure of Invention In order to solve the defects in the prior art, the invention aims to provide a method and a system for generating a child autism rehabilitation training strategy, which are used for carrying out intelligent analysis and accurate assessment on symptoms and causes by acquiring multi-mode data, carrying out strategy generation by dynamic knowledge-graph and implicit risk prediction, carrying out scientific strategy optimization by a multi-level optimization fusion algorithm, and finally carrying out continuous self-adaptive optimization by a closed loop feedback mechanism, so that the effects of overcoming the defects of strong subjectivity, strategy templating, adjustment lag and the like of the traditional rehabilitation training are achieved, truly personalized, self-adaptive, scientific and effective remote rehabilitation training support is realized, and a full-flow intelligent solution from objective diagnosis, preventive intervention to dynamic optimization is provided for autism children. The technical aim of the invention is realized by the following technical scheme: In a first aspect, a method for generating a rehabilitation training strategy for infantile autism is provided, which includes the following steps: s1, acquiring multi-modal data of a patient, and performing correlation analysis of symptoms and causes through an improved deep learning model to obtain a symptom list and cause assumptions, wherein the multi-modal data comprises physiological data, behavior data and environmental data; S2, constructing a dynamic knowledge graph containing symptoms, causes and strategy mapping relations based on historical data, and carrying out multi-step reasoning and implicit risk prediction on the dynamic knowledge graph by taking a symptom list and cause assumptions as starting points to obtain a candidate strategy set containing explicit association and implicit association; S3, performing strategy coordination and conflict detection on the candidate strategy set by adopting a multi-level optimization fusion algorithm, and sequencing according to progressive behavior logic to obtain a rehabilitation training strategy conforming to individual characteristics; And S4, receiving feedback data after executing the rehabilitation training strategy, and dynamically updating the dynamic knowledge graph based on the feedback data to form a closed loop. Further, the association analysis merges a time sequence convolution network and a gating cross-modal attention mechanism, wherein: performing symptom judgment through a multi-branch feature extraction network according to the multi-mode data to generate the symptom list; Extracting local key features of each mode data in a time window before symptom occurrence by using the time sequence con