Digital family-medical collaborative rehabilitation method, device, equipment and medium

Abstract

The invention relates to a digital family-medical collaborative rehabilitation method, device and equipment and a medium. The method comprises the following steps: firstly, acquiring an original rehabilitation behavior data stream of a home-end sensor network, then performing data cleaning on the data stream, repairing missing values through an interpolation method and performing standardization processing to generate a rehabilitation behavior feature set; then, extracting a patient activity mode and a physiological index change trend from the feature set to obtain a dynamic mode cluster of key rehabilitation indexes; and finally, establishing a rehabilitation guidance knowledge graph based on the dynamic mode cluster, and if the cognitive competence evaluation result of the patient is lower than a preset standard, adjusting the guidance content complexity to generate an adaptive rehabilitation guidance scheme. According to the method, the normalization and accuracy of family rehabilitation are improved, the cooperation efficiency of the family end and the medical end is enhanced, rehabilitation guidance fitting the individual state is provided for a patient, meanwhile, the utilization of medical resources is optimized, and the problems of insufficient data integration, poor scheme adaptability and the like in traditional family rehabilitation are solved.

Description

Technical Field

[0001] This invention belongs to the field of medical information technology, and in particular relates to a digital home-medical collaborative rehabilitation method, device, equipment and medium. Background Technology

[0002] With the aging population and increasing demand for chronic disease management, the reliance on digital technology in home-based rehabilitation is gradually increasing. Traditional home-based rehabilitation models suffer from fragmented data collection, insufficient ability to dynamically adjust rehabilitation plans, and low efficiency in collaboration between the home and medical ends. Existing technologies have limited means of integrating and analyzing home-based rehabilitation behavioral data, lacking real-time responses to trends in patients' physiological indicators and differences in cognitive abilities, making it difficult to accurately match rehabilitation guidance to individual needs. Summary of the Invention

[0003] Therefore, it is necessary to address the aforementioned technical issues by providing a digital home-medical collaborative rehabilitation method, device, equipment, and medium that can improve the standardization and accuracy of home rehabilitation and provide patients with rehabilitation guidance tailored to their individual conditions.

[0004] Firstly, this application provides a digital home-medical collaborative rehabilitation method, including:

[0005] Acquire raw rehabilitation behavior data streams from the home-based sensor network.

[0006] Data cleaning is performed on the rehabilitation behavior data stream, missing values ​​are repaired by interpolation and standardized to generate a rehabilitation behavior feature set.

[0007] By extracting patient activity patterns and physiological indicator trends from rehabilitation behavior characteristics, dynamic pattern clusters of key rehabilitation indicators are obtained.

[0008] A rehabilitation guidance knowledge graph is built based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan.

[0009] In one embodiment, data cleaning is performed on the rehabilitation behavior data stream, interpolation is used to repair missing values, and standardization is performed to generate a rehabilitation behavior feature set, including:

[0010] Receive rehabilitation behavior data stream; the rehabilitation behavior data stream includes time-series sensor data and action tags.

[0011] Sliding window segmentation is performed on time-series sensor data to obtain segmented data.

[0012] Outlier correction is performed on segmented data; outlier correction uses an interquartile range threshold for judgment, and segmented data exceeding the threshold is marked as missing data blocks.

[0013] Missing data blocks are repaired using cubic spline interpolation to generate a continuous data stream after interpolation.

[0014] Frequency domain features are extracted based on action tags; frequency domain features include energy spectral density and dominant frequency components.

[0015] Temporal feature extraction is performed on the interpolated continuous data stream, and the frequency domain features and temporal features are merged to obtain a multidimensional feature vector; the temporal features include mean, variance and zero-crossing rate.

[0016] The multidimensional feature vectors are standardized to generate a rehabilitation behavior feature set; the rehabilitation behavior feature set contains the standardized multidimensional feature vectors and the corresponding rehabilitation stage labels.

[0017] In one embodiment, patient activity patterns and physiological indicator trends are extracted from the rehabilitation behavior feature set to obtain a dynamic pattern cluster of key rehabilitation indicators, including:

[0018] Clustering algorithms are used to process activity sequences in the rehabilitation behavior feature set to obtain activity pattern segmentation data; the activity pattern segmentation data identifies the time intervals of different rehabilitation stages.

[0019] Based on the time interval, the physiological indicator time series data of the rehabilitation behavior feature set are extracted to obtain segmented physiological indicator data.

[0020] Analyze the trend fluctuation characteristics of segmented physiological index data; the trend fluctuation characteristics include slope changes and variance thresholds.

[0021] If the slope change exceeds the preset range and the variance threshold is lower than the preset threshold, then abnormal nodes of key rehabilitation indicators are identified.

[0022] By integrating segmented data from abnormal nodes and activity patterns, an indicator correlation network is constructed.

[0023] A graph embedding algorithm is used to process the index association network to obtain dynamic pattern clusters; the dynamic pattern clusters represent the dynamic change patterns of key rehabilitation indicators.

[0024] In one embodiment, segmented data from abnormal nodes and activity patterns are fused to construct an indicator correlation network, including:

[0025] Get the attribute set of the abnormal node; the attribute set includes node type and abnormal threshold.

[0026] Extract the time interval features of the activity pattern segment data; the time interval features consist of the start time and duration of the rehabilitation phase.

[0027] A node-stage mapping table is generated based on the attribute set and time interval features; the node-stage mapping table contains the correspondence between abnormal nodes and recovery stages.

[0028] A graph database is used to store node-stage mapping tables to construct the topology of the indicator association network.

[0029] A community detection algorithm is executed based on the topology to identify clusters of anomalous nodes with strong correlations.

[0030] A dynamic weight matrix is ​​generated based on the cluster of abnormal nodes to update the edge connection strength of the index-related network.

[0031] If the edge connection strength exceeds a preset threshold, a re-alignment operation of the segmented data is triggered to obtain aligned segmented data.

[0032] The optimized index association network model is output using a dynamic weight matrix and aligned segmented data.

[0033] In one embodiment, the dynamic weight matrix is ​​represented by the following formula:

[0034]

[0035] W = [ω ij ] n×n

[0036] Where W represents the dynamic weight matrix, ω ij Represents node v i With v j The edge connection strength, S(v) i ,v j ) represents attribute similarity, d c (v i ,v j V represents the topological distance between nodes within the same cluster. ReLU represents the corrected linear unit, which activates weights only when the attribute similarity between nodes across the cluster exceeds a threshold θ. α, β, and γ represent weight coefficients, and θ represents the cross-cluster activation threshold. V = {v1, v2, ..., v...} n} represents the set of abnormal nodes, C = {C1, C2, ..., C} k} indicates that the community detection algorithm divides the nodes into k clusters.

[0037] In one embodiment, the method establishes a rehabilitation guidance knowledge graph based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than a preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan, including:

[0038] Personalized rehabilitation need feature vectors are extracted from dynamic pattern cluster data; the need feature vectors include parameters of motor function recovery rate and cognitive training responsiveness.

[0039] By associating demand feature vectors with rehabilitation action libraries and cognitive training material libraries, and combining expert experience with clinical case data, a rehabilitation guidance knowledge graph is constructed.

[0040] Obtain patient cognitive ability assessment data; the assessment data includes attention duration and instruction comprehension accuracy.

[0041] If the attention span or the accuracy of instruction comprehension is lower than the preset threshold, the association weights of the training materials in the rehabilitation guidance knowledge graph are adjusted to generate an adaptive rehabilitation guidance plan. The rehabilitation guidance plan includes graded training tasks, multimodal teaching content, and intelligent progress control strategies.

[0042] In one embodiment, adjusting the association weights of training materials in the rehabilitation guidance knowledge graph includes:

[0043] When the cognitive ability assessment result is lower than the preset threshold, the association weights of the training materials in the rehabilitation guidance knowledge graph are dynamically adjusted using the following weight adjustment formula:

[0044]

[0045] Where, ρ ij Indicates patient p i With training material m j The correlation strength, ExpDecay(t) represents the forgetting factor, t represents the evaluation time, and I(m) represents the value of the association. j ,τ d ) represents the difficulty suitability index, τ d Attn(e) represents the center of the current cognitive ability cluster. f ,η i ) represents the expert experience attention mechanism, e f Represents the expert experience vector, η i This represents the feature vector of the patient's personalized rehabilitation needs, where ∈ and ε represent constraint parameters, and μ represents the adjustment intensity coefficient.

[0046] Secondly, this application also provides a digital home-medical collaborative rehabilitation device, the device comprising:

[0047] The data acquisition and processing module is used to acquire the raw rehabilitation behavior data stream from the home-based sensor network; it is also used to perform data cleaning on the rehabilitation behavior data stream, use interpolation to repair missing values ​​and perform standardization processing to generate a rehabilitation behavior feature set.

[0048] The dynamic pattern analysis module is used to extract patient activity patterns and physiological indicator change trends from the rehabilitation behavior feature set to obtain dynamic pattern clusters of key rehabilitation indicators.

[0049] The rehabilitation guidance generation module is used to build a rehabilitation guidance knowledge graph based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0052] The aforementioned digital home-medical collaborative rehabilitation method, device, computer equipment, and storage medium first acquire raw rehabilitation behavior data streams from a home-based sensor network. Then, the data streams are cleaned, missing values ​​are repaired using interpolation, and the data is standardized to generate a rehabilitation behavior feature set. Next, patient activity patterns and physiological indicator trends are extracted from the feature set to obtain dynamic pattern clusters of key rehabilitation indicators. Finally, a rehabilitation guidance knowledge graph is established based on these dynamic pattern clusters. If the patient's cognitive ability assessment results are lower than a preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan. This improves the standardization and accuracy of home rehabilitation, enhances the collaborative efficiency between the home and medical ends, provides patients with rehabilitation guidance tailored to their individual conditions, optimizes the utilization of medical resources, and helps solve problems such as insufficient data integration and poor plan adaptability in traditional home rehabilitation. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart illustrating a digital home-medical collaborative rehabilitation method provided in an embodiment of the present invention;

[0055] Figure 2 This is a structural block diagram of a digital home-medical collaborative rehabilitation device provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, such as Figure 1 As shown, this application provides a digital home-medical collaborative rehabilitation method, which may include the following steps:

[0059] Step S101: Obtain the raw rehabilitation behavior data stream from the home-side sensor network.

[0060] Specifically, a sensor network (such as wearable devices, motion capture devices, and physiological monitors) deployed in the home environment is used to collect raw data generated by patients in real time during the rehabilitation process. Data types include, but are not limited to, time-series data of limb movement trajectories (such as joint angles and displacement speeds), physiological index data (such as heart rate and electromyography signals), and operation logs of interactions with rehabilitation equipment. The raw data stream is recorded in time-series format, accompanied by action labels (such as "shoulder abduction training" and "finger grasping exercises") to identify specific rehabilitation behaviors.

[0061] Step S102: Perform data cleaning on the rehabilitation behavior data stream, use interpolation to repair missing values ​​and perform standardization processing to generate a rehabilitation behavior feature set.

[0062] First, the raw data stream is preprocessed by using a sliding window technique to segment the continuous data into time-contextual segments for easier analysis. Outliers in the segments are identified and marked as missing values ​​using the interquartile range (IQR) method, and then cubic spline interpolation is used to fill the missing regions, generating a continuous and complete data stream. Next, multi-dimensional features are extracted from the repaired data stream: in time-domain analysis, statistics such as mean, variance, and zero-crossing rate are calculated; in frequency-domain analysis, parameters such as energy spectral density and dominant frequency components are extracted, and these are combined to form a multi-dimensional vector containing both time-series and frequency-domain features. Finally, standardization (such as Z-score normalization) is used to eliminate dimensional differences, generating a structured set of rehabilitation behavior features. Rehabilitation stage labels are also associated to distinguish data features from different training periods.

[0063] Step S103: Extract patient activity patterns and physiological indicator change trends from the rehabilitation behavior feature set to obtain a dynamic pattern cluster of key rehabilitation indicators.

[0064] Clustering algorithms (such as DBSCAN and Gaussian mixture models) are used to analyze activity sequences in the rehabilitation behavior feature set. Based on feature similarity, the data is divided into different activity pattern segments, each corresponding to a specific rehabilitation stage's time interval (e.g., "early joint range of motion training stage" and "mid-term muscle strength enhancement stage"). Based on the segmentation results, corresponding physiological indicator time-series data are extracted, and abnormal states are identified by calculating trend fluctuation characteristics (e.g., slope changes reflect the rate of increase / decrease of indicators, and variance thresholds measure data stability). When the slope change of a physiological indicator exceeds a preset range and the variance is lower than the stability threshold, it is identified as an abnormal node of a key rehabilitation indicator. Furthermore, the attributes of abnormal nodes (e.g., indicator type, abnormal threshold) are correlated with the time interval characteristics (start time, duration) of the activity pattern segments to construct an indicator association network. Community detection algorithms (such as the Louvain algorithm) are used to identify strongly correlated abnormal node clusters, ultimately generating dynamic pattern clusters representing the dynamic changes of indicators.

[0065] Step S104: Establish a rehabilitation guidance knowledge graph based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than the preset standard, adjust the complexity of the guidance content to generate an adaptive rehabilitation guidance plan.

[0066] Specifically, centered on dynamic pattern clusters, the system associates patient-specific rehabilitation needs feature vectors (including motor function recovery rate and cognitive training responsiveness) with rehabilitation action libraries and cognitive training material libraries, incorporating expert experience and clinical case data to construct a rehabilitation guidance knowledge graph containing nodes (patients, training content, expert experience) and edges (relationships and weights). Patient cognitive ability data is acquired through standardized cognitive assessment tools (such as attention span tests and instruction comprehension accuracy tests). If the assessment results are below a preset threshold, a knowledge graph weight adjustment mechanism is triggered: reducing the association weight of high-cognitive-load training materials, increasing the priority of multimodal auxiliary content (such as animation demonstrations and voice guidance), and generating an adaptive solution including graded training tasks (basic-intermediate-advanced), multimodal teaching resources, and intelligent progress control strategies. This solution achieves intelligent adjustment and precise delivery of rehabilitation guidance content by dynamically matching the patient's real-time status.

[0067] The aforementioned digital home-medical collaborative rehabilitation method first acquires raw rehabilitation behavior data streams from a home-based sensor network. Then, it performs data cleaning, repairs missing values ​​using interpolation, and standardizes the data to generate a rehabilitation behavior feature set. Next, it extracts patient activity patterns and physiological indicator trends from the feature set, obtaining dynamic pattern clusters of key rehabilitation indicators. Finally, it establishes a rehabilitation guidance knowledge graph based on these dynamic pattern clusters. If the patient's cognitive ability assessment results are lower than a preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan. This improves the standardization and accuracy of home-based rehabilitation, enhances the collaborative efficiency between the home and medical ends, provides patients with rehabilitation guidance tailored to their individual conditions, optimizes the utilization of medical resources, and helps solve problems such as insufficient data integration and poor plan adaptability in traditional home-based rehabilitation.

[0068] In one embodiment, performing data cleaning on the rehabilitation behavior data stream, using interpolation to repair missing values ​​and performing standardization to generate a rehabilitation behavior feature set may include the following steps:

[0069] Step S201: Receive rehabilitation behavior data stream; the rehabilitation behavior data stream includes time-series sensor data and action tags.

[0070] Step S202: Perform sliding window segmentation on the time series sensor data to obtain segmented data.

[0071] Step S203: Perform outlier correction on the segmented data; outlier correction uses the interquartile range threshold for judgment, and segmented data exceeding the threshold is marked as missing data blocks.

[0072] Step S204: Use cubic spline interpolation to repair missing data blocks and generate interpolated continuous data stream.

[0073] Step S205: Extract frequency domain features based on action tags; frequency domain features include energy spectral density and dominant frequency components.

[0074] Step S206: Perform time-domain feature extraction on the interpolated continuous data stream, merge the frequency-domain features and time-domain features to obtain a multi-dimensional feature vector; the time-domain features include mean, variance and zero-crossing rate.

[0075] Step S207: Perform standardization processing on the multidimensional feature vector to generate a rehabilitation behavior feature set; the rehabilitation behavior feature set includes the standardized multidimensional feature vector and the corresponding rehabilitation stage label.

[0076] First, a rehabilitation behavior data stream containing time-series sensor data and action tags is received. Then, sliding window segmentation is performed on the time-series sensor data to obtain segmented data. Next, outlier correction is performed on the segmented data. Using an interquartile range threshold, segments exceeding the threshold are marked as missing data blocks, and cubic spline interpolation is used to repair these missing data blocks, generating an interpolated continuous data stream. Then, frequency domain features such as energy spectral density and dominant frequency components are extracted based on the action tags. Simultaneously, time domain features such as mean, variance, and zero-crossing rate are extracted from the interpolated continuous data stream, and these are combined to form a multidimensional feature vector. Finally, the multidimensional feature vector is standardized to generate a rehabilitation behavior feature set containing standardized multidimensional feature vectors and corresponding rehabilitation stage labels.

[0077] This embodiment improves the quality and usability of home-based rehabilitation data through a systematic cleaning, repair, and feature extraction mechanism. Sliding window segmentation and outlier correction ensure data continuity and reliability; time-frequency domain feature fusion comprehensively characterizes rehabilitation behavior features; and standardization eliminates dimensional differences, providing a structured data foundation for subsequent rehabilitation stage division and dynamic pattern analysis. It effectively solves common problems in home rehabilitation data collection, such as fragmentation, excessive noise, and incomplete features.

[0078] In one embodiment, extracting patient activity patterns and physiological indicator change trends from a set of rehabilitation behavior features to obtain a dynamic pattern cluster of key rehabilitation indicators may include the following steps:

[0079] Step S301: Use a clustering algorithm to process the activity sequences in the rehabilitation behavior feature set to obtain activity pattern segmentation data; the activity pattern segmentation data identifies the time intervals of different rehabilitation stages.

[0080] Step S302: Extract time-series data of physiological indicators from the set of rehabilitation behavior characteristics according to the time interval to obtain segmented physiological indicator data.

[0081] Step S303: Analyze the trend fluctuation characteristics of the segmented physiological index data; the trend fluctuation characteristics include slope changes and variance thresholds.

[0082] Step S304: If the slope change exceeds the preset range and the variance threshold is lower than the preset threshold, then the abnormal node of the key rehabilitation indicator is determined.

[0083] Step S305: Integrate the segmented data of abnormal nodes and activity patterns to construct an indicator correlation network.

[0084] Step S306: The graph embedding algorithm is used to process the index association network to obtain dynamic pattern clusters; the dynamic pattern clusters represent the dynamic change patterns of key rehabilitation indicators.

[0085] Furthermore, after constructing the indicator association network, the topological information and node attributes in the network structure need to be mapped to a low-dimensional vector space. Using graph embedding algorithms (such as DeepWalk and Node2Vec), abnormal nodes and their relationships are transformed into continuous vector representations, preserving the structural similarity and semantic associations between nodes. In the vector space, nodes with similar change patterns are clustered into the same region, forming dynamic pattern clusters. Each pattern cluster corresponds to the coordinated change pattern of key indicators at a specific rehabilitation stage, such as the "heart rate-movement frequency association pattern in the early joint range of motion training stage" or the "attention-task completion decay pattern under cognitive fatigue." By analyzing the distribution and evolution of pattern clusters, the stage-specific characteristics of the rehabilitation process can be quantified, providing data-driven pattern references for the development of personalized rehabilitation programs.

[0086] Specifically, a clustering algorithm is used to process the activity sequences in the rehabilitation behavior feature set to obtain segmented activity pattern data that identifies different rehabilitation stages and time intervals. Then, physiological indicator time-series data are extracted from the rehabilitation behavior feature set according to the time intervals to form segmented physiological indicator data. Next, the trend fluctuation characteristics of the segmented physiological indicator data are analyzed, including slope changes and variance thresholds. If the slope change exceeds a preset range and the variance threshold is lower than a preset threshold, it is identified as an abnormal node of the key rehabilitation indicator. Subsequently, the abnormal nodes are fused with the activity pattern segmented data to construct an indicator association network. Finally, a graph embedding algorithm is used to process the indicator association network to obtain dynamic pattern clusters representing the dynamic changes of key rehabilitation indicators.

[0087] This embodiment realizes the correlation analysis between rehabilitation stages and abnormal physiological indicators. It automatically divides rehabilitation stages using clustering algorithms, reducing the cost of manual annotation; accurately identifies abnormal nodes in physiological indicators through trend fluctuation feature analysis, improving the sensitivity of rehabilitation monitoring; and the dynamic pattern clusters constructed based on graph embedding algorithms can effectively capture the spatiotemporal correlation patterns of key indicators, providing a quantitative basis for the dynamic assessment and intervention of the rehabilitation process, and helping to realize the transformation from "experience-based rehabilitation" to "data-driven rehabilitation".

[0088] In one embodiment, fusing segmented data from abnormal nodes and activity patterns to construct an indicator correlation network may include the following steps:

[0089] Step S401: Obtain the attribute set of the abnormal node; the attribute set includes node type and abnormal threshold.

[0090] Step S402: Extract the time interval features of the activity pattern segment data; the time interval features consist of the start time and duration of the rehabilitation phase.

[0091] Step S403: Generate a node-stage mapping table based on the attribute set and time interval features; the node-stage mapping table contains the correspondence between abnormal nodes and recovery stages.

[0092] Step S404: Use a graph database to store node-stage mapping tables and construct the topology of the indicator association network.

[0093] Step S405: Execute the community detection algorithm based on the topology to identify clusters of abnormal nodes with strong correlations.

[0094] Step S406: Generate a dynamic weight matrix based on the abnormal node cluster to update the edge connection strength of the index-related network.

[0095] Furthermore, a dynamic weight matrix is ​​generated based on the abnormal node clusters to update the edge connection strength of the indicator association network. In this step, after identifying the abnormal node clusters, the degree of association between nodes within and across clusters needs to be quantified. A dynamic weight matrix is ​​constructed by calculating parameters such as attribute similarity (e.g., matching degree of node type and abnormal threshold) and topological distance (the length of the connection path in the network structure) between nodes, combined with pre-set weight coefficients. Each element in the matrix corresponds to the edge connection strength between nodes in the indicator association network, and its value reflects the strength of the association between nodes. For nodes within the same cluster, higher weights are assigned based on their temporal relevance and functional similarity during the rehabilitation process; while for nodes across clusters, the weights are dynamically adjusted according to their attribute matching degree and clinical significance, thereby achieving a quantitative update of the edge connection strength in the indicator association network.

[0096] In step S407, if the edge connection strength exceeds a preset threshold, a re-alignment operation of the segmented data is triggered to obtain the aligned segmented data.

[0097] Step S408: Output the optimized index association network model using the dynamic weight matrix and aligned segmented data.

[0098] First, the attribute set of abnormal nodes is obtained, including node type and abnormality threshold. Simultaneously, the time interval features of the activity pattern segmented data are extracted, namely the start time and duration of the recovery phase. Next, a node-phase mapping table is generated based on the attribute set and time interval features to clarify the correspondence between abnormal nodes and recovery phases. Then, a graph database is used to store the mapping table, constructing the topology of the indicator association network. Based on the topology, a community detection algorithm is executed to identify clusters of strongly correlated abnormal nodes, and a dynamic weight matrix is ​​generated to update the edge connection strength of the network based on these clusters. If the edge connection strength exceeds a preset threshold, a re-alignment operation of the segmented data is triggered, resulting in aligned segmented data. Finally, the optimized indicator association network model is output through the dynamic weight matrix and the aligned data.

[0099] This embodiment utilizes a node-stage mapping table to construct a topology that clearly depicts the correspondence between abnormal indicators and the rehabilitation process. The combination of the community detection algorithm and the dynamic weight matrix can effectively identify strong correlation clusters among indicators, enhancing the network model's ability to capture the inherent patterns in rehabilitation data. The segmented data realignment mechanism, through weight-driven dynamic adjustment, ensures that the network model is consistent with the temporal characteristics of the rehabilitation stage, providing a reliable model foundation for accurately analyzing the dynamic evolution of rehabilitation indicators and enhancing the systematicness and accuracy of rehabilitation monitoring.

[0100] In one embodiment, the dynamic weight matrix can be represented by the following formula:

[0101]

[0102] W = [ω ij ] n×n

[0103] Where W represents the dynamic weight matrix, ω ij Represents node v i With v j The edge connection strength, S(v) i ,v j ) represents attribute similarity, d c (v i ,v j V represents the topological distance between nodes within the same cluster. ReLU represents the corrected linear unit, which activates weights only when the attribute similarity between nodes across the cluster exceeds a threshold θ. α, β, and γ represent weight coefficients, and θ represents the cross-cluster activation threshold. V = {v1, v2, ..., v...} n} represents the set of abnormal nodes, C = {C1, C2, ..., C} k} indicates that the community detection algorithm divides the nodes into k clusters.

[0104] Preferably, x i and x j S(v) represents a vector normalized to the time interval features, etc. i ,v j The value range is [0,1], and the larger the value, the more similar the node attributes are.

[0105] This embodiment achieves dynamic calculation of the edge connection strength of the indicator association network by quantifying the similarity of node attributes and the relationship of topological structure. For nodes within the same cluster, local association modeling is strengthened by combining attribute similarity and topological distance to ensure tight connections between strongly correlated indicators. For nodes across clusters, invalid associations are filtered out by activation thresholds, retaining the truly physiologically or pathologically significant cross-stage indicator linkages. The introduction of a dynamic weight matrix enables the indicator association network to dynamically adjust with the rehabilitation stage, accurately reflecting the association patterns of abnormal nodes at different times. This provides a computable model basis for the quantitative analysis and abnormal early warning of the rehabilitation process, improving the scientific rigor and flexibility of data-driven rehabilitation monitoring.

[0106] In one embodiment, the method establishes a rehabilitation guidance knowledge graph based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than a preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan, which may include the following steps:

[0107] Step S501: Extract personalized rehabilitation need feature vectors based on dynamic pattern cluster data; the need feature vectors include parameters of motor function recovery rate and cognitive training response.

[0108] Step S502: Associate the demand feature vector with the rehabilitation action library and the cognitive training material library, and construct a rehabilitation guidance knowledge graph by combining expert experience and clinical case data.

[0109] Furthermore, using the personalized rehabilitation needs feature vector extracted from dynamic pattern cluster data as the core, a basic relational network is formed by establishing mapping relationships between parameters such as motor function recovery rate and cognitive training responsiveness in the vector and the difficulty of movements and muscle group training types in the rehabilitation movement library, as well as the task complexity and training objectives in the cognitive training material library. Simultaneously, rehabilitation program development guidelines from the expert experience library and rehabilitation path data for similar patients from the clinical case library are introduced to weight and enhance the relationships within the basic network. Rehabilitation movements, cognitive materials, expert experience, and clinical cases are incorporated into the network as nodes, and weighted edges are formed between nodes based on the strength of data association. Ultimately, a structured and computable rehabilitation guidance knowledge graph is constructed.

[0110] Step S503: Obtain the patient's cognitive ability assessment results data; the assessment results data include attention duration and instruction comprehension accuracy.

[0111] Step S504: If the attention span or the accuracy of instruction comprehension is lower than a preset threshold, adjust the association weights of the training materials in the rehabilitation guidance knowledge graph to generate an adaptive rehabilitation guidance plan; the rehabilitation guidance plan includes graded training tasks, multimodal teaching content and intelligent progress control strategies.

[0112] First, personalized rehabilitation need feature vectors are extracted from dynamic pattern cluster data, encompassing parameters such as motor function recovery rate and cognitive training responsiveness. Then, the extracted feature vectors are correlated with a rehabilitation movement library and a cognitive training material library, and expert experience and clinical case data are integrated to construct a rehabilitation guidance knowledge graph. Next, patient cognitive ability assessment results are obtained, including attention span and instruction comprehension accuracy. Finally, if attention span or instruction comprehension accuracy falls below a preset threshold, the association weights of training materials in the rehabilitation guidance knowledge graph are adjusted to generate an adaptive rehabilitation guidance plan that includes tiered training tasks, multimodal teaching content, and intelligent progress control strategies.

[0113] This embodiment achieves personalized and dynamic rehabilitation guidance. Based on feature vectors extracted from dynamic pattern clusters, it accurately characterizes the individual rehabilitation needs of patients; the construction of a knowledge graph integrates multi-source data to form a structured rehabilitation knowledge system; and the weights of training materials are adjusted in conjunction with cognitive ability assessments to ensure that the rehabilitation plan matches the patient's current state. Through tiered tasks, multimodal teaching, and intelligent progress control, it effectively solves the problems of strong universality but insufficient individual adaptation in traditional rehabilitation guidance, improving the pertinence and effectiveness of rehabilitation training, while also promoting the collaborative efficiency between home rehabilitation and the medical end.

[0114] In one embodiment, adjusting the association weights of training materials in the rehabilitation guidance knowledge graph may include the following steps:

[0115] When the cognitive ability assessment result is lower than the preset threshold, the association weights of the training materials in the rehabilitation guidance knowledge graph are dynamically adjusted using the following weight adjustment formula:

[0116]

[0117] Where, ρ ij Indicates patient p i With training material m j The correlation strength, ExpDecay(t) represents the forgetting factor, t represents the evaluation time, and I(m) represents the value of the association. j ,τ d ) represents the difficulty suitability index, τ d Attn(e) represents the center of the current cognitive ability cluster. f ,η i ) represents the expert experience attention mechanism, e f Represents the expert experience vector, η i This represents the feature vector of the patient's personalized rehabilitation needs, where ∈ and ε represent constraint parameters, and μ represents the adjustment intensity coefficient.

[0118] Preferably, Where Δt represents the time interval since the last evaluation, T represents the reference period, and Accuracyi Indicates patient p i The accuracy of instruction comprehension. base δ represents the baseline accuracy, and δ represents the adjustment coefficient.

[0119] Preferably, Among them, Attention i This indicates the duration of real-time attention of patient i. base The baseline value representing the duration of attention, d j representing training material m j Difficulty level, d max τ represents the highest difficulty level among all training materials. d Let ξ represent the average cognitive ability parameter of the current cognitive ability cluster center, i.e., the same type of patients (divided by clustering), and let r represent the Gaussian distribution function. j This indicates that the patient's response to the training material m j The cognitive ability requirements of the individual.

[0120] Preferably, Attn(e f ,η i ) = Softmax(η i ·U a ·e f )·e f , among which, U a Let e ​​represent the attention weight matrix. f Represents the expert experience vector, η i This represents the feature vector of a patient's personalized rehabilitation needs.

[0121] This weighting adjustment formula achieves dynamic optimization of the rehabilitation guidance knowledge graph through the fusion of multi-dimensional parameters. The forgetting factor, combined with assessment time, quantifies the decline trend of cognitive ability, ensuring that the weights align with changes in the patient's condition. The difficulty matching index selects suitable training materials based on the centers of cognitive ability clusters, avoiding a disconnect between training difficulty and the patient's ability. Expert experience and attention mechanisms incorporate clinical knowledge, enhancing the professionalism of the program. The synergistic effect of these three elements effectively solves the problems of fixed content and lack of dynamic adjustment in traditional rehabilitation guidance, enabling rehabilitation programs to adapt to the patient's cognitive abilities in real time, improving the effectiveness of rehabilitation training and patient acceptance.

[0122] In one embodiment, such as Figure 2 As shown, this application also provides a digital home-medical collaborative rehabilitation device, which may include:

[0123] The data acquisition and processing module 601 is used to acquire the raw rehabilitation behavior data stream from the home sensor network; it is also used to perform data cleaning on the rehabilitation behavior data stream, use interpolation to repair missing values ​​and perform standardization processing to generate a rehabilitation behavior feature set.

[0124] The dynamic pattern analysis module 602 is used to extract patient activity patterns and physiological indicator change trends from the rehabilitation behavior feature set to obtain dynamic pattern clusters of key rehabilitation indicators.

[0125] The rehabilitation guidance generation module 603 is used to build a rehabilitation guidance knowledge graph based on dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan.

[0126] The aforementioned digital home-medical collaborative rehabilitation device includes a data acquisition and processing module, a dynamic pattern analysis module, and a rehabilitation guidance generation module. The data acquisition and processing module acquires raw rehabilitation behavior data streams from a home-based sensor network, performs data cleaning, repairs missing values ​​using interpolation, and standardizes the data to generate a rehabilitation behavior feature set. The dynamic pattern analysis module extracts patient activity patterns and physiological indicator trends from the rehabilitation behavior feature set to obtain dynamic pattern clusters of key rehabilitation indicators. The rehabilitation guidance generation module establishes a rehabilitation guidance knowledge graph based on these dynamic pattern clusters. If the patient's cognitive ability assessment result is lower than a preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan. This improves the standardization and accuracy of home rehabilitation, enhances the collaborative efficiency between the home and medical ends, provides patients with rehabilitation guidance tailored to their individual conditions, optimizes the utilization of medical resources, and helps solve problems such as insufficient data integration and poor plan adaptability in traditional home rehabilitation.

[0127] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0128] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the digital home-medical collaborative rehabilitation method, apparatus, device and medium as described above.

[0129] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0130] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0131] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A digital home-medical collaborative rehabilitation method, characterized in that, The method includes: Acquire raw rehabilitation behavior data streams from the home-based sensor network; Data cleaning is performed on the rehabilitation behavior data stream, missing values ​​are repaired by interpolation and standardized to generate a rehabilitation behavior feature set; The patient's activity patterns and physiological indicator change trends are extracted from the set of rehabilitation behavior characteristics to obtain a dynamic pattern cluster of key rehabilitation indicators; A rehabilitation guidance knowledge graph is established based on the dynamic pattern cluster. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan.

2. The method according to claim 1, characterized in that, The process of cleaning the rehabilitation behavior data stream, using interpolation to repair missing values ​​and standardizing the data to generate a rehabilitation behavior feature set includes: Receive the rehabilitation behavior data stream; the rehabilitation behavior data stream includes time-series sensor data and action tags; Sliding window segmentation is performed on the time-series sensor data to obtain segmented data; Outlier correction is performed on the segmented data; the outlier correction uses an interquartile range threshold to determine outliers, and the segmented data exceeding the threshold is marked as a missing data block. The missing data block is repaired using cubic spline interpolation to generate a continuous data stream after interpolation. Frequency domain features are extracted based on the action tags; the frequency domain features include energy spectral density and dominant frequency components. Temporal feature extraction is performed on the interpolated continuous data stream, and the frequency domain features and temporal features are merged to obtain a multidimensional feature vector; the temporal features include mean, variance, and zero-crossing rate. The multidimensional feature vectors are standardized to generate a rehabilitation behavior feature set; the rehabilitation behavior feature set includes the standardized multidimensional feature vectors and corresponding rehabilitation stage labels.

3. The method according to claim 1, characterized in that, The process of extracting patient activity patterns and physiological indicator change trends from the set of rehabilitation behavior characteristics yields a dynamic pattern cluster of key rehabilitation indicators, including: Clustering algorithms are used to process the activity sequences in the rehabilitation behavior feature set to obtain activity pattern segmentation data; the activity pattern segmentation data identifies the time intervals of different rehabilitation stages; Based on the time interval, the time series data of physiological indicators in the set of rehabilitation behavior characteristics are extracted to obtain segmented physiological indicator data; Analyze the trend fluctuation characteristics of the segmented physiological index data; the trend fluctuation characteristics include slope changes and variance thresholds; If the slope change exceeds a preset range and the variance threshold is lower than a preset threshold, then an abnormal node of the key rehabilitation indicator is determined. By integrating the abnormal nodes with the segmented data of the activity patterns, an indicator correlation network is constructed. The indicator association network is processed using a graph embedding algorithm to obtain a dynamic pattern cluster; the dynamic pattern cluster represents the dynamic change pattern of key rehabilitation indicators.

4. The method according to claim 3, characterized in that, The process of fusing the abnormal nodes with the segmented data of the activity patterns to construct an indicator correlation network includes: Obtain the attribute set of the abnormal node; the attribute set includes node type and abnormal threshold; Extract the time interval features of the activity pattern segment data; the time interval features consist of the start time and duration of the rehabilitation phase. A node-stage mapping table is generated based on the attribute set and the time interval features; the node-stage mapping table contains the correspondence between the abnormal nodes and the recovery stages; A graph database is used to store the node-stage mapping table, and the topology of the indicator association network is constructed. Based on the aforementioned topology, a community detection algorithm is executed to identify clusters of anomalous nodes with strong correlations. A dynamic weight matrix is ​​generated based on the abnormal node cluster to update the edge connection strength of the index association network; If the edge connection strength exceeds a preset threshold, a re-alignment operation of the segmented data is triggered to obtain aligned segmented data. The optimized index association network model is output using the dynamic weight matrix and the aligned segmented data.

5. The method according to claim 4, characterized in that, The dynamic weight matrix is ​​represented by the following formula: W=[ω ij ] n×n Where W represents the dynamic weight matrix, ω ij Represents node υ i With υ j The edge connection strength, S(υ) i ,υ j ) represents attribute similarity, d c (υ i ,υ j V represents the topological distance between nodes within the same cluster. ReLU represents the corrected linear unit, which activates weights only when the attribute similarity between nodes across the cluster exceeds a threshold θ. α, β, and γ represent weight coefficients, and θ represents the cross-cluster activation threshold. V = {υ1,υ2,...,υ} n } represents the set of abnormal nodes, C = {C1, C2, ..., C} k ) indicates that the community detection algorithm divides the nodes into k clusters.

6. The method according to claim 1, characterized in that, The method establishes a rehabilitation guidance knowledge graph based on the dynamic pattern cluster. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan, including: Personalized rehabilitation need feature vectors are extracted from the dynamic pattern cluster data; the need feature vectors include parameters of motor function recovery rate and cognitive training responsiveness. The aforementioned demand feature vectors are associated with the rehabilitation action library and the cognitive training material library, and a rehabilitation guidance knowledge graph is constructed by combining expert experience and clinical case data. Acquire patient cognitive ability assessment data; the assessment data includes attention duration and instruction comprehension accuracy. If the attention span or the accuracy of instruction comprehension is lower than a preset threshold, the association weights of the training materials in the rehabilitation guidance knowledge graph are adjusted to generate an adaptive rehabilitation guidance plan. The rehabilitation guidance plan includes graded training tasks, multimodal teaching content, and intelligent progress control strategies.

7. The method according to claim 6, characterized in that, The adjustment of the association weights of training materials in the rehabilitation guidance knowledge graph includes: When the cognitive ability assessment result is lower than a preset threshold, the association weights of the training materials in the rehabilitation guidance knowledge graph are dynamically adjusted using the following weight adjustment formula: Where, ρ ij Indicates patient p i With training material m j The correlation strength, ExpDecay(t) represents the forgetting factor, t represents the evaluation time, and I(m) represents the value of the association. j ,τ d ) represents the difficulty suitability index, τ d Attn(e) represents the center of the current cognitive ability cluster. f ,η i ) represents the expert experience attention mechanism, e f Represents the expert experience vector, η i This represents the feature vector of the patient's personalized rehabilitation needs, where ∈ and ε represent constraint parameters, and μ represents the adjustment intensity coefficient.

8. A digital home-medical collaborative rehabilitation device, characterized in that, The device includes: The data acquisition and processing module is used to acquire the raw rehabilitation behavior data stream from the home-based sensor network; it is also used to perform data cleaning on the rehabilitation behavior data stream, use interpolation to repair missing values ​​and perform standardization processing to generate a rehabilitation behavior feature set. The dynamic pattern analysis module is used to extract patient activity patterns and physiological indicator change trends from the rehabilitation behavior feature set to obtain dynamic pattern clusters of key rehabilitation indicators. The rehabilitation guidance generation module is used to establish a rehabilitation guidance knowledge graph based on the dynamic pattern cluster. If the patient's cognitive ability assessment result is lower than the preset standard, the complexity of the guidance content is adjusted to generate an adaptive rehabilitation guidance plan.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images