Joint rehabilitation scheme recommendation system based on knowledge graph

Abstract

The invention discloses a joint rehabilitation scheme recommendation system based on a knowledge graph, and relates to the technical field of data processing, and the system comprises a graph construction module which is used for constructing a joint rehabilitation knowledge graph; the linkage graph construction module is used for acquiring joint associated information of a target user, performing entity and relationship alignment with the joint rehabilitation knowledge graph, and constructing a user joint linkage graph; the video frame analysis module is used for constructing a real-time user joint linkage graph sequence; the interaction analysis module is used for executing multi-scale time sequence interaction analysis on the user joint linkage graph sequence to obtain interaction joint linkage features; and the scheme recommendation module is used for carrying out rehabilitation scheme recommendation on the interactive joint linkage features based on the joint rehabilitation knowledge graph to obtain a target joint rehabilitation recommendation scheme. The technical problems that in the prior art, personalized recommendation cannot be carried out on the joint rehabilitation scheme, and the recommendation reliability is low are solved, and the technical effect of improving the recommendation accuracy of the joint rehabilitation scheme is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a joint rehabilitation scheme recommendation system based on a knowledge graph. BACKGROUND

[0002] With the increasing demand for joint disease treatment, significant progress has been made in the field of joint rehabilitation technology. Traditional joint rehabilitation schemes usually rely on expert experience and fixed treatment plans, lacking the ability of individualization and real-time adjustment. Many rehabilitation methods use regular physical examinations and static assessments, failing to fully consider the dynamic changes of patients during treatment. With the development of artificial intelligence and big data technology, data-driven rehabilitation schemes have gradually been applied, especially in the context of combining individual differences, rehabilitation tasks, joint function and real-time feedback, rehabilitation recommendation systems based on knowledge graphs have gradually become a new direction.

[0003] At present, existing knowledge graphs mostly focus on static knowledge representation, failing to effectively handle the dynamic changes of individual differences in the rehabilitation process, such as pain levels and joint function improvement. And traditional rehabilitation scheme recommendations often lack personalized design, failing to adjust in real time according to real-time data of users, resulting in limited treatment effect.

[0004] There is a technical problem in the prior art that joint rehabilitation schemes cannot be personalized recommended, and the reliability of the recommendation is low. SUMMARY

[0005] The present application provides a joint rehabilitation scheme recommendation system based on a knowledge graph, which is used to solve the technical problem that joint rehabilitation schemes cannot be personalized recommended and the reliability of the recommendation is low in the prior art.

[0006] In view of the above problems, the present application provides a joint rehabilitation scheme recommendation system based on a knowledge graph, which comprises: a graph construction module for constructing a joint rehabilitation knowledge graph, wherein the joint rehabilitation knowledge graph is a multi-level graph structure containing joint position linkage relationships and joint function linkage relationships; a linkage graph construction module for obtaining joint association information of a target user, aligning entities and relationships with the joint rehabilitation knowledge graph, and constructing a user joint linkage graph, wherein the joint association information includes joint angles, angular velocities, gait skeleton sequences, pain scores, physiological loads, and diagnosis and treatment records; a video frame analysis module for obtaining real-time joint gait video frame sequences of the target user, combining the user joint linkage graph to analyze the video frames, and constructing a real-time user joint linkage graph sequence; an interaction analysis module for performing multi-scale time-series interaction analysis on the user joint linkage graph sequence to obtain interaction joint linkage features; and a scheme recommendation module for recommending rehabilitation schemes based on the joint rehabilitation knowledge graph, and obtaining a target joint rehabilitation recommendation scheme.

[0007] In a possible implementation, the graph construction module further comprises: a rehabilitation position linkage layer construction unit for constructing a joint position linkage layer according to joint node types and node position relationships; a rehabilitation function linkage layer construction unit for constructing a joint function linkage layer based on function node linkage relationships between different joint nodes; a rehabilitation target-path layer construction unit for obtaining a plurality of rehabilitation task targets and a plurality of corresponding rehabilitation paths, mapping and associating the rehabilitation targets and paths, and constructing a rehabilitation target-path layer; and a relationship definition unit for defining a plurality of types of entities and corresponding entity attributes, defining relationships of the entities in combination with the joint position linkage layer, the joint function linkage layer, and the rehabilitation target-path layer, and constructing the joint rehabilitation knowledge graph.

[0008] In a possible implementation, the linkage graph construction module further comprises: a standardization processing unit for collecting joint data of the target user through a multi-dimensional Internet of Things sensor, performing data standardization processing, and obtaining the joint association information; an entity diffusion unit for performing entity diffusion on the joint association information from three dimensions of joint entities, function entities, and task entities in combination with the joint rehabilitation knowledge graph, and determining a diffusion entity set; and a relationship alignment unit for performing relationship alignment on the diffusion entity set from position linkage relationships, function linkage relationships, and target-path associations, and constructing the user joint linkage graph.

[0009] In a possible implementation, the video frame analysis module further includes: an image enhancement unit, configured to perform image enhancement on the real-time joint gait video frame sequence to obtain an enhanced real-time joint gait video frame sequence; and a node convolution mapping unit, configured to perform node convolution mapping on each enhanced real-time joint gait video frame in the enhanced real-time joint gait video frame sequence to obtain a real-time user joint linkage graph sequence, with the user joint linkage graph as a framework.

[0010] In a possible implementation, the interaction analysis module further includes: a channel pre-construction unit, configured to pre-construct a time-series analysis multi-scale channel set; a time-series feature analysis unit, configured to perform multi-scale time-series feature analysis on the user joint linkage graph sequence by using the time-series analysis multi-scale channel set to obtain a user joint linkage feature set; and a time-series interaction analysis unit, configured to perform multi-scale time-series interaction analysis on the user joint linkage feature set to obtain the interaction joint linkage feature.

[0011] In a possible implementation, the channel pre-construction unit further includes: a difference degree identification subunit, configured to identify the adjacency difference degree of the user joint linkage graph sequence to obtain a plurality of adjacency difference degrees; an extraction subunit, configured to extract the maximum and minimum values of the plurality of adjacency difference degrees to obtain a maximum adjacency difference degree and a minimum adjacency difference degree; a mean shift identification subunit, configured to identify the mean shift of the plurality of adjacency difference degrees to determine a representative adjacency difference degree; and a channel construction subunit, configured to construct the time-series analysis multi-scale channel set according to the maximum adjacency difference degree, the minimum adjacency difference degree, and the representative adjacency difference degree.

[0012] In a possible implementation, the time-series interaction analysis unit further includes: an extraction subunit, configured to perform multi-group random non-replacement extraction on the user joint linkage feature set, and extract two user joint linkage features in each group to obtain a plurality of user joint linkage feature groups; an interaction matrix construction subunit, configured to respectively calculate an intra-group similarity set of the plurality of user joint linkage feature groups, and construct a plurality of intra-group adjacency interaction matrices; a feature group interaction analysis subunit, configured to perform interaction analysis on the plurality of user joint linkage feature groups by using the plurality of intra-group adjacency interaction matrices to obtain a plurality of interaction user joint linkage feature groups; and a mean processing subunit, configured to perform mean processing on the plurality of interaction user joint linkage feature groups to obtain the interaction joint linkage feature.

[0013] In a possible implementation, the scheme recommendation module further includes: a sample acquisition unit, configured to acquire a plurality of sample interactive joint linkage features, a plurality of sample target joint rehabilitation recommendation schemes identified in combination with the joint rehabilitation knowledge graph, and a plurality of sample feedback information of the plurality of sample target joint rehabilitation recommendation schemes; a scheme division unit, configured to divide the plurality of sample target joint rehabilitation recommendation schemes based on the plurality of sample feedback information, to obtain a plurality of positive sample target joint rehabilitation recommendation schemes and a plurality of negative sample target joint rehabilitation recommendation schemes; a mapping division unit, configured to map the plurality of sample interactive joint linkage features and the plurality of sample target joint rehabilitation recommendation schemes according to the plurality of positive sample target joint rehabilitation recommendation schemes and the plurality of negative sample target joint rehabilitation recommendation schemes, to obtain a positive sample set and a negative sample set; a supervised training unit, configured to perform supervised training on a framework constructed based on a feedforward neural network by using the positive sample set, the negative sample set, and the joint rehabilitation knowledge graph, until training converges, to obtain a rehabilitation scheme recommender; and an identification unit, configured to identify the joint rehabilitation knowledge graph and the interactive joint linkage features by using the rehabilitation scheme recommender, to obtain a target joint rehabilitation recommendation scheme.

[0014] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0015] In the present application, a joint rehabilitation knowledge graph is constructed, wherein the joint rehabilitation knowledge graph is a multi-level graph structure containing joint position linkage relationships and joint function linkage relationships; joint association information of a target user is acquired, and entities and relationships are aligned with the joint rehabilitation knowledge graph to construct a user joint linkage graph, wherein the joint association information includes joint angles, angular velocities, gait skeleton sequences, pain scores, physiological loads, and diagnosis and treatment records; real-time joint gait video frame sequences of the target user are acquired, and video frame analysis is performed in combination with the user joint linkage graph to construct real-time user joint linkage graph sequences; multi-scale time sequence interaction analysis is performed on the user joint linkage graph sequences to obtain interactive joint linkage features; and rehabilitation scheme recommendation is performed on the interactive joint linkage features based on the joint rehabilitation knowledge graph to obtain a target joint rehabilitation recommendation scheme. The technical effect of optimizing the acquisition of joint linkage features and providing accurate, efficient, and personalized rehabilitation schemes is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A structure schematic diagram of a joint rehabilitation scheme recommendation system based on a knowledge graph is provided for an embodiment of the present application.

[0017] Figure 2 A structure schematic diagram of a linkage graph construction module in a joint rehabilitation scheme recommendation system based on a knowledge graph is provided for an embodiment of the present application.

[0018] Figure labeling: Graph construction module 11, Linkage graph construction module 12, Video frame parsing module 13, Interactive parsing module 14, Scheme recommendation module 15. Detailed Implementation

[0019] This application provides a knowledge graph-based joint rehabilitation program recommendation system to address the technical problem that existing joint rehabilitation programs cannot be personalized and have low reliability.

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] It should be noted that the terms "comprising" and "having" are intended to cover non-exclusive inclusion. For example, a process, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to these processes, systems, products, or devices.

[0022] Examples, such as Figure 1 As shown, this application provides a knowledge graph-based joint rehabilitation program recommendation system, wherein the system includes:

[0023] The graph construction module is used to build a joint rehabilitation knowledge graph, which is a multi-level graph structure that includes the linkage relationship between joint positions and the linkage relationship between joint functions.

[0024] Furthermore, the map construction module also includes:

[0025] The rehabilitation position linkage layer construction unit is used to construct the joint position linkage layer according to the joint node type and node position relationship;

[0026] The rehabilitation function linkage layer construction unit is used to construct a joint function linkage layer based on the functional node linkage relationship between different joint nodes;

[0027] The rehabilitation goal-path layer building unit is used to obtain multiple rehabilitation task goals and their corresponding multiple rehabilitation paths, map and associate them, and construct the rehabilitation goal-path layer.

[0028] The relation definition unit is used to define multiple types of entities and their corresponding entity attributes. It combines the joint position linkage layer, joint function linkage layer and rehabilitation goal-path layer to define the relationships between entities and construct the joint rehabilitation knowledge graph.

[0029] In one possible embodiment, functional node linkages are used to describe the collaborative working relationships between different joints. For example, when the knee joint flexes and extends, the hip joint may also need to perform a certain degree of flexion, extension, or stabilization; the functional linkages between the two describe this interdependence. The rehabilitation goal-path layer is used to associate rehabilitation goals, such as improving knee joint range of motion and muscle strength, with the training paths required to actually achieve these goals, such as flexion-extension training and weight-bearing training, thereby ensuring that each rehabilitation goal has specific training steps and pathways to support it.

[0030] A joint position linkage layer was constructed based on the node types of the joints and their positional relationships. By collecting three-dimensional anatomical data of joints such as the knee and hip joints, such as MRI or X-ray data and motion capture data, the relative position of each joint in space was determined. For example, the relative positions of the knee and hip joints can be captured using gait analysis devices such as Vicon to obtain precise joint positional relationships, thereby providing spatial coordinate data support for subsequent atlas analysis.

[0031] The functional linkage layer is constructed by defining the functional linkage relationships between different joints, such as the synergistic effect of knee flexion and extension and hip flexion and extension, and then constructing the joint function linkage layer. For example, the joint function linkage layer can describe that when the angle of knee flexion and extension changes from 0° to 80°, the angle of hip flexion and extension ranges from 0° to 40°, and the coordinated movement of the two reflects the functional linkage relationship between the knee and hip joints.

[0032] Preferably, a rehabilitation goal-path layer is constructed by linking multiple rehabilitation goals, such as knee joint range of motion and muscle strength, with corresponding rehabilitation pathways, such as flexion-extension training and weight-bearing training, through a rehabilitation goal-path layer construction unit. This lays the groundwork for subsequent selection of suitable training pathways. For example, if knee joint range of motion is low, knee flexion training will be recommended as the first stage of rehabilitation based on the goal-path layer, with the intensity and difficulty of training gradually increased. The relationship definition unit is used to define the relationship between joint position, functional linkage, and rehabilitation pathways and goals.

[0033] Preferably, joint entities may have attributes such as location, type, and range of motion; functional entities may have attributes such as movement pattern and range of motion; and task entities may have attributes such as training goal, intensity, and duration. Based on the relationships between different entities in the joint position linkage layer, joint function linkage layer, and rehabilitation goal-path layer, relationships are defined to obtain a multi-level graph structure containing joint position linkage relationships and joint function linkage relationships, thus completing the construction of a joint rehabilitation knowledge graph.

[0034] The linkage graph construction module is used to obtain the joint association information of the target user, align the entities and relationships with the joint rehabilitation knowledge graph, and construct the user joint linkage graph. The joint association information includes joint angles, angular velocities, gait skeleton sequences, pain scores, physiological loads, and medical records.

[0035] Furthermore, such as Figure 2 As shown, the linkage diagram construction module also includes:

[0036] The standardization processing unit is used to collect joint data of the target user through multi-dimensional Internet of Things sensors, perform data standardization processing, and obtain the joint association information;

[0037] The entity diffusion unit is used to diffuse the joint-related information from three dimensions: joint entities, functional entities, and task entities, in conjunction with the joint rehabilitation knowledge graph, to determine the diffusion entity set.

[0038] The relationship alignment unit is used to align the relationships of the diffusion entity set based on location linkage relationships, functional linkage relationships, and target-path associations, and to construct the user joint linkage graph.

[0039] In one possible embodiment, joint-related information comprises various data related to the target user's joint function, including joint angles, angular velocities, gait skeleton sequences, pain scores, physiological load, and medical records. This data reflects changes in the user's joint status, mobility, and rehabilitation process. Joint angles refer to the angular changes of a joint during flexion, extension, and rotation, while angular velocity is the rate of joint movement. This helps to understand the range and speed of joint movement in the target user. Gait skeleton sequences are gait data obtained through sensors or video analytics, typically represented as a skeletal model, where each skeletal node represents a joint in the human body. The gait sequence represents the temporal changes of a joint during walking or other movements.

[0040] Joint data from users is collected using multi-dimensional IoT sensors, such as IMUs, cameras, and heart rate monitors. The collected data is standardized using a minimum-maximum normalization (MPM) method to obtain joint association information. Then, this joint association information is integrated with existing entities in a joint rehabilitation knowledge graph across three dimensions. For example, if a knee joint angle of 30° is detected and the knee is in a flexed position, the knee flexion / extension functional entity is matched with this angle data, generating a new knee flexion / extension functional entity, which is then associated with other joints, tasks, and goals in the graph. Through this diffusion process, user data is continuously connected to entities in the knowledge graph, forming a dynamic and personalized set of diffused entities.

[0041] Based on the positional, functional, and target-path relationships in the graph, the system aligns the relationships of the diffuse entity set. For example, if there is a linkage between knee flexion / extension and hip flexion / extension, the system will automatically match the corresponding functional linkage in the graph based on the user's joint angle data. Through this relationship alignment, the user's joint data can be accurately matched with the entities, functions, and targets in the graph, generating a complete user joint linkage graph.

[0042] The video frame parsing module is used to acquire the real-time joint gait video frame sequence of the target user, and combine it with the user joint linkage diagram to parse the video frames and construct a real-time user joint linkage diagram sequence.

[0043] Furthermore, the video frame parsing module also includes:

[0044] An image enhancement unit is used to traverse the real-time joint gait video frame sequence to perform image enhancement, thereby obtaining an enhanced real-time joint gait video frame sequence.

[0045] The node convolutional mapping unit is used to perform node convolutional mapping on each enhanced real-time joint gait video frame in the enhanced real-time joint gait video frame sequence, using the user joint linkage graph as a framework, to obtain a real-time user joint linkage graph sequence.

[0046] In one embodiment, a continuous sequence of video frames is acquired using a camera device, such as an RGB camera, a depth camera, or a motion capture system. Each frame contains the position information of the user's joints during movement, thereby obtaining the real-time joint gait video frame sequence. Image processing techniques, such as brightness adjustment, noise reduction, and contrast enhancement, are used to enhance the image of the real-time joint gait video frame sequence, ensuring that the joint information in the video frames is not affected by environmental factors, such as insufficient light or blurring, thereby improving the accuracy of joint recognition.

[0047] The user-defined joint mapping map matches the joint positions in video frames with the joint nodes in the mapping map. Assuming the knee flexion-extension angle is 40°, a convolution operation maps the knee flexion-extension angle data from the video frame to the knee joint node in the user-defined joint mapping map. The corresponding node position is then found in the mapping map using the convolution operation, and this data is updated in the mapping map.

[0048] The interaction parsing module is used to perform multi-scale temporal interaction parsing on the user joint linkage diagram sequence to obtain interactive joint linkage features;

[0049] Furthermore, the interactive parsing module also includes:

[0050] Channel pre-construction units are used to pre-construct multi-scale channel sets for time series analysis;

[0051] The temporal feature parsing unit is used to call the multi-scale channel set of temporal analysis to perform multi-scale temporal feature parsing on the user joint linkage diagram sequence to obtain the user joint linkage feature set.

[0052] The temporal interaction parsing unit is used to perform multi-scale temporal interaction parsing on the user joint linkage feature set to obtain the interactive joint linkage features.

[0053] Furthermore, the pre-built channel units also include:

[0054] The difference identification subunit is used to identify the adjacency difference of the user joint linkage graph sequence and obtain multiple adjacency differences.

[0055] Extraction sub-units are used to extract the maximum and minimum values ​​of the multiple adjacency differences to obtain the maximum and minimum adjacency differences;

[0056] The mean drift identification subunit is used to identify the mean drift of the multiple adjacency differences and determine the representative adjacency difference.

[0057] The channel construction subunit is used to construct a multi-scale channel set for time series analysis based on the maximum adjacency difference, minimum adjacency difference, and representative adjacency difference.

[0058] Furthermore, the timing interaction parsing unit also includes:

[0059] The extraction subunit is used to perform multiple random non-replacement extractions on the user joint linkage feature set, extracting two user joint linkage features in each group to obtain multiple user joint linkage feature groups.

[0060] The interaction matrix construction sub-unit is used to calculate the intra-group similarity set of multiple user joint linkage feature groups and construct multiple intra-group adjacency interaction matrices.

[0061] The feature group interaction parsing subunit is used to perform interaction parsing on the multiple user joint linkage feature groups using multiple intra-group adjacency interaction matrices to obtain multiple interactive user joint linkage feature groups.

[0062] The mean processing subunit is used to perform mean processing on multiple interactive user joint linkage feature groups to obtain interactive joint linkage features.

[0063] In one embodiment, two consecutive frames of joint linkage graphs from a user joint linkage graph sequence are read, and attribute data of corresponding joint nodes in the two graphs are extracted, such as knee joint angle, hip joint angular velocity, and edge relationship data between nodes, such as the positional linkage weights of the knee and hip joints. Then, the distance between the two graphs is calculated using the Euclidean distance formula, and the calculation result is used as the adjacency difference. Based on this principle, Euclidean distance is calculated for every two adjacent user joint linkage graphs in the sequence to obtain multiple adjacency differences.

[0064] Furthermore, the mean shift algorithm is used to identify the mean shift of the multiple adjacency differences to determine the representative adjacency difference. A kernel function is set, such as the Gaussian kernel function K(x)=exp(-x² / (2σ²)), where σ is the bandwidth parameter and the iteration termination threshold. Then, an initial center point is randomly selected from the multiple adjacency differences, and the mean shift vector of all data in the neighborhood of this point is calculated, i.e., the density gradient direction. The center point is moved to the mean position, and this process is repeated until the center point's movement distance is less than the iteration termination threshold, resulting in a density peak. This reveals the central tendency of the adjacency differences, providing a core basis for the intermediate scale division of multi-scale channels. Furthermore, by combining the maximum adjacency difference, the minimum adjacency difference, and the representative adjacency difference, a multi-scale channel set for time series analysis is constructed.

[0065] In other words, the maximum adjacency difference, minimum adjacency difference, and representative adjacency difference are used as the receptive fields for a single analysis, and the receptive fields of the feedforward neural network are updated accordingly, thereby obtaining the multi-scale channel set for time-series analysis.

[0066] In one embodiment, a reasonable number of sampling groups is determined based on the total number of features to ensure coverage of a sufficient number of feature combinations while avoiding duplication. Then, following the principle of random selection without replacement, two different features are sequentially selected from the feature set to form a group. This operation is repeated until extraction is complete, ultimately resulting in multiple user joint linkage feature groups. The two features within each group are decomposed into multiple core dimensions such as amplitude, trend of change, and key node response. For each dimension, the consistency between the two features is compared, and the similarity of that dimension is calculated. The higher the similarity, the stronger the association between the two features in that dimension. The similarities of all dimensions are summarized to form an intra-group similarity set. Finally, based on the similarity set, the similarity of each dimension is used as matrix elements to construct an intra-group adjacency interaction matrix that reflects the strength of multi-dimensional associations between features.

[0067] Based on the similarity of each dimension in the matrix, weights are assigned to different dimensions of the two features, with higher weights for dimensions with higher similarity and lower weights for dimensions with lower similarity. Then, according to the assigned weights, the information of each dimension of the two features is enhanced and fused, highlighting the feature information of strongly correlated dimensions while weakening irrelevant information of weakly correlated dimensions, ultimately resulting in the interactive user joint linkage feature group corresponding to each feature group.

[0068] Then, the overall mean of multiple interactive user joint linkage feature groups is calculated to obtain the interactive joint linkage features.

[0069] The solution recommendation module is used to recommend rehabilitation solutions based on the joint rehabilitation knowledge graph and the interaction joint linkage features to obtain a target joint rehabilitation recommendation solution.

[0070] Furthermore, the solution recommendation module also includes:

[0071] The sample acquisition unit is used to acquire multiple sample interaction joint linkage features and multiple sample target joint rehabilitation recommendation schemes for scheme identification by combining joint rehabilitation knowledge graph, as well as multiple sample feedback information of multiple sample target joint rehabilitation recommendation schemes.

[0072] The scheme division unit is used to divide the target joint rehabilitation recommendation schemes of multiple samples based on the feedback information of the multiple samples, so as to obtain multiple positive sample target joint rehabilitation recommendation schemes and multiple negative sample target joint rehabilitation recommendation schemes.

[0073] The mapping and partitioning unit is used to map and partition the multiple sample interactive joint linkage features and multiple sample target joint rehabilitation recommendation schemes according to multiple positive sample target joint rehabilitation recommendation schemes and multiple negative sample target joint rehabilitation recommendation schemes, so as to obtain a positive sample set and a negative sample set.

[0074] The supervised training unit is used to supervise the training of the framework based on the feedforward neural network using the positive sample set, the negative sample set and the joint rehabilitation knowledge graph until the training converges and a rehabilitation program recommender is obtained.

[0075] The identification unit is used to identify the joint rehabilitation knowledge graph and the interactive joint linkage features using the rehabilitation program recommender to obtain a target joint rehabilitation recommendation program.

[0076] In one embodiment, a large amount of relevant data from historical joint rehabilitation users is collected through data accumulation or screening of public datasets. This includes the joint linkage diagram sequence for each user, the sample interactive joint linkage features extracted from it, and the sample target joint rehabilitation recommendation program recommended for them based on knowledge graph. At the same time, feedback information from users after using the sample program is collected through questionnaires, follow-up records, and sensor monitoring of rehabilitation effects to clarify the actual suitability of each sample program, such as significant effect, excessive intensity, or discomfort during movement.

[0077] Feedback on rehabilitation showing significant effects, appropriate movement, and reasonable intensity was categorized as positive samples; feedback on ineffective rehabilitation, excessive / insufficient intensity, movement discomfort, or increased joint pain was categorized as negative samples. The feedback information corresponding to the target joint rehabilitation recommendation plan for each sample was compared one by one, and all sample plans were divided into positive and negative target joint rehabilitation recommendation plans according to the classification criteria. Furthermore, the interaction joint linkage features of each sample were bound one-to-one with its corresponding target joint rehabilitation recommendation plan, forming feature-plan mapping pairs. Based on the previous classification results, feature-plan mapping pairs bound to positive sample plans were classified into the positive sample set, and feature-plan mapping pairs bound to negative sample plans were classified into the negative sample set.

[0078] The positive and negative sample sets are input into a framework constructed from a feedforward neural network for supervised training. Based on the joint linkage features of the input samples, corresponding rehabilitation plans are predicted. By calculating the difference between the predicted results and the actual situation, the model parameters are adjusted in reverse, and the training process is repeated until the loss value is reduced to a preset threshold, resulting in a stable rehabilitation plan recommender. The joint rehabilitation knowledge graph and the target user's joint linkage features are simultaneously input into the trained rehabilitation plan recommender. The recommender then uses model calculations to discover rehabilitation recommendations for the target joint.

[0079] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0080] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0081] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A knowledge graph-based joint rehabilitation program recommendation system, characterized in that, The system includes: The graph construction module is used to build a joint rehabilitation knowledge graph, which is a multi-level graph structure that includes the linkage relationship between joint positions and the linkage relationship between joint functions. The linkage graph construction module is used to obtain the joint association information of the target user, align the entities and relationships with the joint rehabilitation knowledge graph, and construct the user joint linkage graph. The joint association information includes joint angles, angular velocities, gait skeleton sequences, pain scores, physiological loads, and medical records. The video frame parsing module is used to acquire the real-time joint gait video frame sequence of the target user, and combine it with the user joint linkage diagram to parse the video frames and construct a real-time user joint linkage diagram sequence. The interaction parsing module is used to perform multi-scale temporal interaction parsing on the user joint linkage diagram sequence to obtain interactive joint linkage features; The solution recommendation module is used to recommend rehabilitation solutions based on the joint rehabilitation knowledge graph and the interaction joint linkage features to obtain a target joint rehabilitation recommendation solution.

2. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 1, characterized in that, The map construction module also includes: The rehabilitation position linkage layer construction unit is used to construct the joint position linkage layer according to the joint node type and node position relationship; The rehabilitation function linkage layer construction unit is used to construct a joint function linkage layer based on the functional node linkage relationship between different joint nodes; The rehabilitation goal-path layer building unit is used to obtain multiple rehabilitation task goals and their corresponding multiple rehabilitation paths, map and associate them, and construct the rehabilitation goal-path layer. The relation definition unit is used to define multiple types of entities and their corresponding entity attributes. It combines the joint position linkage layer, joint function linkage layer and rehabilitation goal-path layer to define the relationships between entities and construct the joint rehabilitation knowledge graph.

3. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 1, characterized in that, The linkage diagram construction module also includes: The standardization processing unit is used to collect joint data of the target user through multi-dimensional Internet of Things sensors, perform data standardization processing, and obtain the joint association information; The entity diffusion unit is used to diffuse the joint-related information from three dimensions: joint entities, functional entities, and task entities, in conjunction with the joint rehabilitation knowledge graph, to determine the diffusion entity set. The relationship alignment unit is used to align the relationships of the diffusion entity set from the perspectives of location linkage, function linkage, and target-path association, and to construct the user joint linkage graph.

4. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 1, characterized in that, The video frame parsing module also includes: An image enhancement unit is used to traverse the real-time joint gait video frame sequence to perform image enhancement, thereby obtaining an enhanced real-time joint gait video frame sequence. The node convolutional mapping unit is used to perform node convolutional mapping on each enhanced real-time joint gait video frame in the enhanced real-time joint gait video frame sequence, using the user joint linkage graph as a framework, to obtain a real-time user joint linkage graph sequence.

5. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 1, characterized in that, The interactive parsing module also includes: Channel pre-construction units are used to pre-construct multi-scale channel sets for time series analysis; The temporal feature parsing unit is used to call the multi-scale channel set of temporal analysis to perform multi-scale temporal feature parsing on the user joint linkage diagram sequence to obtain the user joint linkage feature set. The temporal interaction parsing unit is used to perform multi-scale temporal interaction parsing on the user joint linkage feature set to obtain the interactive joint linkage features.

6. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 5, characterized in that, The pre-built channel unit also includes: The difference identification subunit is used to identify the adjacency difference of the user joint linkage graph sequence and obtain multiple adjacency differences. Extraction sub-units are used to extract the maximum and minimum values ​​of the multiple adjacency differences to obtain the maximum and minimum adjacency differences; The mean drift identification subunit is used to identify the mean drift of the multiple adjacency differences and determine the representative adjacency difference. The channel construction subunit is used to construct a multi-scale channel set for time series analysis based on the maximum adjacency difference, minimum adjacency difference, and representative adjacency difference.

7. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 5, characterized in that, The timing interaction parsing unit also includes: The extraction subunit is used to perform multiple random non-replacement extractions on the user joint linkage feature set, extracting two user joint linkage features in each group to obtain multiple user joint linkage feature groups. The interaction matrix construction sub-unit is used to calculate the intra-group similarity set of multiple user joint linkage feature groups and construct multiple intra-group adjacency interaction matrices. The feature group interaction parsing subunit is used to perform interaction parsing on the multiple user joint linkage feature groups using multiple intra-group adjacency interaction matrices to obtain multiple interactive user joint linkage feature groups. The mean processing subunit is used to perform mean processing on multiple interactive user joint linkage feature groups to obtain interactive joint linkage features.

8. The knowledge graph-based joint rehabilitation program recommendation system as described in claim 1, characterized in that, The solution recommendation module also includes: The sample acquisition unit is used to acquire multiple sample interaction joint linkage features and multiple sample target joint rehabilitation recommendation schemes for scheme identification by combining joint rehabilitation knowledge graph, as well as multiple sample feedback information of multiple sample target joint rehabilitation recommendation schemes. The scheme division unit is used to divide the target joint rehabilitation recommendation schemes of multiple samples based on the feedback information of the multiple samples, so as to obtain multiple positive sample target joint rehabilitation recommendation schemes and multiple negative sample target joint rehabilitation recommendation schemes. The mapping and partitioning unit is used to map and partition the multiple sample interactive joint linkage features and multiple sample target joint rehabilitation recommendation schemes according to multiple positive sample target joint rehabilitation recommendation schemes and multiple negative sample target joint rehabilitation recommendation schemes, so as to obtain a positive sample set and a negative sample set. The supervised training unit is used to supervise the training of the framework based on the feedforward neural network using the positive sample set, the negative sample set and the joint rehabilitation knowledge graph until the training converges and a rehabilitation program recommender is obtained. The identification unit is used to identify the joint rehabilitation knowledge graph and the interactive joint linkage features using the rehabilitation program recommender to obtain a target joint rehabilitation recommendation program.

Images