Knowledge graph-driven intelligent decision-making method and system for mastopathy rehabilitation training

Abstract

The invention relates to the technical field of data processing, in particular to a knowledge graph-driven mastopathy rehabilitation training intelligent decision-making method and a knowledge graph-driven mastopathy rehabilitation training intelligent decision-making system. Comprising the following steps: acquiring a current recovery state data set of a patient, extracting a dynamic perception variable group, and determining an individualized recovery trend vector; boundary division and weight adjustment are carried out according to the recovery trend vector, a trigger condition set is generated, and a proper training starting point is judged by combining a pre-established knowledge base matching decision support rule; if the monitoring requirement is met, activating a trigger mechanism to generate a preliminary scheme signal, and fusing individual difference data to determine a final rehabilitation training starting instruction; the instruction sequence is adjusted, the precision is verified through feedback circulation, and finally a precise rehabilitation training scheme is generated through optimization. The problems that an existing rehabilitation training scheme is insufficient in individuation and inaccurate in starting time are solved, dynamic adjustment based on the real-time recovery state of the patient and individual differences is achieved, and precise starting and optimization of rehabilitation training are ensured.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a knowledge graph-driven intelligent decision-making method and system for breast disease rehabilitation training. Background Technology

[0002] Postoperative rehabilitation training plays a crucial role in promoting functional recovery, alleviating postoperative symptoms, and improving quality of life for patients undergoing breast cancer surgery. However, traditional rehabilitation methods typically employ standardized training plans, lacking precise adjustments to meet the individualized needs of patients and failing to fully consider their postoperative recovery status, individual differences, and dynamic changes during the rehabilitation process. This "one-size-fits-all" approach not only easily leads to poor rehabilitation outcomes but may also increase unnecessary physical burden and even trigger new health problems.

[0003] Currently, with the development of medical technology, more and more medical devices and sensors are being used to monitor patients' physiological status in real time. However, how to effectively transform this multi-dimensional monitoring data into personalized rehabilitation training decisions remains a challenge that urgently needs to be addressed. Traditional decision support methods are mostly based on expert experience or simplified rules, failing to fully utilize patients' real-time data and personalized needs, resulting in insufficient accuracy and adaptability in rehabilitation training decisions.

[0004] To address the aforementioned issues, this invention proposes a knowledge graph-driven intelligent decision-making method for breast disease rehabilitation training. This method aims to generate personalized rehabilitation training plans by real-time collection of multi-dimensional postoperative monitoring data from patients, combined with knowledge graph analysis and decision support. By integrating sensor-collected data with decision rules in an established knowledge base, the system can dynamically adjust the timing and intensity of rehabilitation training based on the patient's recovery status and individual differences, thereby improving the accuracy and effectiveness of rehabilitation training. Summary of the Invention

[0005] This invention provides a knowledge graph-driven intelligent decision-making method and system for breast disease rehabilitation training. By combining real-time multi-dimensional monitoring data, individual difference analysis, decision support rules, and feedback mechanisms, it optimizes the generation process of rehabilitation training programs, thereby effectively improving the personalization, accuracy, and safety of rehabilitation training.

[0006] In a first aspect, the present invention provides a knowledge graph-driven intelligent decision-making method for breast disease rehabilitation training, comprising:

[0007] Step S1: Collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain the patient's current recovery status dataset; classify the patient's current recovery status dataset, extract dynamic perception variable groups, and determine individualized recovery trend vectors;

[0008] Step S2: Divide the individualized recovery trend vector into boundaries, adjust the vector weights, and obtain the trigger condition set; determine the appropriate training start point by matching decision support rules with a pre-established knowledge base.

[0009] Step S3: If the suitable training initiation point meets the monitoring requirements, the triggering mechanism is activated to generate a preliminary plan signal, integrates individual difference data, and determines the final rehabilitation training initiation instruction; based on the final rehabilitation training initiation instruction, potential changes are predicted, trend deviations are extracted, and an adjusted instruction sequence is obtained;

[0010] Step S4: Verify the accuracy of the adjusted instruction sequence through a feedback loop, iteratively optimize the condition set, and generate a precise rehabilitation training plan.

[0011] As a preferred embodiment of the present invention, step S1, obtaining the patient's current recovery status dataset, includes:

[0012] Data on wound healing progress, pain scores, and upper limb mobility indicators of breast cancer patients after surgery are collected in real time using sensor devices. These data are preprocessed to remove outliers, generating a standardized multi-dimensional monitoring dataset. The standardized multi-dimensional monitoring dataset is then integrated to form a continuous monitoring stream containing timestamps. Finally, a dataset representing the patient's current recovery status is constructed based on this continuous monitoring stream data.

[0013] As a preferred embodiment of the present invention, step S1, determining the individualized recovery trend vector, includes:

[0014] The patient's current recovery status dataset is input into a pre-trained random forest algorithm model; the random forest algorithm model is used to classify the patient's current recovery status dataset to identify individual differences; key dynamic perception variable groups are extracted based on the classification results, the dynamic perception variable groups contain core indicators related to the recovery status; feature weights are calculated for the dynamic perception variable groups to generate a vector reflecting individual recovery characteristics; and an individualized recovery trend vector is determined based on the vector.

[0015] As a preferred embodiment of the present invention, in step S2, the triggering condition set is obtained, including:

[0016] The individualized recovery trend vector is input into the support vector machine algorithm model; the support vector machine algorithm model is used to perform boundary division of the individualized recovery trend vector using multi-dimensional monitoring indicators, and the boundary division result is determined. The corresponding weights in the individualized recovery trend vector are adjusted according to the boundary division result and the pain score in the dynamic perception variable group; a trigger condition set is generated according to the adjusted individualized recovery trend vector and the boundary division result.

[0017] As a preferred embodiment of the present invention, step S2, determining a suitable training starting point by matching decision support rules with a pre-established knowledge base, includes:

[0018] The trigger condition set is input into a pre-established rehabilitation training knowledge base; the knowledge graph query module retrieves decision support rules related to the trigger condition set from the rehabilitation training knowledge base; the values ​​of each indicator in the trigger condition set are analyzed according to the decision support rules to determine whether the logical conditions for training initiation are met; if the logical conditions are met, the current time point is determined to be a suitable training initiation point.

[0019] As a preferred embodiment of the present invention, step S3, determining the final rehabilitation training initiation command, includes:

[0020] The appropriate training initiation point is verified by multi-dimensional monitoring requirements to confirm that all indicators are within the preset range; if the verification is successful, the triggering mechanism is activated to generate a preliminary plan signal; the latest individual difference data is obtained through the dynamic perception update module; the individual difference data is fused with the preliminary plan signal, the signal parameters are adjusted, and the final rehabilitation training initiation command is determined.

[0021] As a preferred embodiment of the present invention, step S3, obtaining the adjusted instruction sequence, includes:

[0022] The final rehabilitation training initiation command is input into a time series analysis model; the potential changes of the final rehabilitation training initiation command in future time periods are predicted using the time series analysis model; trend deviation data related to the prediction results are extracted from the continuous monitoring stream of the patient's current recovery status dataset; the final rehabilitation training initiation command is corrected based on the trend deviation data to generate an adjusted command sequence.

[0023] As a preferred embodiment of the present invention, step S4, generating a precise rehabilitation training plan, includes:

[0024] The adjusted instruction sequence is input into the decision support feedback loop module; the accuracy of the triggering mechanism of the adjusted instruction sequence is verified by the decision support feedback loop module; if the verification result shows that the deviation exceeds a preset threshold, the triggering condition set is iteratively adjusted; a new instruction sequence is regenerated based on the iteratively adjusted triggering condition set; through multiple feedback loop verifications, the accuracy of the new instruction sequence is ensured to meet the requirements.

[0025] Secondly, the present invention also provides a knowledge graph-driven intelligent decision-making system for breast disease rehabilitation training, used to implement the above-mentioned method, the system comprising:

[0026] The data acquisition unit is used to collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain a dataset of the patient's current recovery status; the data acquisition unit is used to classify and process the dataset of the patient's current recovery status, extract dynamic perception variable groups, and determine an individualized recovery trend vector.

[0027] The starting point determination unit is used to perform boundary division on the individualized recovery trend vector, adjust the vector weights, and obtain a set of triggering conditions; and to determine a suitable training starting point by matching decision support rules with a pre-established knowledge base.

[0028] The initiation command determination unit is used to activate the triggering mechanism to generate a preliminary plan signal, integrate individual difference data, and determine the final rehabilitation training initiation command when the monitoring requirements are met at the appropriate training initiation point.

[0029] The instruction optimization unit is used to predict potential changes based on the final rehabilitation training initiation instruction, extract trend deviations, and obtain an adjusted instruction sequence.

[0030] The training program generation unit is used to verify the accuracy of the adjusted instruction sequence through feedback loops, iteratively optimize the condition set, and generate accurate rehabilitation training programs.

[0031] Thirdly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0032] The beneficial effects of this invention are as follows:

[0033] This invention uses sensor devices to collect multi-dimensional monitoring data from breast cancer patients after surgery in real time, including wound healing progress, pain scores, and upper limb mobility. It extracts individualized recovery trend vectors to reflect the patient's recovery status, ensuring that the patient's rehabilitation data is accurately captured and transformed into a vector usable for decision-making. A Support Vector Machine (SVM) model is used to delimit the boundaries of the recovery trend vector, dynamically adjusting the weights of each monitoring indicator to generate a trigger condition set. This trigger condition set combines individual differences, recovery status, and recovery trends, ensuring that the timing and intensity of training initiation can be precisely adjusted according to the patient's real-time recovery progress. The trigger condition set is combined with decision support rules in a knowledge base. A knowledge graph query module matches relevant rules in the rehabilitation training knowledge base to determine whether the conditions for initiating training are met. The system identifies suitable training initiation points and generates preliminary program signals. It also integrates individual difference data to further optimize rehabilitation training instructions. A time series analysis model predicts instructions, extracts trend deviations, and adjusts the instruction sequence to ensure the training plan adapts to potential future changes. Through a feedback loop mechanism, the accuracy of the adjusted instruction sequence is verified, and the trigger condition set is iteratively optimized to ensure the final generation of an accurate rehabilitation training program. The synergy between these technical solutions solves the problem of lack of personalization and precision in traditional rehabilitation training. By monitoring data in real time and making individualized adjustments, it can flexibly respond to the patient's recovery status and individual differences, providing customized rehabilitation training programs. In addition, through multiple feedback optimizations, the training program can be continuously improved, enhancing the effectiveness and safety of rehabilitation training. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart of a knowledge graph-driven intelligent decision-making method for breast disease rehabilitation training, as shown in the embodiment.

[0036] Figure 2 This is a structural diagram of a knowledge graph-driven intelligent decision-making system for breast disease rehabilitation training in an embodiment.

[0037] Figure 3 This is a schematic diagram illustrating time series analysis and training intensity adjustment in the embodiment. Detailed Implementation

[0038] This invention provides a knowledge graph-driven intelligent decision-making method and system for breast disease rehabilitation training. The terms "first," "second," "third," "fourth," etc. (if applicable) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0039] For ease of understanding, the specific process of the embodiments of the present invention will be described below, such as... Figure 1 As shown in the figure, an intelligent decision-making method for breast disease rehabilitation training driven by knowledge graphs in an embodiment of the present invention includes:

[0040] Step S1: Collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain the patient's current recovery status dataset; classify the patient's current recovery status dataset, extract dynamic perception variable groups, and determine individualized recovery trend vectors;

[0041] In step S1, the patient's current recovery status dataset is obtained, including:

[0042] Data on wound healing progress, pain scores, and upper limb mobility indicators of breast cancer patients after surgery are collected in real time using sensor devices. These data are preprocessed to remove outliers, generating a standardized multi-dimensional monitoring dataset. The standardized multi-dimensional monitoring dataset is then integrated to form a continuous monitoring stream containing timestamps. Finally, a dataset representing the patient's current recovery status is constructed based on this continuous monitoring stream data.

[0043] Specifically, in this embodiment, multi-dimensional monitoring data of breast cancer patients after surgery is collected in real time using sensor devices to construct a dataset of the patient's current recovery status. The sensor devices used can capture data on indicators such as the patient's wound healing degree, pain score, and upper limb mobility, and the data for each monitoring indicator is collected by different types of sensors. For example, wound healing degree data is monitored by optical sensors to detect color changes and tissue density in the wound area; pain score data is recorded by pain assessment sensors to record the patient's subjective pain score, and is further validated by physiological indicators such as heart rate variability; and upper limb mobility data is quantified by motion tracking sensors to quantify the patient's joint range of motion and force output. In addition, the sensor devices can be integrated into a smart bandage to ensure the real-time nature of data acquisition and the accuracy of multi-dimensional monitoring.

[0044] After data acquisition, the data undergoes preprocessing to ensure its quality. Statistical methods, such as box plots, are used to identify and remove outliers caused by sensor errors or external interference. The outlier-removed data is then standardized using min-max normalization to scale the monitoring data across different dimensions to eliminate differences in data units and magnitudes, ensuring the reliability of subsequent analysis. After preprocessing, the system integrates the standardized data to form a continuous monitoring stream containing timestamps. This dataset includes not only real-time values ​​of various indicators but also timestamp information to ensure temporal continuity. To further improve data completeness, missing time points are filled using linear interpolation and other methods to ensure data stream continuity, laying the foundation for subsequent data analysis.

[0045] After constructing a dataset of patients' current recovery status, key features are extracted from continuous monitoring stream data, such as the average rate of change in wound healing, peak pain scores, and changes in upper limb mobility. Further analysis of this data identifies the patient's recovery stage, such as the early healing phase or the functional recovery phase, thus supporting the generation of individualized recovery trend vectors. Individual patient differences, such as age and gender, are also considered to ensure personalized adaptation of the recovery trend vectors. For example, for patients of different age groups, the weights of upper limb mobility indicators are adjusted based on differences in bone mineral density to improve the accuracy and adaptability of the recovery trend vectors. A completeness and validity analysis of the dataset is performed to ensure the accuracy and reliability of the data. The system calculates dataset integrity metrics such as coverage percentage to determine if there are serious data gaps and triggers a data supplementation mechanism for data below a threshold. For example, when dataset integrity is low, historical monitoring data is automatically reviewed to supplement missing parts and improve data integrity. Furthermore, correlation analysis, such as Pearson correlation coefficient, is used to assess the relationships between various monitoring indicators to ensure data logic consistency. For instance, the high correlation between wound healing progress and pain scores provides a reliable basis for subsequent recovery trend vector calculations. These technical solutions not only improve data quality but also lay a solid foundation for subsequent decision support rule matching and training initiation command generation, thereby enabling personalized and precise rehabilitation training programs.

[0046] Further, in step S1, determining the individualized recovery trend vector includes:

[0047] The patient's current recovery status dataset is input into a pre-trained random forest algorithm model; the random forest algorithm model is used to classify the patient's current recovery status dataset to identify individual differences; key dynamic perception variable groups are extracted based on the classification results, the dynamic perception variable groups contain core indicators related to the recovery status; feature weights are calculated for the dynamic perception variable groups to generate a vector reflecting individual recovery characteristics; and an individualized recovery trend vector is determined based on the vector.

[0048] Specifically, the patient's current recovery status dataset is input into a pre-trained random forest algorithm model. This recovery status dataset includes multi-dimensional data collected in real-time by sensor devices, such as wound healing progress, pain scores, upper limb mobility, recovery stage, and indicators like the patient's age, gender, and surgical type. The model is trained on a large amount of historical patient recovery data and corresponding labeled data. The labeled data represents the individual difference classification results corresponding to the historical patient recovery data. The model can identify individual differences based on the patient's recovery status dataset. Random forest is an ensemble learning method that constructs multiple decision trees and combines their predictions to improve overall prediction accuracy and robustness. In the initial stage of model training, the historical dataset is divided into multiple subsets to improve the model's generalization ability. To improve the effectiveness of rehabilitation, a "bootstrap" method is typically used to sample the training set. This involves drawing samples with replacement from the original dataset to generate multiple different subsets. Each subset is used to train a decision tree. Each decision tree splits the data based on different features, generating the tree structure. Finally, a majority voting mechanism determines the outcome of each decision. During training, the model continuously optimizes the splitting rules based on historical data on recovery stages and states, such as wound healing progress, pain scores, upper limb mobility, recovery stage, age, and surgical type. It also calculates the contribution of each feature to the splitting nodes in the decision tree to identify which features have a significant impact on classification. Through continuous iteration, the model enhances its ability to identify individual differences, ensuring it can accurately distinguish different recovery states and support individualized rehabilitation training programs.

[0049] Based on the classification results, i.e., the individual differences in recovery status corresponding to the input data, core indicators highly correlated with recovery status are selected to form a dynamic perception variable set. This variable set is further processed using methods such as information gain to ensure that the extracted features play a key role in monitoring the recovery status. For example, variables such as wound healing degree and pain score are often selected as core indicators because they directly reflect the patient's recovery progress. Then, time-series correlation analysis is performed on the above dynamic perception variable set to ensure that they can accurately reflect the dynamic changes in the patient's recovery process, and their coreness is further verified by calculating the correlation between variables.

[0050] After feature extraction, feature weights are calculated for each variable in the dynamically perceived variable set to generate a vector that accurately reflects the patient's recovery characteristics. Principal component analysis (PCA) is used to standardize the extracted features and calculate the covariance matrix. By extracting the principal components, the weights of each variable are determined. These weights are obtained through eigenvector projection, reflecting the contribution of each variable to the recovery state. The generated multidimensional vector is also weighted to represent the importance of different features in the recovery process, for example, in the form of [weight 1 * index 1 value, weight 2 * index 2 value], reflecting individual recovery characteristics, such as pain-dominated recovery. Based on the feature weight calculation results, a patient-specific recovery vector is calculated using a linear combination method, which involves weighting the indicators in the multidimensional vector. After normalization, a weighted sum is calculated, and the aforementioned multidimensional vector and recovery vector are combined into a recovery trend vector. This provides support for the formulation of subsequent rehabilitation training programs. The trend vector directly reflects the patient's current recovery direction and intensity. For example, when the weighted sum is positive, it indicates that the patient's recovery is good, providing data support for judging the timing of training initiation and adjusting the intensity of rehabilitation training. It can also update the patient's recovery status in real time and dynamically adjust the training program according to different recovery progresses, ensuring that each patient's rehabilitation process is personalized, precise, and efficient. The above technical solution, through data collection, classification, feature extraction, weight calculation, and trend vector generation, effectively achieves accurate monitoring and analysis of the patient's recovery status, ensuring personalized adjustments to the rehabilitation training program, thereby significantly improving the effectiveness of postoperative rehabilitation training for breast disease patients.

[0051] Step S2: Divide the individualized recovery trend vector into boundaries, adjust the vector weights, and obtain the trigger condition set; determine the appropriate training start point by matching decision support rules with a pre-established knowledge base.

[0052] In step S2, the triggering condition set is obtained, including:

[0053] The individualized recovery trend vector is input into the support vector machine algorithm model; the support vector machine algorithm model is used to perform boundary division of the individualized recovery trend vector using multi-dimensional monitoring indicators, and the boundary division result is determined. The corresponding weights in the individualized recovery trend vector are adjusted according to the boundary division result and the pain score in the dynamic perception variable group; a set of triggering conditions is generated based on the adjusted individualized recovery trend vector and the boundary division result.

[0054] Specifically, in this embodiment, the generation of the trigger condition set is achieved based on the support vector machine (SVM) algorithm model through further discrimination and parameter adjustment of the individualized recovery trend vector. Specifically, the individualized recovery trend vector formed by the random forest algorithm is used as input. The input includes multi-dimensional monitoring indicators such as wound healing degree, pain score, and upper limb mobility. This recovery trend vector, as a quantitative expression of the overall recovery state, is passed to the pre-trained SVM algorithm model. The SVM algorithm, through its supervised learning characteristics, constructs a maximum-margin separating hyperplane in a high-dimensional space to delineate the boundaries of different recovery states. In the context of breast cancer rehabilitation, this algorithm... By using nonlinear mapping mechanisms such as radial basis function kernels to map trend vectors to a high-dimensional feature space, the system can more sensitively distinguish between normal and abnormal ranges for each monitoring indicator. For example, in the postoperative monitoring scenario of breast cancer patients, assuming a patient's upper limb mobility index is 45 degrees, the support vector machine algorithm determines its critical range to be 30-60 degrees through boundary division. If it exceeds this range, it is marked as requiring attention, thereby adjusting the recovery trend vector as early as possible. The training data of the above support vector set algorithm model consists of the recovery trend vectors of historical patients and the corresponding boundary division results, that is, distinguishing which indicators are in the normal recovery range, which are in the high-risk range and their corresponding critical ranges.

[0055] After inputting the trend vector, the support vector machine model first analyzes the monitoring indicators in each dimension of the vector, treating them as multi-dimensional feature points. Based on the boundary structure learned during the training phase, it divides the range of change of each monitoring indicator into boundaries, forming a critical range representing the normal state and the risk state. At this point, monitoring variables such as wound healing degree, pain score, and upper limb mobility are mapped to their corresponding high-dimensional spatial feature positions. The model judges the recovery state based on its relative position from the hyperplane. For example, when upper limb mobility is within the normal range corresponding to the training set, the indicator is marked as requiring no intervention. If its value is close to the hyperplane or exceeds the boundary, it is identified as a high-risk range of abnormal recovery, increasing the weight of the corresponding indicator in the recovery trend vector. This allows the rehabilitation training program to capture potential deterioration trends in advance and avoid secondary injuries caused by inappropriate training intensity.

[0056] After boundary delineation, the trend vector is further weighted by combining real-time pain data from patients in the dynamically perceived variable group. If the pain score exceeds a preset safety threshold, the weight of the pain-related components is increased to strengthen the influence of pain in the overall trend, making the trend vector more reflective of the patient's current actual discomfort and suppressing other secondary factors that may interfere with the judgment. Since pain is a key variable affecting the timing of training initiation during post-mastopathy surgery rehabilitation, the above weight adjustment mechanism enables the individualized recovery trend vector to truly respond to patient differences. When pain increases, the expression of the pain variable in the trend vector is automatically increased, making the trend vector as a whole biased towards "recovery risk," thereby effectively avoiding incorrect training initiation when pain has not been relieved. Subsequently, the distribution characteristics of the individualized recovery trend vector are recalculated based on the updated weights. That is, by weighting and summing each monitored variable according to the new weights, a new quantitative expression of the overall recovery state is obtained, which not only changes the direction of the trend vector but also its multidimensional representation. The distribution pattern of the trend vector makes it more closely reflect the dynamic changes in the patient's actual recovery status. After the recovery trend vector is updated, it is passed to the trigger condition set module. The trigger condition set module determines whether the current conditions for training initiation are met based on the updated trend vector, combined with the weight, threshold, and boundary division conditions of each monitoring indicator. For example, if the updated trend vector shows that the wound healing degree has reached the set threshold, such as 80%, and the pain score is low (e.g., below the set threshold of 5 points), and the upper limb mobility is greater than the set minimum recovery range, such as 30 degrees, then the training initiation conditions are determined to be met. Furthermore, based on the recovery trend vector and the rules in the trigger condition set, a new trigger condition set is generated by comparing each recovery indicator with the preset critical range. The trigger condition set includes the recovery indicators and their corresponding adjusted weights, as well as the training initiation conditions corresponding to each recovery. For example, if the recovery of upper limb mobility and the pain score meet the corresponding trigger training conditions, then training is initiated, and the patient's gender and age are also added to the trigger condition set.

[0057] The above technical solution can obtain an optimized set of triggering conditions that accurately reflects the individual recovery status of patients, laying the foundation for subsequent knowledge graph rule matching. This enables the decision-making process for initiating rehabilitation training to shift from fixed threshold judgment to dynamic and personalized judgment, thereby strengthening the technical solution's ability to adapt to the differences in recovery among patients after breast disease surgery and ensuring the scientific and safe timing of rehabilitation training initiation.

[0058] Further, in step S2, a suitable training starting point is determined by matching decision support rules with a pre-established knowledge base, including:

[0059] The trigger condition set is input into a pre-established rehabilitation training knowledge base; the knowledge graph query module retrieves decision support rules related to the trigger condition set from the rehabilitation training knowledge base; the values ​​of each indicator in the trigger condition set are analyzed according to the decision support rules to determine whether the logical conditions for training initiation are met; if the logical conditions are met, the current time point is determined to be a suitable training initiation point.

[0060] Specifically, since the aforementioned trigger condition set can only preliminarily determine the training initiation conditions corresponding to the aforementioned recovery indicators and cannot accurately obtain the training initiation point, the trigger condition set is input into a pre-established rehabilitation training knowledge base. First, the knowledge graph query module retrieves decision support rules related to the trigger condition set. Specifically, the knowledge graph organizes rehabilitation training-related data into an entity-relationship-entity structure, constructing the graph's nodes and edges. Nodes represent rehabilitation indicators such as wound healing degree, pain score, and upper limb mobility, while edges represent the logical relationships between these indicators, such as the causal relationship between pain score and mobility. This effectively maps the patient's dynamic monitoring data and personalized recovery trend vector, identifying the relationship between the patient's current recovery status and the target recovery trend. The system employs decision support rules highly correlated with the recovery state. During the query process, the knowledge graph query module traverses the graph using the SPARQL query language to identify recovery indicators in the trigger condition set, such as pain scores. Based on the query results, it extracts a set of decision support rules related to the recovery indicators. These rules include conditional thresholds and weights related to training initiation. For example, if the pain score exceeds the threshold and the weight exceeds the set weight, the rules retrieved from the knowledge graph may indicate a delay in training initiation to avoid aggravating the wound burden. The knowledge graph can adjust the decision rules according to individual patient differences. For example, for patients of different age groups, rules containing age factors will be prioritized, thereby ensuring that the judgment of training initiation is more personalized and accurate.

[0061] Based on the aforementioned decision support rules, the analysis of each recovery indicator in the trigger condition set determines whether the logical conditions for training initiation are met. Specifically, each monitoring indicator, i.e., the aforementioned recovery indicator such as pain score, is compared with the threshold in the rules, and a logical expression is applied to evaluate the combined conditions. For example, if the pain score is below a certain value and the upper limb mobility exceeds a certain threshold, the conditions for training initiation are considered met. Finally, a Boolean value is output, indicating whether the conditions for training initiation are met. If the conditions are met, it is further determined whether the current time point is a suitable training initiation point. The aforementioned time point is recorded by recording the current timestamp and marking it as the initiation point to ensure timeliness and provide a basis for subsequent rehabilitation training decisions.

[0062] After determining the appropriate training initiation point, the relevant parameters of the initiation point, such as the values ​​of various indicators in the trigger condition set and the timestamp, are stored in the log database to form a reference dataset of activation signals. Through this recording method, the triggering mechanism can obtain effective data support, ensuring the accurate initiation of the rehabilitation training program, and thus realizing the dynamic adjustment and optimization of the patient's rehabilitation training plan. The above technical solution, by combining deep semantic query of knowledge graph and matching of decision support rules, can realize personalized, data-driven rehabilitation training decisions based on multi-dimensional monitoring data and individual differences, significantly improving the accuracy and safety of rehabilitation training.

[0063] Step S3: If the suitable training initiation point meets the monitoring requirements, the triggering mechanism is activated to generate a preliminary plan signal, integrates individual difference data, and determines the final rehabilitation training initiation instruction; based on the final rehabilitation training initiation instruction, potential changes are predicted, trend deviations are extracted, and an adjusted instruction sequence is obtained;

[0064] In step S3, the final rehabilitation training initiation instruction is determined, including:

[0065] The appropriate training initiation point is verified by multi-dimensional monitoring requirements to confirm that all indicators are within the preset range; if the verification is successful, the triggering mechanism is activated to generate a preliminary plan signal; the latest individual difference data is obtained through the dynamic perception update module; the individual difference data is fused with the preliminary plan signal, the signal parameters are adjusted, and the final rehabilitation training initiation command is determined.

[0066] Specifically, the determination of the final rehabilitation training initiation command first involves verifying the multi-dimensional monitoring requirements of the appropriate training initiation point to confirm that all monitoring indicators, i.e., recovery indicators, are within the preset range, ensuring the safety and effectiveness of the training initiation. Specifically, sensor devices are used to collect data on the patient's wound healing progress, pain score, and upper limb mobility in real time, and these are compared with the corresponding preset ranges. If all indicators meet the preset range, i.e., the training initiation range, the verification is confirmed as successful, ensuring data reliability. This activates the trigger mechanism and generates a preliminary rehabilitation training plan signal. Furthermore, the latest individual difference characteristics and corresponding recovery trend vectors are re-obtained through step S1, and the variables in the recovery trend vector are boundary-defined and weighted using a support vector machine algorithm. For example, if the pain score exceeds a set threshold, the weights of pain-related variables are adjusted to increase their influence in the recovery trend, ensuring that the treatment plan prioritizes pain management. This allows the system to dynamically adapt to individual patient differences, such as age and recovery ability, thereby improving data accuracy and the effectiveness of personalized adjustments. The extracted individual difference data, i.e., the aforementioned individual difference characteristics, are then fused with the preliminary plan signal to adjust signal parameters. Finally, time series analysis methods are applied to further refine the signal. The system measures potential changes in individual differences and adjusts parameters in the initial program signal based on prediction results. For example, if a patient's upper limb mobility index shows a positive recovery trend but a high pain score, the pain weighting parameter will be adjusted during fusion to make the signal more focused on progressive training. In another embodiment, for patients with strong mobility in the recovery trend vector, the fusion will strengthen the parameters of the initiation timing, accelerate program generation, and predict deviations through time series analysis to ensure that the parameters reflect true individual differences, which is beneficial to improving rehabilitation efficiency. If the deviation exceeds a preset threshold during the fusion process, a feedback loop is performed to iteratively optimize the parameters of the initial program signal to ensure the precision and accuracy of the rehabilitation training initiation command. Finally, the adjusted signal parameters generate the final rehabilitation training initiation command, which includes the patient's recovery indicators such as wound healing degree, pain score, upper limb mobility, and suggestions for the initial training initiation signals such as training timing and intensity. This command will guide the implementation of subsequent rehabilitation training programs. Through the above technical solutions, scientific and personalized rehabilitation training initiation timing and programs can be provided based on dynamic perception and real-time data updates, ensuring that patients receive training at the optimal time and avoiding initiation at inappropriate times, thereby promoting better rehabilitation outcomes.

[0067] Further, in step S3, the adjusted instruction sequence is obtained, including:

[0068] The final rehabilitation training initiation command is input into a time series analysis model; the potential changes of the final rehabilitation training initiation command in future time periods are predicted using the time series analysis model; trend deviation data related to the prediction results are extracted from the continuous monitoring stream of the patient's current recovery status dataset; the final rehabilitation training initiation command is corrected based on the trend deviation data to generate an adjusted command sequence.

[0069] Specifically, in this embodiment, the potential changes of the instruction within a future time period are predicted by inputting the final rehabilitation training initiation instruction into a time series analysis model. Specifically, the final rehabilitation training initiation instruction is first obtained. This instruction is generated based on a set of triggering conditions and contains information about suitable training initiation points. The time series analysis model is a model trained on a sequence of historical rehabilitation training initiation instructions. The instruction is then converted into a time series format, including serializing the timestamps and related recovery indicator values ​​in the instruction to ensure the temporality and coherence of the data. The converted data is then input into the time series analysis model. In the time series analysis model, the input rehabilitation training initiation instruction is combined with historical rehabilitation training initiation instructions input within adjacent preset time periods to form an instruction sequence, using A... The RIMA model is used to capture the autocorrelation and trend components of the instruction sequence. The ARIMA model eliminates non-stationarity by performing differencing on the instruction sequence and, based on estimates of the differencing order, autoregressive order, and moving average order, uses the minimization of the AIC criterion to fit parameters, thereby generating trend predictions. Based on the prediction results of the time series analysis model, further analysis is conducted on potential future changes, especially for variables with significant dynamic changes during rehabilitation, such as pain scores and upper limb mobility. In this process, by setting the prediction time range, the time series analysis model can generate a predicted sequence of future trends and calculate the expected fluctuations of the aforementioned variables. To ensure the accuracy of the prediction results, the confidence interval of the prediction results is also evaluated to make reasonable estimates of uncertainty, for example, such as... Figure 3 As shown, if the prediction indicates that the pain score may rise in the future, it is possible to identify potential recovery risks and adjust the training intensity or timing in advance to avoid the risks of training too early or inappropriately.

[0070] Simultaneously, trend deviation data related to the prediction results are extracted from the continuous monitoring stream of the patient's current recovery status dataset. A time-series dataset is formed using real-time collected data on wound healing progress, pain scores, and upper limb mobility. The root mean square error formula is used to calculate the deviation between the predicted sequence and the actual monitored values. This deviation data serves as a crucial basis for further adjustment instructions, enabling the system to dynamically respond to individual patient differences and ensuring that each training instruction accurately reflects the patient's current recovery status. For example, if the monitoring stream shows a decline in upper limb mobility while the predicted value is stable, this deviation is extracted and incorporated into the correction process to optimize the initiation conditions for rehabilitation training. Based on the extracted trend deviation data, the final rehabilitation training initiation instruction is revised, adjusting various parameters in the instruction sequence to ensure that the instructions dynamically adapt to the patient's recovery progress. During the adjustment process, various parameters in the deviation data are weighted and adjusted. For instance, if the deviation exceeds a set threshold, the training initiation time may be delayed to ensure that rehabilitation training starts when the patient's condition is most suitable, thereby maximizing the rehabilitation effect.

[0071] After the above correction and optimization process, an adjusted instruction sequence is generated and its stability is analyzed to ensure that it can be stably applied to subsequent decision support rule verification. Specifically, the variance and autocorrelation coefficient of the adjusted instruction sequence are calculated to assess its stability. Simulation tests are then conducted to match the sequence with rules in the knowledge graph query module to verify whether the initiation conditions for rehabilitation training are met. If the analysis results show that the stability of the sequence meets the requirements, the final rehabilitation training initiation instruction can be determined to ensure accurate initiation of training and compliance with individualized rehabilitation needs.

[0072] Step S4: Verify the accuracy of the adjusted instruction sequence through a feedback loop, iteratively optimize the condition set, and generate a precise rehabilitation training plan; specifically including:

[0073] The adjusted instruction sequence is input into the decision support feedback loop module; the accuracy of the triggering mechanism of the adjusted instruction sequence is verified by the decision support feedback loop module; if the verification result shows that the deviation exceeds a preset threshold, the triggering condition set is iteratively adjusted; a new instruction sequence is regenerated based on the iteratively adjusted triggering condition set; through multiple feedback loop verifications, the accuracy of the new instruction sequence is ensured to meet the requirements.

[0074] Specifically, the adjusted instruction sequence is input into the decision support feedback loop module. This feedback loop module captures sequence data through a receiving interface to ensure data integrity and provide reliable input for subsequent processing. The feedback loop module verifies the accuracy of the triggering mechanism of the adjusted instruction sequence, that is, the degree of matching between the start instruction in the instruction sequence and the actual patient recovery status dataset. The deviation value is calculated by comparing the trend deviation in the sequence with a preset standard. If the deviation exceeds the set range, the accuracy of the triggering mechanism is determined to be insufficient, and further adjustments are required.

[0075] When the verification results show that the deviation exceeds the preset threshold, the trigger condition set will be iteratively adjusted. Specifically, the weight values ​​in the condition set will be gradually modified using the gradient descent method to ensure that the condition set is more adapted to the current patient's recovery status data. Through the above iterative process, the trigger condition set can be updated after each round of verification to reduce future deviations and ensure that the condition set can better reflect individual differences and actual recovery needs. After the iterative adjustment is completed, a new instruction sequence will be regenerated and verified through multiple feedback loops to ensure that the accuracy of the newly generated instruction sequence meets the requirements. An upper limit for the number of loops is set, and the improvement in accuracy is evaluated after each loop to determine whether the preset accuracy standard has been reached. In each round of feedback loops, the deviation is gradually reduced by continuously optimizing the trigger condition set and adjusting the instruction parameters until the accuracy requirements are met.

[0076] After confirming that the new instruction sequence meets the requirements, the precise instruction sequence is output to the dynamic perception update module and integrated with the current recovery status dataset to generate the final rehabilitation training plan. This technical solution ensures the accuracy and personalized adaptation of rehabilitation training instructions, improves the reliability and safety of the training plan, ensures that patients receive training at the optimal time, and avoids recovery risks caused by inappropriate start times. The aforementioned multiple verification mechanism can continuously capture dynamic changes during patient monitoring, improve the accuracy and adaptability of personalized recovery plans, and thus enhance rehabilitation effects. Especially in the postoperative recovery scenario of breast disease patients, it can flexibly respond to the rehabilitation needs of different patients.

[0077] This invention also provides a knowledge graph-driven intelligent decision-making system for breast disease rehabilitation training, used to implement the above-mentioned methods, such as... Figure 2 As shown, the system includes:

[0078] The data acquisition unit is used to collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain a dataset of the patient's current recovery status; the data acquisition unit is used to classify and process the dataset of the patient's current recovery status, extract dynamic perception variable groups, and determine an individualized recovery trend vector.

[0079] The starting point determination unit is used to perform boundary division on the individualized recovery trend vector, adjust the vector weights, and obtain a set of triggering conditions; and to determine a suitable training starting point by matching decision support rules with a pre-established knowledge base.

[0080] The initiation command determination unit is used to activate the triggering mechanism to generate a preliminary plan signal, integrate individual difference data, and determine the final rehabilitation training initiation command when the monitoring requirements are met at the appropriate training initiation point.

[0081] The instruction optimization unit is used to predict potential changes based on the final rehabilitation training initiation instruction, extract trend deviations, and obtain an adjusted instruction sequence.

[0082] The training program generation unit is used to verify the accuracy of the adjusted instruction sequence through feedback loops, iteratively optimize the condition set, and generate accurate rehabilitation training programs.

[0083] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0084] In summary, this invention uses sensor devices to collect multi-dimensional monitoring data from breast cancer patients after surgery in real time, including wound healing progress, pain scores, and upper limb mobility. It extracts individualized recovery trend vectors to reflect the patient's recovery status, ensuring that the patient's rehabilitation data is accurately captured and transformed into a vector usable for decision-making. A Support Vector Machine (SVM) model is used to delimit the boundaries of the recovery trend vector, dynamically adjusting the weights of each monitoring indicator to generate a trigger condition set. This trigger condition set combines individual differences, recovery status, and recovery trends, ensuring that the timing and intensity of training initiation can be precisely adjusted according to the patient's real-time recovery progress. Finally, the trigger condition set is combined with decision support rules in a knowledge base. A knowledge graph query module matches relevant rules in the rehabilitation training knowledge base to determine whether the conditions for initiating training are met. The system identifies suitable training initiation points and generates preliminary program signals, while also integrating individual difference data to further optimize rehabilitation training instructions. A time series analysis model predicts instructions, extracts trend deviations, and adjusts the instruction sequence to ensure the training plan adapts to potential future changes. A feedback loop mechanism verifies the accuracy of the adjusted instruction sequence and iteratively optimizes the trigger condition set, ensuring the final generation of a precise rehabilitation training program. Through the synergy of these technologies, the system addresses the lack of personalization and precision in traditional rehabilitation training. Real-time data monitoring and individualized adjustments allow for flexible responses to patients' recovery status and individual differences, providing customized rehabilitation training programs. Furthermore, multiple feedback optimizations continuously improve the training program, enhancing the effectiveness and safety of rehabilitation training.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A knowledge graph-driven intelligent decision-making method for breast disease rehabilitation training, characterized in that, include: Step S1: Collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain the patient's current recovery status dataset; classify the patient's current recovery status dataset, extract dynamic perception variable groups, and determine individualized recovery trend vectors; Step S2: Divide the individualized recovery trend vector into boundaries, adjust the vector weights, and obtain the trigger condition set; determine the appropriate training start point by matching decision support rules with a pre-established knowledge base. Step S3: If the suitable training initiation point meets the monitoring requirements, the triggering mechanism is activated to generate a preliminary plan signal, integrates individual difference data, and determines the final rehabilitation training initiation instruction; based on the final rehabilitation training initiation instruction, potential changes are predicted, trend deviations are extracted, and an adjusted instruction sequence is obtained; Step S4: Verify the accuracy of the adjusted instruction sequence through a feedback loop, iteratively optimize the condition set, and generate a precise rehabilitation training plan.

2. The method as described in claim 1, characterized in that, In step S1, the patient's current recovery status dataset is obtained, including: Data on wound healing progress, pain scores, and upper limb mobility indicators of breast cancer patients after surgery are collected in real time using sensor devices. These data are preprocessed to remove outliers, generating a standardized multi-dimensional monitoring dataset. The standardized multi-dimensional monitoring dataset is then integrated to form a continuous monitoring stream containing timestamps. Finally, a dataset representing the patient's current recovery status is constructed based on this continuous monitoring stream data.

3. The method as described in claim 2, characterized in that, In step S1, the individualized recovery trend vector is determined, including: The patient's current recovery status dataset is input into a pre-trained random forest algorithm model; the random forest algorithm model is used to classify the patient's current recovery status dataset to identify individual differences; key dynamic perception variable groups are extracted based on the classification results, the dynamic perception variable groups contain core indicators related to the recovery status; feature weights are calculated for the dynamic perception variable groups to generate a vector reflecting individual recovery characteristics; and an individualized recovery trend vector is determined based on the vector.

4. The method as described in claim 1, characterized in that, In step S2, the triggering condition set is obtained, including: The individualized recovery trend vector is input into the support vector machine algorithm model; the support vector machine algorithm model is used to perform boundary division of the individualized recovery trend vector using multi-dimensional monitoring indicators, and the boundary division result is determined. The corresponding weights in the individualized recovery trend vector are adjusted according to the boundary division result and the pain score in the dynamic perception variable group; a trigger condition set is generated according to the adjusted individualized recovery trend vector and the boundary division result.

5. The method as described in claim 4, characterized in that, In step S2, decision support rules are matched with a pre-established knowledge base to determine a suitable training starting point, including: The trigger condition set is input into a pre-established rehabilitation training knowledge base; the knowledge graph query module retrieves decision support rules related to the trigger condition set from the rehabilitation training knowledge base; the values ​​of each indicator in the trigger condition set are analyzed according to the decision support rules to determine whether the logical conditions for training initiation are met; if the logical conditions are met, the current time point is determined to be a suitable training initiation point.

6. The method as described in claim 1, characterized in that, In step S3, the final rehabilitation training initiation instruction is determined, including: The appropriate training initiation point is verified by multi-dimensional monitoring requirements to confirm that all indicators are within the preset range; if the verification is successful, the triggering mechanism is activated to generate a preliminary plan signal; the latest individual difference data is obtained through the dynamic perception update module; the individual difference data is fused with the preliminary plan signal, the signal parameters are adjusted, and the final rehabilitation training initiation command is determined.

7. The method as described in claim 6, characterized in that, In step S3, the adjusted instruction sequence is obtained, including: The final rehabilitation training initiation command is input into a time series analysis model; the potential changes of the final rehabilitation training initiation command in future time periods are predicted using the time series analysis model; trend deviation data related to the prediction results are extracted from the continuous monitoring stream of the patient's current recovery status dataset; the final rehabilitation training initiation command is corrected based on the trend deviation data to generate an adjusted command sequence.

8. The method as described in claim 1, characterized in that, In step S4, a precise rehabilitation training plan is generated, including: The adjusted instruction sequence is input into the decision support feedback loop module; the accuracy of the triggering mechanism of the adjusted instruction sequence is verified by the decision support feedback loop module; if the verification result shows that the deviation exceeds a preset threshold, the triggering condition set is iteratively adjusted; a new instruction sequence is regenerated based on the iteratively adjusted triggering condition set; through multiple feedback loop verifications, the accuracy of the new instruction sequence is ensured to meet the requirements.

9. A knowledge graph-driven intelligent decision-making system for breast disease rehabilitation training, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: The data acquisition unit is used to collect multi-dimensional monitoring data of breast disease patients after surgery through sensor devices to obtain a dataset of the patient's current recovery status; the data acquisition unit is used to classify and process the dataset of the patient's current recovery status, extract dynamic perception variable groups, and determine an individualized recovery trend vector. The starting point determination unit is used to perform boundary division on the individualized recovery trend vector, adjust the vector weights, and obtain a set of triggering conditions; and to determine a suitable training starting point by matching decision support rules with a pre-established knowledge base. The initiation command determination unit is used to activate the triggering mechanism to generate a preliminary plan signal, integrate individual difference data, and determine the final rehabilitation training initiation command when the monitoring requirements are met at the appropriate training initiation point. The instruction optimization unit is used to predict potential changes based on the final rehabilitation training initiation instruction, extract trend deviations, and obtain an adjusted instruction sequence. The training program generation unit is used to verify the accuracy of the adjusted instruction sequence through feedback loops, iteratively optimize the condition set, and generate accurate rehabilitation training programs.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-8.

Images