Postoperative rehabilitation method based on knowledge graph
By constructing a rehabilitation knowledge graph and decision tree model based on knowledge graph, the inaccurate nursing plan caused by relying on experience in the existing technology is solved, and personalized adjustments to postoperative rehabilitation plans are achieved, and the quality of nursing is improved.
Patent Information
- Application Number
- CN202510474079.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-08-01
AI Technical Summary
In the existing postoperative nursing monitoring methods, the patient's rehabilitation plan depends on the experience of medical staff, resulting in insufficient accuracy and inability to accurately match the patient's actual needs, affecting the quality of care.
A knowledge graph-based method is adopted to obtain postoperative rehabilitation data for preprocessing, build a rehabilitation knowledge graph, generate a personalized rehabilitation plan, and dynamically adjust it using the rehabilitation plan decision tree model to ensure that the rehabilitation plan matches the patient's condition.
It has achieved more accurate consideration of patient diversity, and can timely adjust the rehabilitation plan, ensuring the smooth completion of the rehabilitation process, and improving the quality of nursing.
Smart Images
Figure CN120412972A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent medical technology, and in particular to a postoperative rehabilitation method based on a knowledge graph. Background Art
[0002] In the existing postoperative care monitoring methods, the rehabilitation plan for patients is usually formulated based on the current various data of the patients. After the plan is executed, medical staff will judge the rehabilitation period and the current rehabilitation progress of the patients by combining their own experience and the various index data of the patients. However, this judgment method may have insufficient accuracy due to more reference to their own experience, and may misjudge the rehabilitation progress of the patients, resulting in insufficient matching degree between the nursing and rehabilitation plan given when nursing the patients and the physical state of the patients, reducing the nursing quality of the patients. With the gradual rise of "intelligent medicine", intelligent medical solutions are being piloted in some big cities. The pilot hospitals build a regional medical information platform for health records and use the most advanced Internet of Things technology to effectively realize intelligent postoperative nursing. However, in the determination of weights and the selection of each gene in the existing technical solutions, data elements are not linked, and each element is independent, and the obtained prediction model is not comprehensive.
[0003] Therefore, providing a postoperative rehabilitation method based on a knowledge graph to solve the difficulties existing in the prior art is an urgent problem to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides a postoperative rehabilitation method based on a knowledge graph, which can timely adjust the rehabilitation plan and ensure the smooth completion of the patient's rehabilitation process.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] A postoperative rehabilitation method based on a knowledge graph includes the following steps:
[0007] Obtain postoperative rehabilitation data, and preprocess the postoperative rehabilitation data to obtain a preprocessed data set;
[0008] Identify entities in the preprocessed data set, form entity pairs by taking two entities as a group, obtain the sentence vector of the sentence according to the entity pair; extract the fusion feature of the sentence according to the sentence vector, predict the entity relationship of the entity pair according to the fusion feature, and construct a rehabilitation knowledge graph according to the entity pair and the corresponding entity relationship;
[0009] Map the patient data to the rehabilitation knowledge graph, generate the postoperative recovery data of the patient according to the rehabilitation knowledge graph, and generate a rehabilitation plan based on the postoperative recovery data;
[0010] Perform corresponding rehabilitation training according to the rehabilitation plan.
[0011] Optionally, the preprocessing includes: cleaning the postoperative rehabilitation data to remove data with large errors, and then performing text mining on the cleaned postoperative rehabilitation data to obtain a preprocessed data set.
[0012] Optionally, the text mining includes:
[0013] extracting feature data of the text from the cleaned postoperative rehabilitation data, obtaining text feature data including term frequency and text weight;
[0014] extracting keywords from the text feature data to obtain text keywords, and performing entity recognition on the text keywords to obtain text mining data.
[0015] Optionally, the sentence vector includes word-position vectors of multiple individual words, and the word-position vector of an individual word includes the word vector of the individual word and the relative position vector of the individual word and the entity pair;
[0016] The entity pair includes a first entity and a second entity, and the relative position vector of the individual word and the entity pair includes: calculating the relative distances of the individual word with respect to the first entity and the individual word with respect to the second entity respectively, and converting the relative distances into vectors by using a preset position vector matrix.
[0017] Optionally, extracting the representation features of a sentence includes: extracting the representation features of an individual sentence, putting the representation features of a single sentence into individual channels of a convolutional network respectively to generate feature vectors, and generating a fused feature based on the feature vectors.
[0018] Optionally, generating a rehabilitation plan includes:
[0019] generating a first rehabilitation plan according to the patient's postoperative recovery data;
[0020] obtaining historical postoperative rehabilitation stage data and historical rehabilitation plan data, and constructing a rehabilitation plan decision tree according to the historical postoperative rehabilitation stage data and the historical rehabilitation plan data, so as to obtain a rehabilitation plan decision tree model;
[0021] generating a second rehabilitation plan for the postoperative rehabilitation stage data by using the rehabilitation plan decision tree model;
[0022] generating a rehabilitation plan by integrating the first rehabilitation plan and the second rehabilitation plan.
[0023] Optionally, the rehabilitation plan decision tree model includes:
[0024] extracting the feature data of the historical postoperative rehabilitation stage data and performing node selection to obtain node data, and generating feature tree data based on the node data;
[0025] Calculate the association degree of the node data in the feature tree data to obtain node association data, and calculate the association degree by combining the node data in the feature tree data with the historical rehabilitation plan data to obtain rehabilitation plan association data;
[0026] Dynamically adjust and adjust the spatial relationship of the node association data based on the rehabilitation plan association data to obtain a preliminary rehabilitation decision tree;
[0027] Construct a rehabilitation plan decision tree model by using the historical postoperative rehabilitation stage data and the historical rehabilitation plan data in combination with the preliminary rehabilitation decision tree.
[0028] As can be seen from the above technical solutions, compared with the prior art, the present invention provides a postoperative rehabilitation method based on a knowledge graph, which has the following beneficial effects: 1) The present invention mines the features between the relationships of entities in sentences from multiple aspects and angles, and through a fusion means, a knowledge graph can be constructed more completely; 2) By using the rehabilitation management plan decision tree model, the system can generate a more personalized rehabilitation plan and better meet the rehabilitation needs of patients; 3) The present invention can more accurately consider the diversity of patient conditions, can timely adjust the rehabilitation plan, and ensure the smooth completion of the patient's rehabilitation process. Description of the Drawings
[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.
[0030] Figure 1 It is a flowchart of a postoperative rehabilitation method based on a knowledge graph disclosed by the present invention. Detailed Embodiments
[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.
[0032] Refer to Figure 1 As shown, the present invention discloses a postoperative rehabilitation method based on a knowledge graph, including the following steps:
[0033] Obtain postoperative rehabilitation data, and preprocess the postoperative rehabilitation data to obtain a preprocessed data set;
[0034] Identify entities in the preprocessed dataset, form entity pairs by taking entities two by two, and obtain the sentence vector of the sentence according to the entity pairs; extract the fusion features of the sentence according to the sentence vector, predict the entity relationship of the entity pairs according to the fusion features, and construct a rehabilitation knowledge graph according to the entity pairs and the corresponding entity relationships;
[0035] Map the patient data to the rehabilitation knowledge graph, generate the patient's postoperative recovery data according to the rehabilitation knowledge graph, and generate a rehabilitation plan based on the postoperative recovery data;
[0036] Perform corresponding rehabilitation training according to the rehabilitation plan.
[0037] Furthermore, the postoperative rehabilitation data is obtained from the hospital record data, including the data in the initial postoperative rehabilitation stage and the postoperative rehabilitation tracking data.
[0038] Furthermore, the preprocessing includes: cleaning the postoperative rehabilitation data, removing the data with large errors, and then performing text mining on the cleaned postoperative rehabilitation data to obtain the preprocessed dataset.
[0039] Furthermore, the text mining includes:
[0040] Extract features from the cleaned postoperative rehabilitation data to obtain text feature data including text frequency and text weight;
[0041] Extract keywords from the text feature data to obtain text keywords, and perform entity recognition on the text keywords to obtain text mining data.
[0042] Furthermore, there may be multiple pairs of entity pairs, and there is a corresponding entity relationship between each pair of entity pairs for the construction and improvement of the knowledge graph. The entity relationship prediction can be performed for each pair of entity pairs separately. When performing the entity relationship prediction between a certain pair of entity pairs, the entire sentence text to be processed needs to be vectorized.
[0043] Furthermore, the sentence vector includes the word-position vectors of multiple individual words, and the word-position vector of an individual word includes the word vector of the individual word and the relative position vector of the individual word and the entity pair;
[0044] The entity pair includes a first entity and a second entity. The relative position vector of an individual word and the entity pair includes: calculating the relative distances of the individual word relative to the first entity and the individual word relative to the second entity respectively, and converting the relative distances into vectors by using a preset position vector matrix.
[0045] Specifically, the traditional sentence vector model is directly composed of the word vectors of the words contained in the sentence. However, only relying on the word vectors of the words themselves cannot accurately represent the relationships between the words, and it has little effect on the judgment of entity relationships. Therefore, relative position vectors of the words relative to the entity pair are introduced in addition to the word vectors, so as to reflect the degree of association between each individual word and the two entities, making the vector representation of the individual word no longer isolated and better reflecting the connections between the words. Introducing the relative position features between the words into entity relationship prediction will undoubtedly improve the prediction accuracy.
[0046] Furthermore, when a sentence is long, words that are far from a certain entity in the entity pair have a high probability of having nothing to do with the entity relationship to be predicted. Therefore, in the actual calculation process, an effective threshold range can be set for the relative distance, that is, the values of the preset minimum distance and maximum distance are set, and the relative distances outside the threshold range are regarded as invalid distances, reducing the influence of words that have little relationship with the entity in representing the sentence features. If the relative distance is less than the minimum distance, the relative distance is replaced with the minimum distance; if the relative distance is greater than the maximum distance, the relative distance is replaced with the maximum distance.
[0047] Furthermore, since the importance of each feature vector in the fusion features for entity relationship extraction is different, the global average pooling method is adopted to aggregate the global feature information into one layer of aggregated information;
[0048] To limit the complexity of the model and introduce a non-linear matrix into the calculation at the same time, two fully connected networks are used to obtain the screening parameters corresponding to the feature vectors to determine whether the feature vector is a noise feature. In order to make full use of the aggregated information and obtain the global feature dependence distribution, when performing feature screening, a flexible and non-mutually exclusive threshold mechanism needs to be selected to obtain the screening parameters for judging whether the feature vector is noise;
[0049] Judge whether the feature vector is a noise feature according to the screening parameters. If so, mask the noise feature and update the feature vector;
[0050] According to the updated feature vector, update the original representation of the feature, making the feature parameters more targeted.
[0051] Furthermore, the representation features of the sentence extracted include: extracting the representation features of a single sentence, putting the representation features of a single sentence into separate channels of the convolutional network respectively to generate feature vectors, and generating fusion features based on the feature vectors.
[0052] Further, mapping patient data to the rehabilitation knowledge graph includes: preprocessing the data and performing time series analysis. For example, statistical methods such as the sliding window method and the moving average method can be used to perform trend analysis on patient data in time series, identify the focus of users' attention on different keywords at different stages of rehabilitation, and the characteristics of each stage can include the patient's physiological indicators, emotional state, activity level, etc. Machine learning techniques, such as clustering algorithms, are used to divide the time series analysis data into different rehabilitation stages, and different rehabilitation stages are divided according to characteristics such as the patient's activity level, pain assessment, and diet situation and then mapped.
[0053] Further, generating a rehabilitation plan includes:
[0054] Generating a first rehabilitation plan based on the patient's postoperative recovery data;
[0055] Obtaining historical postoperative rehabilitation stage data and historical rehabilitation plan data, and constructing a rehabilitation plan decision tree based on the historical postoperative rehabilitation stage data and the historical rehabilitation plan data, so as to obtain a rehabilitation plan decision tree model;
[0056] Using the rehabilitation plan decision tree model to generate a second rehabilitation plan for the postoperative rehabilitation stage data;
[0057] Integrating the first rehabilitation plan and the second rehabilitation plan to generate a rehabilitation plan.
[0058] Further, performing association rule mining on the patient's postoperative recovery data to obtain association rule data, integrating the retrieved association rule data and the postoperative rehabilitation stage data to obtain the patient's postoperative health data, and using the preset rehabilitation plan template data to generate a first rehabilitation plan for the patient's postoperative health data, so as to obtain the first rehabilitation plan.
[0059] Further, the rehabilitation plan decision tree model includes:
[0060] Extracting the characteristics of the historical postoperative rehabilitation stage data and performing node selection to obtain node data, and generating feature tree data based on the node data;
[0061] Calculating the correlation degree of the node data of the feature tree data to obtain node correlation data, and calculating the correlation degree of the node data of the feature tree data combined with the historical rehabilitation plan data to obtain rehabilitation plan correlation data;
[0062] Dynamically adjusting and spatially adjusting the node correlation data according to the rehabilitation plan correlation data to obtain a preliminary rehabilitation decision tree;
[0063] Using the historical postoperative rehabilitation stage data and the historical rehabilitation plan data combined with the preliminary rehabilitation decision tree to construct a rehabilitation plan decision tree model.
[0064] Specifically, the model input includes the patient's physiological parameters (such as blood pressure, heart rate), rehabilitation effect evaluation (such as improvement in motor ability), symptom description (such as pain sensation), etc. The decision tree model predicts the rehabilitation measures to be taken at different rehabilitation stages by analyzing these features. Specifically, if the model finds that the patient has high blood pressure but good improvement in motor ability, through correlation processing of historical data, it is mapped to generate an increase in exercise volume and adjustment of drug treatment.
[0065] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A postoperative rehabilitation method based on a knowledge graph, characterized in that It includes the following steps: Obtain postoperative rehabilitation data, preprocess the postoperative rehabilitation data, and obtain a preprocessed data set; Identify entities in the preprocessed data set, form entity pairs by taking two entities as a group, obtain the sentence vector of the sentence according to the entity pair; extract the fusion feature of the sentence according to the sentence vector, predict the entity relationship of the entity pair according to the fusion feature, and construct a rehabilitation knowledge graph according to the entity pair and the corresponding entity relationship; Map the patient data to the rehabilitation knowledge graph, generate the patient's postoperative recovery data according to the rehabilitation knowledge graph, and generate a rehabilitation plan based on the postoperative recovery data; Perform corresponding rehabilitation training according to the rehabilitation plan.
2. The postoperative rehabilitation method based on a knowledge graph according to claim 1, wherein The preprocessing includes: performing data cleaning on the postoperative rehabilitation data, removing data with large errors, and then performing text mining on the cleaned postoperative rehabilitation data to obtain a preprocessed data set.
3. The postoperative rehabilitation method based on a knowledge graph according to claim 2, wherein The text mining includes: Extract features from the cleaned postoperative rehabilitation data to obtain text feature data including term frequency and text weight; Extract keywords from the text feature data to obtain text keywords, perform entity recognition on the text keywords, and obtain text mining data.
4. The postoperative rehabilitation method based on a knowledge graph according to claim 1, wherein The sentence vector includes the word-position vectors of multiple individual words, and the word-position vector of an individual word includes the word vector of the individual word and the relative position vector of the individual word and the entity pair; The entity pair includes a first entity and a second entity, and the relative position vector of the individual word and the entity pair includes: respectively calculating the relative distances of the individual word relative to the first entity and the individual word relative to the second entity, and converting the relative distances into vectors by using a preset position vector matrix.
5. The postoperative rehabilitation method based on a knowledge graph according to claim 1, wherein Extracting the representation feature of the sentence includes: extracting the representation feature of a single sentence, putting the representation features of individual sentences into separate channels of a convolutional network respectively to generate feature vectors, and generating a fusion feature based on the feature vectors.
6. The postoperative rehabilitation method based on a knowledge graph according to claim 1, wherein Generating a rehabilitation plan includes: Generating a first rehabilitation plan according to the patient's postoperative recovery data; Obtain historical postoperative rehabilitation stage data and historical rehabilitation plan data, and construct a rehabilitation plan decision tree according to the historical postoperative rehabilitation stage data and the historical rehabilitation plan data, so as to obtain a rehabilitation plan decision tree model; Use the rehabilitation plan decision tree model to generate a second rehabilitation plan for the postoperative rehabilitation stage data; Integrate the first rehabilitation plan and the second rehabilitation plan to generate a rehabilitation plan.
7. The postoperative rehabilitation method based on a knowledge graph according to claim 6, wherein The rehabilitation plan decision tree model includes: Extract the features of the historical postoperative rehabilitation stage data and perform node selection to obtain node data, and generate feature tree data based on the node data; Calculate the correlation degree of the node data of the feature tree data to obtain node correlation data, and calculate the correlation degree by combining the node data of the feature tree data with the historical rehabilitation plan data to obtain rehabilitation plan correlation data; Dynamically adjust and adjust the spatial relationship of the node correlation data based on the rehabilitation plan correlation data to obtain a preliminary rehabilitation decision tree; Construct a rehabilitation plan decision tree model by using the historical postoperative rehabilitation stage data and the historical rehabilitation plan data in combination with the preliminary rehabilitation decision tree.