A method for recommending cancer pain care for cancer patients

Abstract

The present invention discloses a method for recommending a cancer pain care plan for tumor patients, which belongs to the field of medical care technology, and includes the following steps: obtaining video monitoring data, audio monitoring data, heart rate monitoring data, and blood pressure monitoring data during the cancer pain observation period of the tumor patient; extracting and obtaining cancer pain recognition data; using a trained cancer pain recognition machine learning model set to perform cancer pain recognition and obtain cancer pain recognition results; obtaining the self-assessed cancer pain site and self-assessed cancer pain score obtained by the patient after self-assessment through cancer pain test self-assessment questions; obtaining the cancer pain score, and combining the self-assessed cancer pain site to obtain the cancer pain recognition evaluation result of the tumor patient; constructing a knowledge graph of a cancer pain care recommendation plan for tumor patients; according to the cancer pain recognition evaluation result of the tumor patient, based on the cancer pain care recommendation plan knowledge graph for the tumor patient, a recommended cancer pain care plan for the tumor patient is obtained. The present invention solves the problem that the cancer pain assessment accuracy of the tumor patient is low and the recommended care plan is not targeted enough.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical care, and in particular relates to a method for recommending a cancer pain care plan for tumor patients. Background Art

[0002] With the continuous advancement of medical technology, the survival time of cancer patients has been greatly improved. Prolonging the life of patients is no longer the only goal of cancer treatment. More attention should be paid to the long-term benefits of patients and improving their quality of life. Patients at different stages of cancer and cured patients all have cancer pain, and cancer pain has become a problem covering the entire course of cancer patients. However, the current management of cancer pain is not optimistic, and the situation of delayed medical treatment is serious. The main reasons include misunderstandings about cancer pain, inconvenience in medical treatment and high treatment costs. If cancer pain is not effectively controlled, it will not only aggravate the patient's condition, accelerate disease progression, deteriorate physiological functions, and force interruption of anti-tumor treatment, but also affect the patient's mental and psychological functions, social activities and quality of life, bringing huge psychological and economic burdens to patients, families and society.

[0003] The current status of cancer pain management focuses on immediate control of cancer pain, with less attention paid to long-term cancer pain management, psychological and social needs of patients. There are also problems such as low assessment accuracy, poor real-time performance, and lack of homogeneity in care plans. The current recommendations for cancer pain care plans for cancer patients are mainly based on a single-dimensional plan recommendation for existing cancer pain conditions, making it difficult to provide targeted care plans based on the patient's condition. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for recommending cancer pain care plans for tumor patients, which solves the problems of low accuracy in cancer pain assessment for tumor patients and insufficient specificity of recommended care plans.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] The present invention provides a method for recommending a cancer pain care plan for tumor patients, comprising the following steps:

[0007] S1. Obtaining video monitoring data, audio monitoring data, heart rate monitoring data and blood pressure monitoring data of tumor patients during cancer pain observation;

[0008] S2. Extracting cancer pain recognition data based on the video monitoring data, audio monitoring data, heart rate monitoring data, and blood pressure monitoring data;

[0009] S3, performing cancer pain recognition based on the cancer pain recognition data using a set of trained cancer pain recognition machine learning models to obtain a cancer pain recognition result;

[0010] S4, obtaining the self-assessed cancer pain location and self-assessed cancer pain score obtained by the patient through self-assessment questions of the cancer pain test;

[0011] S5, performing weighted summation of the cancer pain recognition result and the self-assessed cancer pain score to obtain the cancer pain score, and combining the self-assessed cancer pain location to obtain the cancer pain recognition assessment result of the tumor patient;

[0012] S6. Construct a knowledge graph of recommended cancer pain care plans for cancer patients;

[0013] S7. According to the cancer pain identification and assessment results of tumor patients, a recommended cancer pain care plan for tumor patients is obtained based on the knowledge graph of the recommended cancer pain care plan for tumor patients.

[0014] The beneficial effects of the present invention are as follows: the present invention provides a method for recommending cancer pain care plans for tumor patients, which obtains the video monitoring data, audio monitoring data, heart rate monitoring data and blood pressure monitoring data of tumor patients, and extracts the cancer pain manifestation data from the physical signs and physiological data of the cancer pain experienced by the tumor patients, as the data to be recognized by the cancer pain, thereby accurately learning the relationship between the cancer pain manifestation characteristics of the tumor patients and the cancer pain degree score based on the comprehensive cancer pain manifestation data of the tumor patients, thereby accurately predicting the corresponding cancer pain degree; the present invention realizes cancer pain recognition according to different cancer pain manifestation data of tumor patients through the trained cancer pain recognition machine learning model, and obtains a comprehensive cancer pain recognition result; the present invention The invention obtains the identification data of the cancer pain self-assessed by the tumor patients, namely the physical and psychological cancer pain they have experienced. The invention obtains the identification data of the cancer pain self-assessed by the tumor patients, namely the physical and psychological cancer pain they have experienced. The invention obtains the cancer pain identification and evaluation results of the tumor patients, and constructs a knowledge graph of the recommended cancer pain care plans for the tumor patients. The invention realizes matching based on the characteristics of the pain degree and pain range in the cancer pain identification and evaluation results of the tumor patients, and obtains the corresponding recommended cancer pain care plans for the tumor patients. The invention can accurately assess the cancer pain of the tumor patients and recommend the cancer pain care plans in a targeted manner, so that the tumor patients can benefit in the long term and improve their quality of life.

[0015] Furthermore, the S2 comprises the following steps:

[0016] S21, intercepting a number of facial images and a number of full-body images of cancer patients from the video monitoring data;

[0017] S22, extracting sound spectrum data of tumor patients from the audio monitoring data;

[0018] S23, using the tumor patient's facial image, the tumor patient's full body image, the tumor patient's voice spectrum data, and the heart rate and blood pressure monitoring signals as the cancer pain recognition data.

[0019] The beneficial effect of adopting the above further scheme is: the present invention extracts the facial image, full-body image and sound spectrum data of the tumor patient from the video monitoring data and the audio monitoring data, and combines the heart rate and blood pressure monitoring signals to form the cancer pain recognition data, so as to comprehensively consider the manifestation characteristics of the cancer pain experienced by the tumor patient and provide a basis for accurately evaluating the cancer pain of the tumor patient.

[0020] Furthermore, the set of cancer pain recognition machine learning models trained in S3 includes a trained expression cancer pain recognition network model, a trained action cancer pain recognition network model, an audio cancer pain recognition neural network model, and a heart rate and blood pressure cancer pain recognition neural network model;

[0021] The cancer pain recognition result is the average value of the cancer pain degree score obtained by the expression cancer pain recognition network model identifying the facial image of the tumor patient, the cancer pain degree score obtained by the action cancer pain recognition network model identifying the whole body image of the tumor patient, the cancer pain degree score obtained by the audio cancer pain recognition neural network model identifying the sound spectrum data of the tumor patient, and the cancer pain degree score obtained by the heart rate and blood pressure cancer pain recognition neural network model identifying the heart rate and blood pressure monitoring signals.

[0022] The beneficial effects of adopting the above further scheme are: the present invention provides a set of cancer pain recognition machine learning models to respectively learn the relationship between cancer pain manifestations and cancer pain severity, which can be better generalized to new data, has better tolerance to noise and outliers, and can reduce the sensitivity of a single model to noisy data, thereby achieving accurate assessment of cancer pain severity and obtaining high-quality cancer pain recognition results.

[0023] Further, the S6 comprises the following steps:

[0024] S61. Obtain the data source of structured cancer pain care recommendation plan and the data source of unstructured cancer pain care recommendation plan;

[0025] S62, respectively cleaning the structured cancer pain care solution data source and the unstructured cancer pain care solution data source to remove duplicate, erroneous and inconsistent data, and obtaining a stand-by structured cancer pain care recommendation solution data source and a stand-by unstructured cancer pain care recommendation solution data source;

[0026] S63, performing entity alignment and attribute filling on the data source of the structured cancer pain care recommendation plan to be used according to the cancer pain patient information database and the care plan database, to obtain knowledge fusion data;

[0027] S64, performing entity extraction, relationship extraction, and attribute extraction on the unstructured cancer pain care solution data source to be used to obtain knowledge extraction data;

[0028] S65. Perform coreference resolution and entity disambiguation on the knowledge fusion data and the knowledge extraction data, and build an initial database;

[0029] S66. Convert the entities and relationships in the initial database into entity triplets in the Neo4j database to complete the construction of the knowledge graph of recommended cancer pain care plans for cancer patients.

[0030] The beneficial effect of adopting the above-mentioned further scheme is as follows: the present invention integrates various types of cancer pain care recommendation plan data sources, including structured and unstructured data sources, which helps to more comprehensively understand the relationship between the cancer pain degree and cancer pain range characteristics of cancer pain patients and the care plan, so that the constructed cancer pain care recommendation plan knowledge graph can accurately provide the cancer pain care plan required by cancer patients.

[0031] Further, the S62 comprises the following steps:

[0032] S621, identifying and locating entities in the unstructured cancer pain care plan data source to be used, and classifying the entities into predefined entity categories as entity extraction data, wherein the predefined entity categories include cancer pain care plan, cancer pain score, and cancer pain location;

[0033] S622. A statistical learning-based method uses a deep learning algorithm to learn features of relationships between entities, and uses the learned features to extract relationships between entities from a data source of unstructured cancer pain care solutions to be used as relationship extraction data;

[0034] S623, extracting the attributes of each entity from the unstructured cancer pain care program data source to be used using an attribute extraction deep learning network as attribute extraction data;

[0035] S624. Use the entity extraction data, relationship extraction data and attribute extraction data as knowledge extraction data.

[0036] Further, the S7 comprises the following steps:

[0037] S71. Define the bipartite diagram of cancer patients and cancer pain care plan:

[0038] ,

[0039] Among them, G represents the bipartite graph of cancer patients and cancer pain care plans, p represents cancer patients, represents the interactive relationship between tumor patients and cancer pain care programs, i represents the cancer pain care program, represents belonging to, P represents the set of tumor patients, I represents the set of cancer pain care programs, represents constraints, where the interaction relationship is the input cancer pain recognition evaluation results of tumor patients;

[0040] S72, adding the tumor patient set and the interaction relationship to the cancer pain care recommendation plan knowledge graph for tumor patients to form an attention network of the cancer pain care knowledge graph;

[0041] S73. Use the knowledge representation model to learn the vector representations of nodes and edges in the attention network of the cancer pain care knowledge graph, and obtain the first-order embedding vector representation of tumor patients and cancer pain care plans;

[0042] S74, according to the first-order embedding vector representation of the tumor patients and the cancer pain care plan, using the attention embedding propagation module to update the node representation in the attention network of the cancer pain care knowledge graph, and obtain the aggregated attention vector representation of several tumor patients and cancer pain care plans, wherein the attention embedding propagation module includes several propagation layers, each propagation layer recursively propagates the embedding vector of the neighboring node to update the node representation;

[0043] S75, concatenating the first-order embedding vector representation of each tumor patient and the cancer pain care plan, the representation of each propagation layer output, and the aggregated attention vector representation to obtain a matching score between the tumor patient and the cancer pain care plan;

[0044] S76. The cancer pain care plan corresponding to the tumor patient with the highest matching score is used as the recommended cancer pain care plan for the tumor patient.

[0045] The beneficial effects of adopting the above-mentioned further scheme are as follows: the present invention introduces a set of tumor patients and interactive relationships into the knowledge graph, forming an attention network of the cancer pain care knowledge graph, which can better capture the complex relationship between tumor patients and cancer pain care plans, so as to improve the accuracy of matching recommended care plans, and make the recommended care plans more targeted by embedding the attention propagation module and splicing relevant information vectors.

[0046] Further, the S74 includes the following steps:

[0047] S741. Construct a header entity triple set:

[0048] ,

[0049] in, Represents a set of head entity triples;

[0050] S742. According to the head entity triple set, the amount of information of the tail entity vector of the head entity triple is obtained:

[0051] ,

[0052] ,

[0053] in, The amount of information in the tail entity vector of the head entity triplet, represents the attention weight, represents the tail entity vector, represents the head entity vector, Represents the relationship vector between entities;

[0054] S743. According to the first-order embedding vector representation of tumor patients and cancer pain care plans, the representations of tumor patients and cancer pain care plans output by all propagation layers are aggregated based on the information amount of the tail entity vector of the head entity triplet to obtain the aggregated attention vector representation of tumor patients and cancer pain care plans.

[0055] The beneficial effect of adopting the above further scheme is: the attention degree weight set by the present invention takes into account the differences between nodes and the offset of vectors, so that the attention weight between nodes that are closer in the relationship space is higher, and more information is transmitted between entities that are actually more similar, effectively reducing the loss of information contained in the entity.

[0056] Furthermore, the calculation expression of the matching score between the tumor patient and the cancer pain care plan in S75 is as follows:

[0057] ,

[0058] ,

[0059] ,

[0060] in, represents the matching score between cancer patients and cancer pain care programs, represents the target tumor patient after splicing, T represents transposition, The target cancer pain care program after splicing is expressed as follows: Represents the first-order embedding vector representation of tumor patients, represents the representation of the tumor patient output by the first propagation layer, || represents concatenation, Represents the aggregated attention vector representation of tumor patients, The first-order embedding vector representation of the cancer pain care plan, represents the cancer pain care plan output by the first propagation layer, The aggregated attention vector representation of the cancer pain care plan, L represents the total number of propagation layers.

[0061] The beneficial effect of adopting the above further scheme is: the present invention fully considers the different orders of embedded vector representations of tumor patients and cancer pain care plans, thereby ensuring the accuracy of the matching score and providing a guarantee for accurately recommending targeted care plans for tumor patients.

[0062] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0064] Figure 1 This is a flowchart of the steps of a method for recommending a cancer pain care plan for tumor patients in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with 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 the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0066] like Figure 1 As shown, in one embodiment of the present invention, the present invention provides a method for recommending a cancer pain care plan for tumor patients, comprising the following steps:

[0067] S1. Obtaining video monitoring data, audio monitoring data, heart rate monitoring data and blood pressure monitoring data of tumor patients during cancer pain observation;

[0068] The tumor patients in this embodiment have the following characteristics: confirmed malignant tumor, NRS score greater than or equal to 4 points, complete information, no history of mental illness, and normal communication ability, intelligence and pain perception.

[0069] S2. Extracting cancer pain recognition data based on the video monitoring data, audio monitoring data, heart rate monitoring data, and blood pressure monitoring data;

[0070] The S2 comprises the following steps:

[0071] S21, intercepting a number of facial images and a number of full-body images of cancer patients from the video monitoring data;

[0072] S22, extracting sound spectrum data of tumor patients from the audio monitoring data;

[0073] S23, using the tumor patient's facial image, the tumor patient's full body image, the tumor patient's voice spectrum data, and the heart rate and blood pressure monitoring signals as the cancer pain recognition data.

[0074] S3, performing cancer pain recognition based on the cancer pain recognition data using a set of trained cancer pain recognition machine learning models to obtain a cancer pain recognition result;

[0075] The set of machine learning models for cancer pain recognition trained in S3 includes a trained expression cancer pain recognition network model, a trained action cancer pain recognition network model, an audio cancer pain recognition neural network model, and a heart rate and blood pressure cancer pain recognition neural network model;

[0076] The cancer pain recognition result is the average value of the cancer pain degree score obtained by the expression cancer pain recognition network model identifying the facial image of the tumor patient, the cancer pain degree score obtained by the action cancer pain recognition network model identifying the whole body image of the tumor patient, the cancer pain degree score obtained by the audio cancer pain recognition neural network model identifying the sound spectrum data of the tumor patient, and the cancer pain degree score obtained by the heart rate and blood pressure cancer pain recognition neural network model identifying the heart rate and blood pressure monitoring signals.

[0077] The method for constructing and training the facial cancer pain recognition network model comprises the following steps:

[0078] Obtain cancer pain facial images of several different tumor patients experiencing cancer pain, and set a cancer pain degree score label for each cancer pain facial image, that is, the true category of the cancer pain facial image;

[0079] An image feature extraction algorithm is used to extract a cancer pain expression feature vector from a cancer pain face image to form a cancer pain expression feature data set. In this embodiment, a directional gradient histogram (HOG) feature extraction algorithm is used as an image feature extraction algorithm. The shape and texture information in the image is captured by calculating the gradient direction histogram of a local area of ​​the image, and the local histogram is contrast normalized, which can effectively improve the robustness to illumination changes and shadows in the cancer pain face image.

[0080] Select the kernel function of the SVM model, set the penalty parameters and kernel function parameters;

[0081] The SVM model is trained according to the cancer pain expression feature dataset, and the optimal parameter combination is found through cross-validation to find multiple separation surfaces, thereby obtaining a trained expression cancer pain recognition network model; the trained expression cancer pain recognition network model can accurately identify the cancer pain degree score corresponding to the input tumor patient's facial image.

[0082] The method for constructing and training the action cancer pain recognition network model comprises the following steps:

[0083] Obtain whole-body cancer pain images of several different tumor patients during the cancer pain process, and set a cancer pain degree score label for each whole-body cancer pain image, that is, the true category of the whole-body cancer pain image;

[0084] The image feature extraction algorithm is used to extract the cancer pain action feature vectors in the whole-body cancer pain images to form a cancer pain action feature dataset;

[0085] Select the kernel function of the SVM model, set the penalty parameters and kernel function parameters;

[0086] The SVM model is trained according to the cancer pain motion feature dataset, and the optimal parameter combination is found through cross-validation to find multiple separation surfaces, thereby obtaining a trained motion cancer pain recognition network model. The trained motion cancer pain recognition network model can accurately identify the cancer pain degree score corresponding to the input whole-body image of the tumor patient.

[0087] The method for constructing and training the audio cancer pain recognition neural network model comprises the following steps:

[0088] Obtain audio recordings of cancer pain experienced by several different tumor patients, convert each audio recording into a Mel spectrum, and set a cancer pain degree score label for each Mel spectrum, that is, the true category of the audio recording;

[0089] The audio feature convolutional neural network is used to extract the cancer pain emotion feature vector of the Mel spectrum to form a cancer pain sound feature dataset;

[0090] Using the cancer pain sound feature data set, the audio cancer pain recognition neural network is trained so that the audio cancer pain recognition neural network can learn the mapping relationship between the cancer pain emotional feature vector and the cancer pain degree score label, and obtain a trained audio cancer pain recognition neural network; the trained audio cancer pain recognition neural network can accurately identify the cancer pain degree score corresponding to the input tumor patient sound spectrum data. The audio cancer pain recognition neural network uses a convolutional neural network.

[0091] The method for constructing and training the heart rate, blood pressure, cancer pain recognition model network model comprises the following steps:

[0092] Obtain continuous heart rate and blood pressure signals of several different tumor patients during the process of experiencing cancer pain, and set cancer pain degree score labels for each heart rate and blood pressure signal, that is, the true category of the heart rate and blood pressure signals;

[0093] Filtering the heart rate and blood pressure signals, and extracting cancer pain heart rate and blood pressure feature vectors from the filtered heart rate and blood pressure signals to form a cancer pain heart rate and blood pressure feature data set;

[0094] A heart rate and blood pressure cancer pain recognition neural network model composed of a convolutional neural network and a recurrent neural network is trained using the cancer pain heart rate and blood pressure feature data set; the convolutional neural network is used to extract the spatial features of the cancer pain heart rate and blood pressure feature vectors in the cancer pain heart rate and blood pressure feature data set, and then the recurrent neural network is used to process the heart rate and blood pressure signals in the time series to obtain the time characteristics of the signals, and the recognition and classification of the heart rate and blood pressure signals are realized through multi-classification functions; the trained heart rate and blood pressure cancer pain recognition neural network model can accurately identify the cancer pain degree score corresponding to the input heart rate and blood pressure monitoring signal.

[0095] S4. Obtain the self-assessed cancer pain location and self-assessed cancer pain score obtained by the patient through self-assessment questions of the cancer pain test; the cancer pain test self-assessment questions include physical cancer pain scoring questions, psychological cancer pain scoring questions and cancer pain location selection questions.

[0096] S5. Perform weighted summation on the cancer pain recognition result and the self-assessed cancer pain score to obtain the cancer pain score, and combine the self-assessed cancer pain location to obtain the cancer pain recognition assessment result of the tumor patient; when performing weighted summation on the cancer pain recognition result and the self-assessed cancer pain score, the weighted summation is performed based on a preset first weighting coefficient and a second weighting coefficient.

[0097] The cancer pain score includes 1-3 points for mild cancer pain, 4-6 points for moderate cancer pain, and 7-10 points for severe cancer pain;

[0098] The corresponding characteristics of mild cancer pain are tolerable, normal life and normal sleep; the corresponding characteristics of moderate cancer pain are obvious cancer pain, unbearable pain, need for analgesia, and disturbed sleep; the corresponding characteristics of severe cancer pain are intense cancer pain, unbearable pain, need for analgesia, severe disturbance of sleep, accompanied by autonomic nervous system disorder or passive body position.

[0099] The cancer pain sites include systemic cancer pain, local cancer pain and cancer pain after surgery or radiotherapy; local cancer pain includes bone pain, headache or cancer pain in specific parts of the body;

[0100] The corresponding characteristics of systemic cancer pain are extensive tumor metastasis, side effects of treatment, etc.; the corresponding characteristics of local cancer pain are tumor compression, invasion of surrounding tissues or nerves, etc.; the corresponding characteristics of cancer pain after surgery or radiotherapy are tissue damage, nerve damage, etc.

[0101] S6. Construct a knowledge graph of recommended cancer pain care plans for cancer patients;

[0102] The S6 comprises the following steps:

[0103] S61. Obtain the data source of structured cancer pain care recommendation plan and the data source of unstructured cancer pain care recommendation plan;

[0104] In this embodiment, based on medical books and pathological data, the cancer pain score, self-assessed cancer pain location and personalized cancer pain care plan of each patient are obtained one by one, organized into structured data, and stored in a table in a relational database to obtain a structured cancer pain care recommendation plan data source; a cancer pain care plan questionnaire survey is conducted on cancer pain care nursing staff to obtain the cancer pain characteristics and cancer pain care plans of different cancer pain patients, and the cancer pain characteristics and cancer pain care plans that are logically consistent are used as the cancer pain care recommendation plan questionnaire survey results to form an unstructured cancer pain care plan data source.

[0105] S62, respectively cleaning the structured cancer pain care solution data source and the unstructured cancer pain care solution data source to remove duplicate, erroneous and inconsistent data, and obtaining a stand-by structured cancer pain care recommendation solution data source and a stand-by unstructured cancer pain care recommendation solution data source;

[0106] The S62 comprises the following steps:

[0107] S621, identifying and locating entities in the unstructured cancer pain care plan data source to be used, and classifying the entities into predefined entity categories as entity extraction data, wherein the predefined entity categories include cancer pain care plan, cancer pain score, and cancer pain location;

[0108] S622. A statistical learning-based method uses a deep learning algorithm to learn features of relationships between entities, and uses the learned features to extract relationships between entities from a data source of unstructured cancer pain care solutions to be used as relationship extraction data;

[0109] S623. Utilize the attribute extraction deep learning network to extract the attributes of each entity from the unstructured cancer pain care plan data source to be used as the attribute extraction data; in this embodiment, the deep learning network is trained and predicted based on the cancer pain patient care plan, and the trained deep learning network is used as the attribute extraction deep learning network to realize the attribute extraction task.

[0110] S624. Use the entity extraction data, relationship extraction data and attribute extraction data as knowledge extraction data.

[0111] S63, performing entity alignment and attribute filling on the data source of the structured cancer pain care recommendation plan to be used according to the cancer pain patient information database and the care plan database, to obtain knowledge fusion data;

[0112] S64, performing entity extraction, relationship extraction, and attribute extraction on the unstructured cancer pain care solution data source to be used to obtain knowledge extraction data;

[0113] S65. Perform coreference resolution and entity disambiguation on the knowledge fusion data and the knowledge extraction data, and build an initial database;

[0114] S66. Convert the entities and relationships in the initial database into entity triplets in the Neo4j database to complete the construction of the knowledge graph of recommended cancer pain care plans for cancer patients.

[0115] S7. According to the cancer pain identification and assessment results of tumor patients, a recommended cancer pain care plan for tumor patients is obtained based on the knowledge graph of the recommended cancer pain care plan for tumor patients.

[0116] The S7 comprises the following steps:

[0117] S71. Define the bipartite diagram of cancer patients and cancer pain care plan:

[0118] ,

[0119] Among them, G represents the bipartite graph of cancer patients and cancer pain care plans, p represents cancer patients, represents the interactive relationship between tumor patients and cancer pain care programs, i represents the cancer pain care program, represents belonging to, P represents the set of tumor patients, I represents the set of cancer pain care programs, Represents a constraint, wherein the interactive relationship is inputting the cancer pain recognition and evaluation results of the tumor patient; in this embodiment, by inputting the cancer pain recognition and evaluation results of the tumor patient, the cancer pain degree and cancer pain range confirmed by the tumor patient can be used to accurately recommend a cancer pain care plan for the tumor patient.

[0120] S72, adding the tumor patient set and the interaction relationship to the cancer pain care recommendation plan knowledge graph for tumor patients to form an attention network of the cancer pain care knowledge graph;

[0121] S73. Use the knowledge representation model to learn the vector representations of nodes and edges in the attention network of the cancer pain care knowledge graph, and obtain the first-order embedding vector representation of tumor patients and cancer pain care plans;

[0122] The calculation expression of the projection score of the knowledge identification model is as follows:

[0123] ,

[0124] , ,

[0125] in, represents the projection score of the knowledge representation model, h represents the head entity, t represents the tail entity, represents the mapped head entity vector, r represents the relationship between entities, represents the tail entity vector after mapping, represents the square operation of the two-norm, Represents the projection matrix from the entity space to the relational space; the knowledge representation model provided by this scheme can effectively handle many-to-many relationships by mapping entities to different relational spaces, and is highly flexible and easy to train; the rationality of the triple can be measured by the projection score of the knowledge identification model, and the rationality of the triple is evaluated by calculating the distance between the sum of the head entity vector and the relation vector after projection to the relational space and the tail entity vector, where the lower the score, the more reasonable the triple; the higher the score, the more unreasonable the triple.

[0126] S74. Based on the first-order embedding vector representation of tumor patients and cancer pain care plans, the attention embedding propagation module is used to update the node representation in the attention network of the cancer pain care knowledge graph to obtain the aggregated attention vector representation of several tumor patients and cancer pain care plans, wherein the attention embedding propagation module includes several propagation layers, each propagation layer recursively propagates the embedding vector of the neighboring node to update the node representation; in this scheme, the attention embedding propagation module determines the influence of each node on its neighboring node through the similarity between adjacent nodes, and each node updates its embedding vector representation according to the embedding vector of its neighboring node and the corresponding attention score weight, thereby establishing the association relationship between the nodes; the representations output by different propagation layers emphasize the connection information of different orders, which can effectively enrich the embedding vector representation of tumor patients and cancer pain care plans;

[0127] The S74 comprises the following steps:

[0128] S741. Construct a header entity triple set:

[0129] ,

[0130] in, Represents a set of head entity triples;

[0131] S742. According to the head entity triple set, the amount of information of the tail entity vector of the head entity triple is obtained:

[0132] ,

[0133] ,

[0134] in, The amount of information in the tail entity vector of the head entity triplet, represents the attention weight, represents the tail entity vector, represents the head entity vector, Represents the relationship vector between entities; in this scheme, the attention degree weight can represent the amount of information transferred from the tail entity vector to the head entity vector based on the relationship vector between entities.

[0135] S743. According to the first-order embedding vector representation of tumor patients and cancer pain care plans, the representations of tumor patients and cancer pain care plans output by all propagation layers are aggregated based on the information amount of the tail entity vector of the head entity triplet to obtain the aggregated attention vector representation of tumor patients and cancer pain care plans.

[0136] In this scheme, a dual interactive aggregator is used to aggregate the representations of tumor patients and cancer pain care solutions output by each propagation layer to obtain an aggregated attention vector representation of tumor patients and cancer pain care solutions;

[0137] The calculation expression of the dual interaction aggregator is as follows:

[0138] ,

[0139] in, represents a dual interaction aggregator, represents the second activation function, represents the first trainable weight matrix, represents the second trainable weight matrix, Represents the vector element product operation. This scheme aggregates tumor patients and cancer pain care schemes of all propagation layers, so that the propagated information can pay more attention to the correlation between the information amount of the head entity vector and the tail entity vector of the head entity triple, thereby transmitting more information between entities.

[0140] S75, concatenating the first-order embedding vector representation of each tumor patient and the cancer pain care plan, the representation of each propagation layer output, and the aggregated attention vector representation to obtain a matching score between the tumor patient and the cancer pain care plan;

[0141] The calculation expression of the matching score between the tumor patient and the cancer pain care plan in S75 is as follows:

[0142] ,

[0143] ,

[0144] ,

[0145] in, represents the matching score between cancer patients and cancer pain care programs, represents the target tumor patient after splicing, T represents transposition, The target cancer pain care program after splicing is expressed as follows: Represents the first-order embedding vector representation of tumor patients, represents the representation of the tumor patient output by the first propagation layer, || represents concatenation, Represents the aggregated attention vector representation of tumor patients, The first-order embedding vector representation of the cancer pain care plan, represents the cancer pain care plan output by the first propagation layer, The aggregated attention vector representation of the cancer pain care plan, L represents the total number of propagation layers.

[0146] S76. The cancer pain care plan corresponding to the tumor patient with the highest matching score is used as the recommended cancer pain care plan for the tumor patient.

[0147] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for recommending a cancer pain care plan for tumor patients, characterized in that: The steps include: S1. Obtaining video monitoring data, audio monitoring data, heart rate monitoring data and blood pressure monitoring data of tumor patients during cancer pain observation; S2. Extracting cancer pain recognition data based on the video monitoring data, audio monitoring data, heart rate monitoring data, and blood pressure monitoring data; S3, performing cancer pain recognition based on the cancer pain recognition data using a set of trained cancer pain recognition machine learning models to obtain a cancer pain recognition result; S4, obtaining the self-assessed cancer pain location and self-assessed cancer pain score obtained by the patient through self-assessment questions of the cancer pain test; S5, performing weighted summation of the cancer pain recognition result and the self-assessed cancer pain score to obtain the cancer pain score, and combining the self-assessed cancer pain location to obtain the cancer pain recognition assessment result of the tumor patient; S6. Construct a knowledge graph of recommended cancer pain care plans for cancer patients; S7. According to the cancer pain identification and assessment results of tumor patients, based on the knowledge graph of the recommended cancer pain care plan for tumor patients, a recommended cancer pain care plan for tumor patients is obtained; The S7 comprises the following steps: S71. Define the bipartite diagram of cancer patients and cancer pain care plan: ， in, G represents a bipartite graph of cancer patients and cancer pain care plans, p Indicates that cancer patients Represents the interactive relationship between cancer patients and cancer pain care programs, i Cancer pain care program, Indicates belonging to, P represents the set of tumor patients, I Represents a collection of cancer pain care solutions. represents constraints, where the interaction relationship is the input cancer pain recognition evaluation results of tumor patients; S72, adding the tumor patient set and the interaction relationship to the cancer pain care recommendation plan knowledge graph for tumor patients to form an attention network of the cancer pain care knowledge graph; S73. Use the knowledge representation model to learn the vector representations of nodes and edges in the attention network of the cancer pain care knowledge graph, and obtain the first-order embedding vector representation of tumor patients and cancer pain care plans; S74, according to the first-order embedding vector representation of the tumor patients and the cancer pain care plan, using the attention embedding propagation module to update the node representation in the attention network of the cancer pain care knowledge graph, and obtain the aggregated attention vector representation of several tumor patients and cancer pain care plans, wherein the attention embedding propagation module includes several propagation layers, each propagation layer recursively propagates the embedding vector of the neighboring node to update the node representation; S75, concatenating the first-order embedding vector representation of each tumor patient and the cancer pain care plan, the representation of each propagation layer output, and the aggregated attention vector representation to obtain a matching score between the tumor patient and the cancer pain care plan; S76. The cancer pain care plan corresponding to the tumor patient with the highest matching score is used as the recommended cancer pain care plan for the tumor patient.

2. The method for recommending cancer pain care plans for tumor patients according to claim 1, characterized in that: The S2 comprises the following steps: S21, intercepting a number of facial images and a number of full-body images of cancer patients from the video monitoring data; S22, extracting sound spectrum data of tumor patients from the audio monitoring data; S23, using the tumor patient's facial image, the tumor patient's full body image, the tumor patient's voice spectrum data, and the heart rate and blood pressure monitoring signals as the cancer pain recognition data.

3. The method for recommending cancer pain care plans for tumor patients according to claim 2, characterized in that: The set of machine learning models for cancer pain recognition trained in S3 includes a trained expression cancer pain recognition network model, a trained action cancer pain recognition network model, an audio cancer pain recognition neural network model, and a heart rate and blood pressure cancer pain recognition neural network model; The cancer pain recognition result is the average value of the cancer pain degree score obtained by the expression cancer pain recognition network model identifying the facial image of the tumor patient, the cancer pain degree score obtained by the action cancer pain recognition network model identifying the whole body image of the tumor patient, the cancer pain degree score obtained by the audio cancer pain recognition neural network model identifying the sound spectrum data of the tumor patient, and the cancer pain degree score obtained by the heart rate and blood pressure cancer pain recognition neural network model identifying the heart rate and blood pressure monitoring signals.

4. The method for recommending cancer pain care plans for tumor patients according to claim 3, characterized in that: The S6 comprises the following steps: S61. Obtain the data source of structured cancer pain care recommendation plan and the data source of unstructured cancer pain care recommendation plan; S62, respectively cleaning the structured cancer pain care solution data source and the unstructured cancer pain care solution data source to remove duplicate, erroneous and inconsistent data, and obtaining a stand-by structured cancer pain care recommendation solution data source and a stand-by unstructured cancer pain care recommendation solution data source; S63, performing entity alignment and attribute filling on the data source of the structured cancer pain care recommendation plan to be used according to the cancer pain patient information database and the care plan database, to obtain knowledge fusion data; S64, performing entity extraction, relationship extraction, and attribute extraction on the unstructured cancer pain care solution data source to be used to obtain knowledge extraction data; S65. Perform coreference resolution and entity disambiguation on the knowledge fusion data and the knowledge extraction data, and build an initial database; S66. Convert the entities and relationships in the initial database into entity triplets in the Neo4j database to complete the construction of the knowledge graph of recommended cancer pain care plans for cancer patients.

5. The method for recommending cancer pain care plans for tumor patients according to claim 4, characterized in that: The S62 comprises the following steps: S621, identifying and locating entities in the unstructured cancer pain care plan data source to be used, and classifying the entities into predefined entity categories as entity extraction data, wherein the predefined entity categories include cancer pain care plan, cancer pain score, and cancer pain location; S622. A statistical learning-based method uses a deep learning algorithm to learn features of relationships between entities, and uses the learned features to extract relationships between entities from a data source of unstructured cancer pain care solutions to be used as relationship extraction data; S623, extracting the attributes of each entity from the unstructured cancer pain care program data source to be used using an attribute extraction deep learning network as attribute extraction data; S624. Use the entity extraction data, relationship extraction data and attribute extraction data as knowledge extraction data.

6. The method for recommending cancer pain care plans for tumor patients according to claim 1, characterized in that: The S74 comprises the following steps: S741. Construct a header entity triple set: ， in, Represents a set of head entity triples; S742. According to the head entity triplet set, the amount of information of the tail entity vector of the head entity triplet is obtained: ， ， in, The amount of information in the tail entity vector of the head entity triplet, represents the attention weight, represents the tail entity vector, represents the head entity vector, Represents the relationship vector between entities; S743. According to the first-order embedding vector representation of tumor patients and cancer pain care plans, the representations of tumor patients and cancer pain care plans output by all propagation layers are aggregated based on the information amount of the tail entity vector of the head entity triplet to obtain the aggregated attention vector representation of tumor patients and cancer pain care plans.

7. The method for recommending cancer pain care plans for tumor patients according to claim 6, characterized in that: The calculation expression of the matching score between the tumor patient and the cancer pain care plan in S75 is as follows: ， ， ， in, represents the matching score between cancer patients and cancer pain care programs, The target tumor patient after splicing indicated that T represents transpose, The target cancer pain care program after splicing is expressed as follows: Represents the first-order embedding vector representation of tumor patients, represents the representation of the tumor patient output by the first propagation layer, || represents concatenation, Represents the aggregated attention vector representation of tumor patients, The first-order embedding vector representation of the cancer pain care plan, represents the cancer pain care plan output by the first propagation layer, The aggregated attention vector representation of the cancer pain care plan, L Indicates the total number of propagation layers.

Images