Electronic medical record information processing method, system, device and medium based on deep learning

Abstract

The present application discloses a method, system, device and medium for processing electronic medical record information based on deep learning, and relates to the field of data processing technology. The method includes: obtaining first entity information in a first electronic medical record through a first sub-model in an electronic medical record information processing model; converting the first entity information into first vector information; identifying the association relationship between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model; generating a knowledge graph based on the association relationship, the knowledge graph is used to characterize the key information in the first electronic medical record; identifying the knowledge graph through the electronic medical record information processing model, and generating recommended information corresponding to the electronic medical record, the recommended information is used to characterize the adaptation scheme corresponding to the user status information. The ability to quickly extract information from electronic medical records greatly saves the time doctors spend viewing electronic medical records and improves the efficiency of electronic medical records.

Description

Technical Field

[0001] The present disclosure generally relates to the field of data processing technology, and more particularly to a method, system, device and medium for processing electronic medical record information based on deep learning. Background Art

[0002] With the continuous development of electronic technology, it has become very common in various fields to analyze and calculate data in their field through model training.

[0003] In the medical field, electronic medical records have long been popularized to ensure the accuracy of patient diagnosis. Based on its easy-to-preserve and non-lost characteristics, electronic medical records can retain all the diagnostic process information for each patient in detail. For any patient, in order to ensure the accuracy of the diagnosis, the doctor may need the support of multiple examination data for each diagnosis. Therefore, for the patient, when he finally faces the doctor and the doctor makes a diagnosis for him, the patient's electronic medical record may already contain a large amount of data, and there are many pages that require the doctor to constantly click on the display terminal (such as a computer) to flip through. The doctor needs to view it for a period of time and extract useful information as quickly as possible to make a diagnosis.

[0004] From this we can see that when faced with electronic medical records with large amounts of data, doctors actually need to spend a long time reviewing and processing them, which will increase the length of diagnosis and treatment. Summary of the invention

[0005] In view of the above-mentioned defects or deficiencies in the relevant technology, it is expected to provide an electronic medical record information processing method, system, device and medium based on deep learning, which can solve the problem that when faced with electronic medical records with large amounts of data, doctors actually need to spend a long time to review and process them, which increases the length of diagnosis and treatment. The system can quickly extract information from electronic medical records, greatly saving the time doctors spend reviewing electronic medical records and greatly improving the efficiency of electronic medical records.

[0006] In a first aspect, a method for processing electronic medical record information based on deep learning is provided, the method comprising: Acquire first entity information in the first electronic medical record through the first sub-model in the electronic medical record information processing model, the first entity information including at least one of the following: user status information, a scheme corresponding to the user status information, the first entity information including: past entity information and current entity information, the current entity information including the user status information in the first electronic medical record that is closest to the current time; Converting the first entity information into first vector information; Identify the association relationship between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model, wherein the second sub-model is a multi-entity parallel graph neural network model; Generate a knowledge graph according to the association relationship, wherein the knowledge graph is used to represent key information in the first electronic medical record; The knowledge graph is identified through the electronic medical record information processing model, and recommendation information corresponding to the electronic medical record is generated, where the recommendation information is used to characterize an adaptation solution corresponding to the user status information.

[0007] In the present application, the RoBERTa model in the electronic medical record information processing model is used to obtain user status information and / or a solution corresponding to the user status information in the first electronic medical record, and at the same time, the first entity information: past entity information and current entity information, the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and the first sub-model is the RoBERTa model; then, the above-mentioned first entity information is converted into first vector information; then, the multi-entity parallel graph neural network model in the electronic medical record information processing model is used to identify the association relationship between different first vector information in the first vector information; then, an indicative graph for characterizing the key information in the first electronic medical record is generated according to the association relationship; then, the above-mentioned knowledge graph is identified through the above-mentioned electronic medical record information processing model, and recommendation information corresponding to the electronic medical record is generated for characterizing the adaptation solution corresponding to the user status information. In this way, by using different sub-models in the electronic medical record information processing model, the first electronic medical record is processed and information is screened, and finally the most critical information is screened out to generate a knowledge graph, and the key information is combined to match the adaptation plan corresponding to the user status information for the doctor's reference, thereby greatly improving the efficiency, accuracy and comprehensiveness of doctors' viewing of information.

[0008] In a second aspect, a deep learning-based electronic medical record information processing system is provided, the system comprising: an acquisition unit, configured to acquire first entity information in a first electronic medical record through a first sub-model in an electronic medical record information processing model, wherein the first entity information includes at least one of the following: user status information and a scheme corresponding to the user status information, wherein the first entity information includes: past entity information and current entity information, wherein the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and wherein the first sub-model is a RoBERTa model; An execution unit, configured to convert the first entity information into first vector information; an identification unit, configured to identify an association relationship between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model, wherein the second sub-model is a multi-entity parallel graph neural network model; A generating unit, configured to generate a knowledge graph according to the association relationship, wherein the knowledge graph is used to represent key information in the first electronic medical record; The generating unit is further used to identify the knowledge graph through the electronic medical record information processing model, and generate recommendation information corresponding to the electronic medical record, wherein the recommendation information is used to characterize the adaptation solution corresponding to the user status information.

[0009] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the first aspect is implemented.

[0010] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the program is executed by a processor, the method described in the first aspect is implemented.

[0011] In a fifth aspect, a computer program product is provided. The computer program product includes instructions, and the instructions are executed by a processor to implement the method described in the first aspect.

[0012] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 A flowchart of an electronic medical record information processing method based on deep learning provided in an embodiment of the present application; Figure 2 A schematic diagram of a process for confirming first entity information based on a RoBERTa model provided in an embodiment of the present application; Figure 3 The structure of the multi-entity parallel graph neural network model provided in the embodiment of the present application; Figure 4 The architecture of the self-learning electronic medical record system provided in the embodiment of the present application; Figure 5 The application process of the electronic medical record information processing model provided in the embodiment of the present application; Figure 6 The graphical effect of the knowledge graph provided in the embodiment of the present application; Figure 7 A schematic diagram of an intelligent recommendation scheme provided in an embodiment of the present application; Figure 8 A system schematic diagram of an electronic medical record information processing system based on deep learning provided in an embodiment of the present application; Fig. 9A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0015] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0016] Figure 1 FIG. 1 is a flowchart of an electronic medical record information processing method based on deep learning provided by an embodiment of the present application. Figure 1 As shown, the method includes the following steps 301 to 305: Step 301: Acquire first entity information in a first electronic medical record through a first sub-model in an electronic medical record information processing model.

[0017] In an embodiment of the present application, the above-mentioned first electronic medical record may include a single electronic medical record or may include multiple electronic medical records, which is not limited in this embodiment of the present application.

[0018] In an embodiment of the present application, the first entity information may include entity information of a single electronic medical record, or entity information of multiple electronic medical records. When the first entity information includes entity information of multiple electronic medical records, each electronic medical record corresponds to the entity information of each electronic medical record.

[0019] In an embodiment of the present application, the above-mentioned first entity information includes at least one of the following: user status information, and a scheme corresponding to the user status information.

[0020] In one example, the user status information may include age, gender, medical history, symptoms, etc., and the treatment plans corresponding to the user status information include: medication plan, surgical plan, physical treatment plan, etc.

[0021] It can be understood that in the embodiment of the present application, the above-mentioned user status information includes information used to characterize user symptoms; the solutions corresponding to the above-mentioned user status information include: solutions, mitigation solutions, or improvement solutions for the above-mentioned user symptoms.

[0022] In one example, for the first electronic medical record, the user status information may be a blood sugar level during a specific period of time. When the blood sugar level during the specific period of time is high, the solution corresponding to the user status information includes the name of a drug that can make the blood sugar level return to normal and the method of using the drug.

[0023] For example, the first electronic medical record contains a large amount of data, and the data that needs to be identified and extracted by the RoBERTa model is not all data. It is necessary to screen the data information therein to obtain the first entity information. Based on this, the first entity information is part of the information of the first electronic medical record used to indicate the user status information and the scheme corresponding to the user status information.

[0024] Furthermore, the first entity information is essentially information that can be used to characterize the complete physical condition of the user of the first electronic medical record in the most concise way. For example, in the first electronic medical record, the following information is included: "Wang XX has high blood sugar, and the blood sugar value is XXX", then after the recognition and processing of the RoBERTa model, the first entity information is: "Wang XX; blood sugar value: XXX". That is, the first entity information only includes information in the first electronic medical record that can actually characterize the user's characteristics.

[0025] In this way, since the electronic medical record information processing model needs to process a large number of electronic medical records in actual applications, by screening the first entity information, there is no need to identify and calculate excessive amounts of information in the subsequent process of using the electronic medical record information processing model to generate recommended information, thereby saving processing resources and greatly improving processing efficiency.

[0026] In an embodiment of the present application, the first entity information includes: past entity information and current entity information, and the current entity information includes the user status information closest to the current time in the first electronic medical record.

[0027] It is understandable that, for the first electronic medical record, the information it contains may be multiple medical symptom information and treatment plans of the user to whom the first electronic medical record belongs, that is, the first electronic medical record includes both historical medical symptom information and treatment plans, as well as current medical symptom information.

[0028] Furthermore, in the embodiment of the present application, since the ultimate technical effect to be achieved is to generate recommendation information for the current user status information, the current entity information does not actually include the current solution corresponding to the user status information in the current entity information.

[0029] In an embodiment of the present application, the above-mentioned first sub-model is a RoBERTa model.

[0030] It is understandable that the RoBERTa (Robustly Optimized BERT Approach) model is a pre-trained language model based on the Transformer architecture, proposed by Facebook AI Research (FAIR). The RoBERTa model has made several improvements based on the BERT model to improve the performance and robustness of the model.

[0031] Step 302: Convert the first entity information into first vector information.

[0032] In the embodiment of the present application, after the first entity information is screened out, in order to align and unify the first entity information and subsequently distinguish the association relationship between different entity information in the first entity information, the first entity information needs to be converted into first vector information.

[0033] In one example, the conversion of the first entity information into the first vector information may be accomplished through a RoBERTa model.

[0034] Furthermore, the first entity information may include user status information, such as user symptoms, and a plan corresponding to the user status information, such as a specific drug used.

[0035] Step 303: Identify the association relationship between different first vector information in the first vector information through the second sub-model in the electronic medical record information processing model.

[0036] In an embodiment of the present application, the above-mentioned second sub-model is a multi-entity parallel graph neural network model.

[0037] It is understandable that the above-mentioned multi-entity parallel graph neural network model (MEP-GNN) plays a key role in the embodiments of the present application. Specifically, through the graph convolution layer and the self-attention layer, the model can capture the complex relationship between different entity information in the first entity information, thereby constructing a knowledge graph. At the same time, since the model has been trained with a large number of electronic medical records, it can be used to identify the schemes associated with the corresponding user status information in the first entity information, such as medication schemes. Therefore, the model has the ability to provide schemes that may improve the corresponding status in the user status information, such as proposing lifestyle adjustment suggestions, and can also evaluate the risks corresponding to the current user status information. The multi-entity parallel graph neural network model can generate accurate recommendations through the fusion layer and the output layer, provide comprehensive decision support, and improve the quality and efficiency of the doctor's judgment process.

[0038] It is understandable that for the first electronic medical record, it is very likely to have a lot of previous medical record information, which means that the user of the first electronic medical record has undergone multiple treatments, and the treatment process is progressive and related. Therefore, it is necessary to use a multi-entity parallel graph neural network model to sort out the different first vector information that have been extracted to represent the previous medical record information, and obtain the correlation in the first vector information, so as to better help the doctors who view the first electronic medical record to clarify the content in the future.

[0039] Step 304: Generate a knowledge graph based on the above association relationships.

[0040] In an embodiment of the present application, the above-mentioned knowledge graph is used to characterize the key information in the above-mentioned first electronic medical record.

[0041] It is understandable that after obtaining the above-mentioned association relationship, the relationship between different information in the first electronic medical record can be known. In order to display the information relationship in the first electronic medical record more intuitively and accurately, it can be displayed in the form of a knowledge graph.

[0042] Step 305: Identify the above-mentioned knowledge graph through the above-mentioned electronic medical record information processing model, and generate recommendation information corresponding to the above-mentioned first electronic medical record.

[0043] In the embodiment of the present application, the above-mentioned recommendation information is used to characterize the adaptation solution corresponding to the above-mentioned user status information.

[0044] It is understandable that after obtaining the above-mentioned knowledge graph, the trained electronic medical record information processing model can be used to identify the knowledge graph and generate an adaptation solution corresponding to the current user status information.

[0045] In the method provided in the embodiment of the present application, the user status information and / or the scheme corresponding to the user status information in the first electronic medical record is obtained through the RoBERTa model in the electronic medical record information processing model, and the first entity information: past entity information and current entity information, the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and the first sub-model is the RoBERTa model; then, the above-mentioned first entity information is converted into first vector information; then, the association relationship between different first vector information in the first vector information is identified through the multi-entity parallel graph neural network model in the electronic medical record information processing model; then, an indication graph for characterizing the key information in the first electronic medical record is generated according to the association relationship; then, the above-mentioned knowledge graph is identified through the above-mentioned electronic medical record information processing model, and recommendation information corresponding to the first electronic medical record is generated for characterizing the adaptation scheme corresponding to the user status information. In this way, by using different sub-models in the electronic medical record information processing model, the first electronic medical record is processed and information is screened, and finally the most critical information is screened out to generate a knowledge graph, and the key information is combined to match the adaptation plan corresponding to the user status information for the doctor's reference, thereby greatly improving the efficiency, accuracy and comprehensiveness of doctors' viewing of information.

[0046] In another embodiment of the present application, a specific implementation method for displaying a knowledge graph is also provided. Exemplarily, the specific implementation of "generating a knowledge graph according to the above-mentioned association relationship" mentioned above includes: generating a knowledge graph according to the above-mentioned association relationship, and displaying the above-mentioned knowledge graph in a display interface; the "generating the above-mentioned recommendation information corresponding to the above-mentioned first electronic medical record" mentioned above includes: receiving a first input from a user to a first control; displaying the above-mentioned recommendation information in a display interface corresponding to the above-mentioned first electronic medical record.

[0047] Exemplarily, after the above-mentioned knowledge graph is generated according to the association relationship, the above-mentioned knowledge graph can be directly displayed by executing the display terminal of the present application, so that the user can directly view the association relationship of key information through the visualized knowledge graph, which is more intuitive and accurate, and improves the efficiency of the user's subsequent judgment of the actual physical condition of the patient to whom the first electronic medical record belongs.

[0048] Exemplarily, the first input may be a touch input, a voice input, or a contactless gesture input, which is not limited in the embodiments of the present application.

[0049] It is understandable that after displaying the above-mentioned knowledge graph, there may be a user who needs to view the knowledge graph first to understand the previous user status information, the solutions corresponding to the previous user status information, and the relationship between key information in the current user status information. Therefore, after the user views the aforementioned knowledge graph, the subsequent recommended information can be displayed on the display terminal by inputting the first input to trigger the display terminal to display the subsequent recommended information, thereby ensuring the convenience of users using this method to display recommended information.

[0050] In another embodiment of the present application, a specific implementation method for obtaining the first entity information is also provided. Exemplarily, the "obtaining the first entity information in the first electronic medical record through the first sub-model" mentioned above includes: identifying the first electronic medical record through a position coding formula, confirming the information order of the text information of each first sub-electronic medical record in the first electronic medical record and the information length of the text information; confirming the first entity information in the first electronic medical record according to the information order and information length.

[0051] Exemplarily, the above position encoding formula is a formula in the RoBERTa model for identifying entity information in the first electronic medical record.

[0052] Exemplarily, the text information includes at least one of the following: vocabulary information and phrase information, and the order of the information is used to indicate the contextual relationship of the text information.

[0053] Understandably, the positional encoding formula is used in the RoBERTa model, which is one of the key steps in entity recognition. When processing natural language, positional encoding helps the model understand the order of words in a sentence. This is crucial for accurately identifying entities in electronic medical records, as the context of the entity can provide important information to assist in determining the user's symptom freeze.

[0054] In the RoBERTa model in the related art, position encoding is calculated by the following formulas (1) and (2): (1) (2) In the embodiment of the present application, the position coding formula is as follows. Specifically, considering the length of the text information in the first electronic medical record and the specificity of the text in the electronic medical record, the embodiment of the present application adjusts the position coding formulas (1) and (2) to become formulas (3) and (4): (3) (4) In formula (3) and formula (4), is the character position in the text, represents the dimension index, is the dimension of the encoded vector, It is also the hidden layer dimension of the model, and α is a constant less than 1.

[0055] It can be understood that the hidden layer here is the layer between the input layer and the output layer of the neural network, which is used to extract and transform the features of the input data. In the Transformer architecture, the dimension dmodel of the hidden layer determines the width of the data processed by the model in each hidden layer. This dimension is crucial to the performance and learning ability of the model. It can be seen that in the above formulas (3) and (4), compared with formulas (1) and (2) in the related art, the model adaptive position coding is modified. In this way, according to the information length of the text information in the first electronic medical record, the scale of the position coding can be dynamically adjusted to better capture the contextual information relationship in the long text information in the first electronic medical record. Therefore, the domain-specific adjustment capability of the formula can be utilized to introduce medical domain-specific position coding parameters related to the first electronic medical record. By using medical text data for training in the pre-training stage, the information order and the length of the text information can be adjusted when applying the above formulas (3) and (4).

[0056] For example, in the above formula (3) and formula (4), by setting This parameter can set a domain-specific adjustment factor for the first electronic medical record, which is used to adjust the scale of the position coding. Specifically, it is dynamically adjusted according to the length and complexity of the first electronic medical record.

[0057] Afterwards, according to the above information order and information length, the first entity information of the above first electronic medical record can be obtained using the RoBERTa model.

[0058] Example 1: Figure 2 As shown, Figure 2 The process of confirming the first entity information in the first electronic medical record using the position encoding formula based on the combination of the RoBERTa model and other models is demonstrated. Figure 2In the figure, a stacked attention network architecture is used that is based on the combination of the RoBERTa model and other models and belongs to multi-model fusion. It can be seen that after the input layer 21 of the architecture receives the data of the first electronic medical record, it is encoded by the model encoder 22 of the RoBERTa model, wherein in the encoder 22, the above position encoding formula helps the RoBERTa model understand the position of the word in the sentence. Then, the local and sequence features are processed respectively by the CNN layer 23 and the BiLSTM layer 24 in the stacked attention network architecture, and finally these features are merged in the fusion layer 25, and the final feature representation of the output layer 26 is the first entity information.

[0059] In another embodiment of the present application, a specific implementation method for obtaining the first vector information is also provided. Exemplarily, the aforementioned "identifying the association relationship between different vector information in the first vector information through the second sub-model" includes: identifying the correspondence between different categories of vector information in the first vector information through a multi-entity parallel graph neural network model.

[0060] Exemplarily, the above categories include: user status categories, and solution categories corresponding to the user status categories.

[0061] It can be understood that the above categories correspond to different categories of first entity information.

[0062] Exemplarily, the multi-entity parallel graph neural network model is used to identify the above-mentioned different categories of vectors and assign self-attention scores to the above-mentioned different categories of vectors, and the above-mentioned self-attention scores are determined by pre-setting electronic medical record domain knowledge.

[0063] Exemplarily, in the above-mentioned multi-entity parallel graph neural network model, the vector relationship between different vector information in the first vector information, that is, the correspondence between the above-mentioned different categories of vector information, can be determined through the self-attention mechanism.

[0064] It is understandable that the self-attention mechanism is used for relation extraction in the multi-entity parallel graph neural network model. It allows the model to consider the influence of other related entities in the knowledge graph when processing a pair of entities. This is crucial for understanding complex relationships between entities. For example, it can be used to determine the association between a specific drug and a specific disease. In the multi-entity parallel graph neural network model in the related art, the self-attention score is calculated as follows (5): (5) In the embodiment of the present application, an adjustment is required. The adjusted self-attention score calculation formula introduces domain knowledge to adjust the self-attention coefficient, and dynamically adjusts the weight as shown in (6).

[0065] (6) Is the vector information of the first entity information in the first electronic medical record and the vector information of the first entity information The self-attention score between the two entities, the vector information of the first entity information and the vector information of the first entity information For different vector information. yes and The self-attention coefficient between right The importance of.

[0066] It can be understood that in the above multi-entity parallel graph neural network model, Is Entity and entities The self-attention coefficient between is obtained through the training process. Specifically, it is calculated as shown in the following formula (7): The dynamic weight based on domain knowledge is also obtained through the training process and is calculated as shown in (8): (7) (8) is the activation function, and is the weight matrix used to calculate the self-attention coefficient, Represents a vector concatenation operation. entity The set of first-order neighbors of . Represents the domain knowledge weight adjustment factor, which is used to control the influence of domain knowledge in self-attention calculation. Based on medical knowledge, and entities By introducing dynamic weights based on domain knowledge ,The self-attention score calculation formula can better capture the complex relationships in ,electronic medical records and improve the accuracy of relationship extraction.

[0067] In an embodiment of the present application, by introducing domain knowledge related to the first electronic medical record into the multi-entity parallel graph neural network model, prior knowledge in the medical field is specifically introduced in the calculation of the self-attention coefficient, such as adjusting the relationship weights between entities through a medical ontology or an expert knowledge base. In addition, dynamic weight adjustment is performed to dynamically adjust the self-attention coefficient according to the importance of the entity to highlight the impact of key entities on relationship extraction. This adjustment helps the model to be more accurate and robust when processing entity relationships in the first electronic medical record.

[0068] It should be noted that in machine learning and deep learning, the exponential function is often used in activation functions, such as in the softmax function, which helps convert any real vector into a probability distribution. The exponential function is also used to calculate the normalization constant of continuous probability distributions, such as the Gaussian distribution. express It will self-attention coefficient Converts to non-negative values ​​that can be used to calculate probability distributions.

[0069] Example 2: Figure 3 As shown, Figure 3 The structure of the multi-entity parallel graph neural network model is shown. The model first collects electronic medical record data from two perspectives and generates a knowledge graph: entity interaction perspective 31 and relationship interaction perspective 32. Each perspective processes the corresponding information through a graph convolution layer, which is a graph convolution layer 33 and a graph convolution layer 34 respectively. Then, the information of the two perspectives is fused through a gated aggregation mechanism and input into a fusion layer 35, and finally the embedded representation of the entity is output, which is a knowledge graph 36. Figure 6 As shown, Figure 6 The graphical effect of the knowledge graph displayed to users (such as doctors).

[0070] In another embodiment of the present application, a specific implementation method for obtaining a first electronic medical record by preprocessing is also provided. Exemplarily, before the "obtaining the first entity information in the first electronic medical record by using the RoBERTa model" mentioned above, the above method includes: performing information preprocessing on the first information in the electronic medical record to generate the first electronic medical record.

[0071] Exemplarily, the first information mentioned above includes: personal privacy information in the electronic medical record.

[0072] Exemplarily, the above-mentioned processing methods include: data cleaning, data formatting and data desensitization.

[0073] Exemplarily, the above preprocessing process is used to desensitize the first electronic medical record data from different medical institutions and systems to protect patient privacy, and to clean and format the data of the first electronic medical record for subsequent processing.

[0074] In another embodiment of the present application, a specific implementation method for obtaining an electronic medical record information processing model is also provided. Exemplarily, before the "obtaining the first entity information in the first electronic medical record through the electronic medical record information processing model" mentioned above, the above method also includes: obtaining multiple groups of original electronic medical records and target data corresponding to the above multiple groups of original electronic medical records; inputting the above multiple groups of original electronic medical records into the initial electronic medical record information processing model to generate target data corresponding to the above electronic medical records, and the above target data includes a test knowledge graph and test recommendation information; comparing the above multiple groups of test data with the above multiple groups of target data, and determining the difference between the above multiple groups of test data and the above multiple groups of target data; when the above difference is greater than a preset threshold, adjusting the above initial electronic medical record information processing model according to the above difference, the above initial electronic medical record information processing model includes an initial RoBERTa model and an initial multi-entity parallel graph neural network model, so that the difference between the test data and the target data generated by the above electronic medical record information processing model is less than the preset threshold.

[0075] Exemplarily, the target data includes: a target knowledge graph and target recommendation information. The knowledge graph and recommendation information corresponding to the multiple groups of original electronic medical records are knowledge graphs and recommendation information composed of data in a target format corresponding one-to-one to each sub-medical record in the multiple groups of original electronic medical records. The multiple groups of original electronic medical records include data information of at least two users.

[0076] Exemplarily, the test recommendation information is used to characterize the adaptation solution corresponding to the user status information.

[0077] It can be understood that in the embodiment of the present application, it is necessary to train the initial electronic medical record information processing model multiple times using target data and original data corresponding to multiple groups of original electronic medical records, and compare the difference between each training result and the target result. The difference can be expressed by the above-mentioned difference. Only when the difference is less than the preset threshold can an electronic medical record information processing model that can be put into actual use be obtained.

[0078] It can be understood that in an embodiment of the present application, the above-mentioned electronic medical record information processing model includes multiple sub-models, such as a first sub-model and a second sub-model, wherein the second sub-model is a multi-entity parallel graph neural network model, and the training process of the multi-entity parallel graph neural network model can be completed by applying its loss function.

[0079] Specifically, the alignment loss function is used to optimize the multi-entity parallel graph neural network model so that the distance between positive triples (head entity, relationship, tail entity) is smaller than the distance between negative triples. This helps the electronic medical record information processing model learn more accurate feature representations, thereby improving the accuracy of knowledge discovery.

[0080] In the related art, the alignment loss function is formula (9), which is calculated by the entity alignment loss as follows: (9) represents entity alignment loss; is the set of positive examples of entity alignment; is the set of negative examples for entity alignment, and is an entity in the positive pair, and is the entity in the negative pair, is the marginal hyperparameter, Represents the L2 norm, which is used to calculate the distance between entities.

[0081] In the embodiment of the present application, the adjusted alignment loss function used by the multi-entity parallel graph neural network model is as follows: (10) represents the triplet alignment loss, is the set of positive triples, is the set of negative triples, is the embedding representation of the head entity, relation, and tail entity in the positive triple. and is the embedding representation of the corresponding entity. is the embedding representation of the head entity, relation, and tail entity in the negative triple. is the marginal hyperparameter, the distance function Used to calculate the distance between embedded vectors, usually using Euclidean distance or cosine similarity, It represents the distance between the head entity and the tail entity in the positive triplet in the feature space, usually through the relationship vector To convert the head entity vector To match the tail entity vector , is the distance between the head entity and the tail entity in the negative triplet in the feature space.

[0082] In the embodiment of the present application, the above formula (10) adjusts the balance of the loss function by dynamically adjusting and The value of is adjusted to balance the contribution of entity alignment loss and triple alignment loss in the total loss. On the other hand, by adjusting the weights of positive and negative samples, positive and negative sample weights are introduced to deal with the problem of class imbalance, especially in electronic medical record data, some entity pairs or triples may be far more than other entity pairs or triples. Through these adjustments, the model can better adapt to the characteristics of electronic medical record data and improve the performance of the model in entity recognition, relationship extraction and knowledge graph construction. These adjustments need to be verified and optimized in actual data and application scenarios.

[0083] It can be understood that after the training is completed, the electronic medical record information processing model obtained actually constitutes a self-learning electronic medical record system, such as Figure 4 The architecture shown in the figure. In practical applications, this electronic medical record information processing model can extract data information from the acquired electronic medical records, identify entities, and extract the corresponding recommended information of the current electronic medical record through the similarity algorithm obtained in the previous training process, that is, the features that solve the problems in the current electronic medical record. Then, the electronic medical record information processing model generates recommended information based on these features, that is, a list of recommended medical orders (see the subsequent table content for details), to provide decision support for doctors. This electronic medical record information processing model can automatically learn and recommend the best recommended information based on the specific situation and historical data of the user to whom the electronic medical record belongs.

[0084] The method of electronic medical record information processing based on deep learning in this application can be applied to the electronic medical record information processing system based on deep learning. We can design a specific application scenario, namely medical order analysis. In this scenario, the system first integrates and preprocesses previously generated electronic medical record data from different medical institutions and systems, including desensitization to protect patient privacy, as well as cleaning and formatting for subsequent processing. Next, the system uses the positional encoding formula based on the RoBERTa model to identify medical entities in the medical records, that is, the first test information, such as disease name, symptom description, and drug name, where the positional encoding formula and This helps the model understand the order of words in a sentence, which is crucial for accurately identifying entities. Then, through the self-attention score calculation formula in the multi-entity parallel graph neural network model, the system can identify the relationship between entities, such as the relationship between diseases and symptoms, and the relationship between drugs and treatment plans. The self-attention score calculation formula and Allows the model to consider the influence of other related entities in the knowledge graph when processing the first entity information. The system then constructs the identified entities and extracted relationships into a knowledge graph for easy knowledge storage and query. Using the alignment loss function, the system optimizes the model so that the distance between positive example triplets is smaller than the distance between negative example triplets, which helps the model learn more accurate feature representations, thereby improving the accuracy of knowledge discovery. Entity alignment loss function And triplet alignment loss function For the optimization model, and is the marginal hyperparameter, stands for L2 norm, which is used to calculate the distance between entities. Finally, the system learns the common features of similar patients' treatment plans based on the patient similarity algorithm, and provides the recommended information to doctors in the form of a recommended medical order list, which helps to improve the efficiency of subsequent doctors' judgment on the first electronic medical record. Through this process, the system can effectively extract valuable medical knowledge from electronic medical records, assist medical decision-making and research, and ultimately realize a high-precision medical and intelligent medical service system.

[0085] The following specific example 3 is combined Figure 5 The following is an illustration of the application process of the electronic medical record information processing model of this application: Example 3: S101: Preprocess the data of the first electronic medical record through the electronic medical record information processing model: clean and format the medical history and corresponding treatment plan (previous user status), and the examination results and treatment plan (current user status) in the patient's first electronic medical record, that is, preprocess the data by using natural language processing technology.

[0086] S102: Entity Recognition: Use the pre-trained stacked attention network architecture (the first sub-model in the electronic medical record information processing model, that is, applying the RoBERTa model and the position encoding formula) to identify entity information in the electronic medical record and convert the entity information into vector information.

[0087] S103: Relationship extraction: The relationship between vector information, such as the relationship between symptoms, drugs and treatment plans, is extracted through the second sub-model in the electronic medical record information processing model, namely the multi-entity parallel graph neural network model.

[0088] S104: Knowledge graph construction: Organize entities and relationships in the form of graphs and store them in the database to construct a knowledge graph, such as Figure 6 As shown, this part can be completed by optimizing the alignment loss function.

[0089] S105: Knowledge discovery: Use the second sub-model in the electronic medical record information processing model, namely the multi-entity parallel graph neural network model to analyze the knowledge graph and improve the accuracy of knowledge discovery.

[0090] S106: Self-learning recommendation: By developing a patient similarity algorithm in the electronic medical record information processing model, medical advice recommendations (i.e. the above-mentioned recommendation information) are generated by comparing and matching similar cases.

[0091] The process for a complex case of a 48-year-old woman with gestational hypertension, diabetes, and kidney disease was as follows: Figure 7 As shown, Figure 7 The following is an intelligent recommendation solution based on the electronic medical records and search results of simulated diabetes, kidney disease, hypertension, and obstetrics and gynecology. It should be noted that this result is only a recommendation information, not a diagnostic information, and further manual judgment is required to determine whether it is adoptable information:

[0092] Through the display of the recommended information and knowledge graph, doctors can quickly sort out the patient information in the first electronic medical record and obtain the recommended information of the electronic medical record information processing model, thereby improving judgment efficiency. For example, if the recommended information does include the medicine that the doctor needs to prescribe, it can save typing time and improve efficiency.

[0093] The present application also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the training rule determination method described in the present application is implemented. For example, the following can be executed: Figure 1 The individual steps of the method are shown.

[0094] The present invention provides a computer program product, which includes instructions that are executed by a processor when the processor executes the instructions. Figure 1 The individual steps of the method are shown.

[0095] It should be noted that although the operations of the method of the present invention are described in a particular order in the drawings, this does not require or imply that the operations must be performed in this particular order or that all illustrated operations must be performed to achieve desired results.

[0096] Figure 8 This is a block diagram of an electronic medical record information processing system based on deep learning according to an embodiment of the present application. Figure 8 The system includes an acquisition unit 601, an execution unit 602, an identification unit 603, and a generation unit 604.

[0097] An acquisition unit 601 is used to acquire first entity information in a first electronic medical record through a first sub-model in an electronic medical record information processing model, wherein the first entity information includes at least one of the following: user status information and a scheme corresponding to the user status information, wherein the first entity information includes: past entity information and current entity information, wherein the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and wherein the first sub-model is a RoBERTa model; An execution unit 602 is used to convert the first entity information into first vector information; An identification unit 603, configured to identify associations between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model, wherein the second sub-model is a multi-entity parallel graph neural network model; A generating unit 604 is used to generate a knowledge graph according to the association relationship, wherein the knowledge graph is used to represent key information in the first electronic medical record; The generating unit 604 is further configured to identify the knowledge graph through the electronic medical record information processing model, and generate recommendation information corresponding to the first electronic medical record, wherein the recommendation information is used to characterize an adaptation solution corresponding to the user status information.

[0098] In one embodiment, the acquisition unit 601 is used to identify the first electronic medical record through a position coding formula, confirm the information order of the text information of each first sub-electronic medical record in the first electronic medical record and the information length of the text information, the text information including at least one of the following: vocabulary information and phrase information, the information order is used to indicate the contextual relationship of the text information; and is also used to confirm the first entity information in the first electronic medical record based on the information order and information length.

[0099] In one embodiment, the identification unit 601 is specifically configured to: Through a multi-entity parallel graph neural network model, the correspondence between different categories of vector information in the first vector information is identified, and the categories include: user status categories, and solution categories corresponding to user status categories; the multi-entity parallel graph neural network model is used to identify vectors of different categories and assign self-attention scores to the vectors of different categories, and the self-attention scores are determined by preset electronic medical record domain knowledge.

[0100] In one embodiment, the execution unit 601 is further configured to: Preprocessing the first information in the electronic medical record to generate a first electronic medical record; Wherein, the first information includes: personal privacy information in the electronic medical record.

[0101] In one embodiment, the execution unit 601 is further configured to: Acquire multiple groups of original electronic medical records and target data corresponding to the multiple groups of original electronic medical records, wherein the target data includes: a target knowledge graph and target recommendation information, wherein the knowledge graph and recommendation information corresponding to the multiple groups of original electronic medical records are knowledge graphs and recommendation information composed of data in a target format corresponding to each sub-medical record in the multiple groups of original electronic medical records, and the multiple groups of original electronic medical records include data information of at least two users; Input the multiple groups of original electronic medical records into an initial electronic medical record information processing model to generate target data corresponding to the electronic medical records, wherein the target data includes a test knowledge graph and test recommendation information, and the test recommendation information is used to characterize an adaptation solution corresponding to the user status information; Comparing the multiple sets of test data with the multiple sets of target data, and determining differences between the multiple sets of test data and the multiple sets of target data; When the difference is greater than a preset threshold, the initial electronic medical record information processing model is adjusted according to the difference. The initial electronic medical record information processing model includes an initial RoBERTa model and an initial multi-entity parallel graph neural network model, so that the difference between the test data and the target data generated by the electronic medical record information processing model is less than a preset threshold.

[0102] It should be understood that the units recorded in the electronic medical record information processing system based on deep learning correspond to the various steps in the method described in the accompanying drawings. Therefore, the operations and features described above for the method are also applicable to the electronic medical record information processing system based on deep learning and the units contained therein, and will not be repeated here. The electronic medical record information processing system based on deep learning can be pre-implemented in a browser or other security application of a computer device, or can be loaded into a browser or its security application of a computer device by downloading or the like. The corresponding units in the electronic medical record information processing system based on deep learning can cooperate with the units in the computer device to implement the solution of the embodiment of the present application.

[0103] For the several modules or units mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0104] It should be noted that for the details not disclosed in the deep learning-based electronic medical record information processing system in the embodiments of the present application, please refer to the details disclosed in the above embodiments of the present application, and no further details will be given here.

[0105] Reference below Fig. 9 , Fig. 9 FIG. 1 is a schematic diagram showing a structure of a computer device suitable for implementing an embodiment of the present application. Fig. 9 As shown, the computer system 1700 includes a central processing unit (CPU) 1701, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1702 or the program loaded from the storage part 1708 to the random access memory (RAM) 1703. In the RAM 1703, various programs and data required for the operation instructions of the system are also stored. The CPU 1701, the ROM 1702, and the RAM 1703 are connected to each other through a bus 1704. An input / output (I / O) interface 1705 is also connected to the bus 1704.

[0106] The following components are connected to the I / O interface 1705: an input section 1706 including a keyboard, a mouse, etc.; an output section 1707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1708 including a hard disk, etc.; and a communication section 1709 including a network interface card such as a LAN card, a modem, etc. The communication section 1709 performs communication processing via a network such as the Internet. A drive 1710 is also connected to the I / O interface 1705 as needed. A removable medium 1711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1710 as needed, so that a computer program read therefrom is installed into the storage section 1708 as needed.

[0107] In particular, according to an embodiment of the present application, the above reference flow chart Figure 1 The described process can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 1709, and / or installed from a removable medium 1711. When the computer program is executed by the central processing unit (CPU) 1701, the above-mentioned functions defined in the system of the present application are executed.

[0108] It should be noted that the computer-readable medium shown in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium such as a computer-readable storage medium that can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0109] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operating instructions of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the aforementioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, the boxes represented by two connections can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operating instruction, or can be implemented with a combination of dedicated hardware and computer instructions.

[0110] The units or modules involved in the embodiments described in the present application may be implemented by software or hardware. The units or modules described may also be set in a processor. For example, it may be described as: a processor includes a first receiving module, a second receiving module, and a sending module. The names of these units or modules do not, in some cases, constitute limitations on the units or modules themselves.

[0111] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device. The above computer-readable storage medium stores one or more programs, and when the above programs are used by one or more processors to execute the electronic medical record information processing method based on deep learning described in the present application.

[0112] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) by each other to form a technical solution.

Claims

1. A method for processing electronic medical record information based on deep learning, characterized in that: include: Acquire first entity information in the first electronic medical record through the first sub-model in the electronic medical record information processing model, the first entity information includes at least one of the following: user status information, a scheme corresponding to the user status information, the first entity information includes: past entity information and current entity information, the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and the first sub-model is a RoBERTa model; Converting the first entity information into first vector information; Identify the association relationship between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model, wherein the second sub-model is a multi-entity parallel graph neural network model; Generate a knowledge graph according to the association relationship, wherein the knowledge graph is used to represent key information in the first electronic medical record; The knowledge graph is identified through the electronic medical record information processing model, and recommendation information corresponding to the first electronic medical record is generated, where the recommendation information is used to characterize an adaptation solution corresponding to the user status information.

2. The method according to claim 1, characterized in that The generating of the knowledge graph according to the association relationship comprises: Generate a knowledge graph according to the association relationship, and display the knowledge graph in a display interface; The generating of recommendation information corresponding to the first electronic medical record includes: Receive a first input from a user to a first control; The recommendation information is displayed in a display interface corresponding to the first electronic medical record.

3. The method according to claim 1, characterized in that: The obtaining the first entity information in the first electronic medical record by using the first sub-model includes: Identify the first electronic medical record by using a position coding formula, confirm the information sequence of text information of each first sub-electronic medical record in the first electronic medical record and the information length of the text information, the text information including at least one of the following: vocabulary information and phrase information, the information sequence is used to indicate the contextual relationship of the text information; According to the information sequence and the information length, the first entity information in the first electronic medical record is confirmed.

4. The method according to claim 1, characterized in that The identifying, by using the second sub-model, the association relationship between different vector information in the first vector information includes: Through a multi-entity parallel graph neural network model, the correspondence between different categories of vector information in the first vector information is identified, and the categories include: user status categories, and solution categories corresponding to user status categories; the multi-entity parallel graph neural network model is used to identify vectors of different categories and assign self-attention scores to the vectors of different categories, and the self-attention scores are determined by preset electronic medical record domain knowledge.

5. The method according to claim 1, characterized in that Before obtaining the first entity information in the first electronic medical record through the RoBERTa model, the method includes: Preprocessing the first information in the electronic medical record to generate a first electronic medical record; Wherein, the first information includes: personal privacy information in the electronic medical record.

6. The method according to claim 1, characterized in that Acquiring first entity information in a first electronic medical record through an electronic medical record information processing model, the method further comprising: Acquire multiple groups of original electronic medical records and target data corresponding to the multiple groups of original electronic medical records, wherein the target data includes: a target knowledge graph and target recommendation information, wherein the knowledge graph and recommendation information corresponding to the multiple groups of original electronic medical records are knowledge graphs and recommendation information composed of data in a target format corresponding to each sub-medical record in the multiple groups of original electronic medical records, and the multiple groups of original electronic medical records include data information of at least two users; Input the multiple groups of original electronic medical records into an initial electronic medical record information processing model to generate target data corresponding to the electronic medical records, wherein the target data includes a test knowledge graph and test recommendation information, and the test recommendation information is used to characterize an adaptation solution corresponding to the user status information; Comparing a plurality of sets of test data with the plurality of sets of target data, and determining differences between the plurality of sets of test data and the plurality of sets of target data; When the difference is greater than a preset threshold, the initial electronic medical record information processing model is adjusted according to the difference. The initial electronic medical record information processing model includes an initial RoBERTa model and an initial multi-entity parallel graph neural network model, so that the difference between the test data and the target data generated by the electronic medical record information processing model is less than a preset threshold.

7. An electronic medical record information processing system based on deep learning, characterized in that: include: an acquisition unit, configured to acquire first entity information in a first electronic medical record through a first sub-model in an electronic medical record information processing model, wherein the first entity information includes at least one of the following: user status information and a scheme corresponding to the user status information, wherein the first entity information includes: past entity information and current entity information, wherein the current entity information includes the user status information in the first electronic medical record that is closest to the current time, and wherein the first sub-model is a RoBERTa model; An execution unit, configured to convert the first entity information into first vector information; an identification unit, configured to identify an association relationship between different first vector information in the first vector information through a second sub-model in the electronic medical record information processing model, wherein the second sub-model is a multi-entity parallel graph neural network model; A generating unit, configured to generate a knowledge graph according to the association relationship, wherein the knowledge graph is used to represent key information in the first electronic medical record; The generating unit is further used to identify the knowledge graph through the electronic medical record information processing model, and generate recommendation information corresponding to the electronic medical record, wherein the recommendation information is used to characterize the adaptation solution corresponding to the user status information.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising instructions, characterized in that: When the instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images