Structured processing method and system for electronic health record, medium, and device

Abstract

Provided are a structured processing method and system for an electronic health record, a medium, and a device. The method comprises: acquiring electronic health record data; using a deep learning model to recognize clinical entities in the electronic health record data; mapping the clinical entities to preset data models by means of an information extraction algorithm to obtain mapping results; and verifying the clinical entities and the mapping results by means of an anomaly detection algorithm to obtain structured data. The clinical entities in an electronic health record are accurately and efficiently recognized and extracted, and converted into structured data, and the consistency and accuracy of the structured data can be ensured, thereby laying a foundation for the subsequent deep analysis and utilization of electronic health record data.

Description

Methods, systems, media and equipment for structuring electronic health records Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to a method, system, medium, and device for structuring electronic health records. Background Technology

[0002] An Electronic Health Record (EHR) contains a variety of patient data and information, such as personal information like age, height, and weight, as well as medical history, medication and allergy history, immunization status, laboratory test results, radiological images (such as X-rays), vital signs, medical process records, and payment information. Its purpose is to record patient data generated across all healthcare institutions. This digitally stored information needs to be shareable between different healthcare institutions to ensure patients receive quality medical care from different doctors, hospitals, clinics, and even in different countries. It also allows doctors, other healthcare professionals, and insurance companies to share the patient's medical records across different devices.

[0003] However, EHRs are diverse in nature. For example, different organizations use different recording standards. At the same time, EHRs contain a large amount of text with different structures and technical terms, which leads to problems of low efficiency and accuracy in information extraction and recognition when they are structured.

[0004] To better utilize data in EHRs to drive the development of digital healthcare, solving the complex problem of structuring electronic health record data is a significant challenge we currently face.

[0005] Summary of the Invention

[0006] In view of the deficiencies in the prior art, the purpose of this invention is to provide a method, system, medium and device for structuring electronic health records.

[0007] According to the present invention, a method for structuring electronic health records includes: acquiring electronic health record data; using a deep learning model to identify clinical entities in the electronic health record data to obtain clinical entities; wherein the deep learning model is trained using an entity dataset corresponding to a medical professional dictionary; mapping the clinical entities to a preset data model using an information extraction algorithm to obtain a mapping result; obtaining structured data based on the clinical entities and the mapping result, wherein the mapping result includes the category of the data model to which the clinical entity belongs; and verifying the clinical entities and the mapping result using an anomaly detection algorithm to obtain verified structured data.

[0008] It is easy to understand that the deep learning model in this application is trained on an entity dataset corresponding to a medical professional dictionary. Therefore, compared with other deep learning models, the deep learning model in this application identifies clinical entities in electronic health record data more accurately. This effectively improves the efficiency and accuracy of clinical entity recognition. Furthermore, this application uses an information extraction algorithm to map the identified clinical entities to a pre-defined data model to obtain mapping results, adapting to electronic health record data of different formats or different medical fields, thereby obtaining structured data with high consistency. Furthermore, this application also uses an anomaly detection algorithm to verify the structured data, eliminating structured data with poor consistency, thereby further improving the accuracy and consistency of the structured data in this application.

[0009] Furthermore, the deep learning model identifies clinical entities by: converting each character of the electronic health record data into a character vector and a word vector to which each character belongs; concatenating the character vectors and word vectors to obtain a concatenated vector; extracting features from the concatenated vector to determine a concatenated feature vector; obtaining the probability of each character corresponding to a different label based on the concatenated feature vector; determining the label sequence of the electronic health record data based on the probability of each character corresponding to a different label; and determining the clinical entities in the electronic health record data based on the label sequence of the electronic health record data.

[0010] Furthermore, the deep learning model identifies clinical entities by: converting each character of the electronic health record data into a character vector and a word vector to which each character belongs; concatenating the character vectors and word vectors to obtain a concatenated vector; extracting features from the concatenated vector to determine a concatenated feature vector; obtaining the probability of each character corresponding to different labels based on the concatenated feature vector; processing the probability of each character corresponding to different labels by setting constraints to obtain an optimized probability of each character corresponding to different labels; determining the label sequence of the electronic health record data based on the optimized probability of each character corresponding to different labels; and determining the clinical entities in the electronic health record data based on the label sequence of the electronic health record data.

[0011] It is easy to understand that the constraint condition of this application is to represent the constraint relationship between each label through learned features. By setting this constraint condition, the probability of different labels corresponding to each character is processed, and the label sequence is evaluated using state feature functions and transition feature functions to obtain the optimized probability of each character corresponding to different labels. This ensures that the probability of the label of each character in the output can fully consider the transition dependency relationship of the electronic health record data (i.e., the dependency relationship between characters before and after in the electronic health record data, the dependency relationship of sentences in the electronic health record data, etc.). Based on the optimized probability of each character corresponding to different labels, the label sequence of the electronic health record data is determined, resulting in higher accuracy of the label sequence or clinical entity of the output electronic health record data. This improves the accuracy of the neural network model in recognizing clinical entities in the electronic health record data, thereby ensuring the accuracy of the final structured data.

[0012] Further, the step of mapping the clinical entity to a preset data model using an information extraction algorithm to obtain the mapping result includes: extracting entity attributes of the clinical entity using the information extraction algorithm to obtain the entity attributes of the clinical entity; mapping the clinical entity to a data model of the corresponding category according to the category of the preset data model and the entity attributes of the clinical entity, thereby obtaining the mapping result, wherein the mapping result includes the category of the data model to which the clinical entity belongs.

[0013] Furthermore, the information extraction algorithm is used to extract entity attributes from the clinical entity. The entity attributes of the clinical entity include: performing dependency analysis based on transition to analyze the dependency relationship of the input clinical entity and finding the core words of the electronic health record data; recursively searching for all words that are parallel to the core words to form a predicate list; for each predicate, finding its subject and object according to the dependency relationship to form an information tuple; for information tuples without a subject, searching for a parent tuple that is parallel to its predicate and has a subject; if the parent tuple exists, setting the subject of the information tuple as the subject of the parent tuple; otherwise, no processing is performed; and returning the extraction result to obtain the entity attributes of the clinical entity.

[0014] It is easy to understand that this application determines the entity attributes of clinical entities by fully considering the dependency relationships of the input clinical entities. The entity attributes of clinical entities can be Western medicine diagnosis and treatment, drug research, or traditional Chinese medicine diagnosis and treatment. It is also clear that the entity attributes of clinical entities determined in this application are not based on a single character or semantic meaning of the clinical entity, but rather on the sentence or paragraph in which the clinical entity is located. Therefore, the entity attributes of clinical entities determined in this application are more accurate. This, in turn, leads to higher accuracy in the structured data obtained by this application based on the mapping relationship (i.e., matching relationship) between the categories of the preset data model and the entity attributes of clinical entities.

[0015] Furthermore, the preset data model includes at least one of the following: a disease-drug data model, a traditional Chinese medicine (TCM) diagnosis and treatment data model, and a Western medicine diagnosis and treatment data model; correspondingly, the category of the preset data model includes at least one of the following: the disease-drug data model is categorized as drug research, the TCM diagnosis and treatment data model is categorized as TCM diagnosis and treatment, and the Western medicine diagnosis and treatment data model is categorized as Western medicine diagnosis and treatment.

[0016] Furthermore, the method for verification using an anomaly detection algorithm includes: performing clustering calculations on the clinical entities and their mapping results; based on the clustering calculation results corresponding to the clinical entities and their mapping results, removing clinical entities and their mapping results whose clustering calculation results are greater than a preset value, thereby obtaining verified structured data.

[0017] It is easy to understand that this application verifies structured data by performing clustering calculations on the clinical entities and their mapping results, and obtains verified structured data. That is, the verified structured data removes clinical entities and their mapping results whose clustering calculation results are greater than a preset value, and retains clinical entities and their mapping results whose clustering calculation results are not greater than a preset value. In other words, the verified structured data has better clustering, thereby making the consistency and accuracy of the verified structured data higher.

[0018] According to the present invention, an electronic health record structured processing system includes:

[0019] The acquisition module is used to acquire electronic health record data.

[0020] The natural language processing module is used to identify clinical entities in the electronic health record data using a deep learning model, thereby obtaining the clinical entities.

[0021] The information extraction and mapping module is used to map the clinical entity to a preset data model through an information extraction algorithm to obtain a mapping result, and to obtain structured data based on the clinical entity and the mapping result, wherein the mapping result includes the category of the data model to which the clinical entity belongs.

[0022] The quality control module is used to verify the clinical entities and the mapping results through anomaly detection algorithms to obtain verified structured data.

[0023] According to the present invention, a computer-readable storage medium storing a computer program is provided, wherein when the computer program is executed by a processor, it implements the steps of the electronic health record structuring processing method.

[0024] An electronic device according to the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the electronic health record structuring processing method.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] This invention can accurately and efficiently identify and extract clinical entities in electronic health records and convert them into structured data. Furthermore, by verifying the structured data, this application can further ensure the consistency and accuracy of the obtained structured data, laying the foundation for subsequent in-depth analysis and utilization of electronic health record data. Attached Figure Description

[0027] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0028] Figure 1 is a flowchart of the electronic health record structure processing method of the present invention;

[0029] Figure 2 is a flowchart of entity attribute extraction;

[0030] Figure 3 is a flowchart of the verification process using an anomaly detection algorithm;

[0031] Figure 4 is a schematic diagram of the structure of a deep learning model;

[0032] Figure 5 is a schematic diagram of a loop unit of the LSTM in the embodiment;

[0033] Figure 6 is a schematic diagram of the structure of another deep learning model;

[0034] Figure 7 is a system block diagram of the electronic health record structured processing system of the present invention. Detailed Implementation

[0035] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0036] This application provides a method for structuring electronic health records, specifically including acquiring electronic health record data, using a deep learning model to identify clinical entities in the electronic health record data to obtain clinical entities, mapping the clinical entities to a preset data model using an information extraction algorithm to obtain mapping results, obtaining structured data based on the clinical entities and their mapping results, and verifying the clinical entities and mapping results using an anomaly detection algorithm to obtain verified structured data. The structured data includes the clinical entities and the categories of the data models corresponding to the clinical entities.

[0037] It is easy to understand that this application uses a deep learning model to identify clinical entities in electronic health record data, effectively improving the efficiency and accuracy of clinical entity identification; it uses an information extraction algorithm to map the identified clinical entities to a preset data model to obtain mapping results, adapting to electronic health record data of different formats or different medical fields, thereby obtaining structured data with high consistency; and it uses an anomaly detection algorithm to verify the structured data, thereby further improving the accuracy and consistency of the structured data in this application.

[0038] As shown in Figure 1, a method for structuring electronic health records includes:

[0039] Step 1: Obtain electronic health record data.

[0040] Electronic health record data can be manually entered, imported from a database, or obtained from the network; this invention does not impose any restrictions on this.

[0041] Step 2: Use a deep learning model to identify clinical entities in the electronic health record data to obtain the clinical entities.

[0042] The deep learning model in this application is trained using an entity dataset corresponding to a medical professional dictionary, thereby enabling the deep learning model to more accurately identify clinical entities in health record data. Clinical entities can include symptoms, diagnoses, treatments, medications, and dosages.

[0043] In embodiments of this application, a deep learning model is used to determine the probability of different tags corresponding to each character in the electronic health record data. Based on the probability of different tags corresponding to each character in the electronic health record data, the tag of each character in the electronic health record data is determined, thereby determining the tag sequence of the electronic health record data. Clinical entities in the electronic health record data are determined based on the tag sequence of the electronic health record data.

[0044] Specifically, in some embodiments, using a deep learning model to identify clinical entities in the electronic health record data can include: converting each character of the electronic health record data into a character vector and a word vector to which each character belongs; concatenating the character vectors and word vectors to obtain a concatenated vector; extracting features from the concatenated vector to determine a concatenated feature vector; obtaining the probability of each character corresponding to a different label based on the concatenated feature vector; determining a label sequence of the electronic health record data based on the probability of each character corresponding to a different label; and determining the clinical entities in the electronic health record data based on the label sequence of the electronic health record data.

[0045] Specifically, the deep learning model may include an embedding layer, a feature extraction layer, and a fully connected layer. The embedding layer is used to convert each character of the electronic health record data into a character vector and the word vector to which each character belongs. The feature extraction layer is used to concatenate the character vectors and word vectors to obtain a concatenated vector. The fully connected layer is used to process the concatenated vector through an activation function to obtain the probability of different labels in the sequence annotation corresponding to each character, thereby obtaining the label sequence of the electronic health record data. Based on the label sequence of the electronic health record data, the clinical entities in the electronic health record data are determined.

[0046] In other embodiments, using a deep learning model to identify clinical entities in the electronic health record data can include: converting each character of the electronic health record data into a character vector and a word vector to which each character belongs; concatenating the character vectors and word vectors to obtain a concatenated vector; extracting features from the concatenated vector to determine a concatenated feature vector; obtaining the probability of each character corresponding to different labels based on the concatenated feature vector; processing the probabilities of each character corresponding to different labels by setting constraints to obtain optimized probabilities of each character corresponding to different labels; determining a label sequence of the electronic health record data based on the optimized probabilities of each character corresponding to different labels; and determining the clinical entities in the electronic health record data based on the label sequence of the electronic health record data.

[0047] Specifically, the deep learning model includes an embedding layer, a feature extraction layer, a fully connected layer, and a constraint layer. The embedding layer converts each character in the electronic health record data into a character vector and its corresponding word vector. The feature extraction layer concatenates the character vectors and word vectors to obtain a concatenated vector. The fully connected layer processes the concatenated vector using an activation function to obtain the probability of different labels in the sequence annotation for each character, thus obtaining a label sequence for the electronic health record data. Based on this label sequence, clinical entities in the electronic health record data are determined. The constraint layer learns the constraints between the labels and processes the probabilities of different labels corresponding to each character using these constraints to obtain an optimized label sequence.

[0048] Step 3: Map clinical entities to a pre-defined data model using an information extraction algorithm to obtain mapping results. Based on the clinical entities and the mapping results, structured data is obtained. The mapping results include the category of the pre-defined data model to which the clinical entity belongs.

[0049] Specifically, information extraction algorithms are used to extract entity attributes from clinical entities in electronic health record data. Based on a pre-defined data model and the extracted entity attributes, the clinical entities are mapped to corresponding data models. These pre-defined data models include: disease-drug data models, traditional Chinese medicine (TCM) treatment data models, and Western medicine treatment data models. Correspondingly, the disease-drug data model can be categorized as drug research, the TCM treatment data model as TCM treatment, and the Western medicine treatment data model as Western medicine treatment; the entity attributes include Western medicine treatment, drug research, and TCM treatment.

[0050] As shown in Figure 2, the process of extracting entity attributes from clinical entities in electronic health record data using information extraction algorithms includes:

[0051] S301: Transfer-based dependency analysis performs dependency relationship analysis on input clinical entities to identify core terms in electronic health record data.

[0052] S302: Use recursion to find all words that are parallel to the core word to form a predicate list.

[0053] S303: For each predicate, find its subject and object based on the dependency relationship to form an information tuple.

[0054] S304: For an information tuple without a subject, find a parent tuple that is parallel to its predicate and has a subject. If the parent tuple exists, set the subject of the information tuple as the subject of the parent tuple. If no such parent tuple is found, do not process it.

[0055] S305: Return the extraction results and obtain the entity attributes of the clinical entity.

[0056] As can be easily understood from the above, dependency analysis of transfer is used to analyze the dependency relationships of clinical entities, thereby determining the entity attributes of clinical entities. Based on the entity attributes of clinical entities and the category of the preset data model, the clinical entities are mapped to the corresponding preset data models, thus obtaining the mapping relationship between clinical entities and preset data models, that is, determining the category of the data model to which the clinical entity belongs.

[0057] It is easy to understand further that this application determines the entity attributes of clinical entities by fully considering the dependency relationships of the input clinical entities. It is clear that the entity attributes determined by this application are not based on a single character or semantic meaning of the clinical entity, but rather on the sentence or paragraph in which the clinical entity is located. Therefore, the entity attributes determined by this application are more accurate. Consequently, the structured data obtained by this application based on the mapping relationship (i.e., matching relationship) between the categories of the preset data model and the entity attributes of clinical entities is also more accurate.

[0058] Step 4: Verify the clinical entities and mapping results using anomaly detection algorithms to obtain verified structured data.

[0059] As shown in Figure 3, the structured data obtained by validating clinical entities and mapping results using anomaly detection algorithms can specifically include:

[0060] Step S401: Randomly select k from n clinical entities and their mapping results as initial cluster centers, and determine the categories of the k initial cluster centers.

[0061] Step S402: Based on the nk clinical entities and their mapping results, the k initial cluster centers, and the k categories that serve as the initial cluster centers, determine the categories to which the nk clinical entities and their mapping results belong.

[0062] In some embodiments, taking the category to which the i-th clinical entity and its mapping result belong as an example, determining the category to which the i-th clinical entity and its mapping result belong, based on the i-th clinical entity and its mapping result, k initial cluster centers, and k categories used as initial cluster centers, includes: traversing the k initial cluster centers, calculating the distance between the i-th clinical entity and its mapping result and each initial cluster center, and taking the distance between the k initial cluster centers and the i-th clinical entity and its mapping result x... i The category corresponding to the initial cluster center with the smallest distance is taken as the category of the i-th clinical entity.

[0063] Specifically, iterate through the k initial cluster centers, and determine the mapping result x between the j-th initial cluster center and the i-th clinical entity. i minimum distance ci It can be calculated using formula (1):

[0064] Where, x i Let u be the i-th clinical entity and its mapping result; j Let ||x be the j-th initial cluster center; i -u j || 2 Let x be the L2 norm of the difference between the i-th clinical entity and its mapping result and the j-th initial cluster center, i.e., the Euclid distance between the i-th clinical entity and its mapping result and the j-th initial cluster center; arg min is the value used to iterate and calculate the x-th clinical entity and its mapping result. i The Euclid distance from each initial cluster center is used to determine the mapping result x between the j-th initial cluster center and the i-th clinical entity among the k initial cluster centers. i The Euclid distance is the smallest.

[0065] It is easy to understand from formula (1) that the category corresponding to the j-th initial cluster center is the category of the i-th clinical entity and its mapping result.

[0066] As can be seen from the above, by determining the category to which the nk clinical entities and their mapping results belong, k initial cluster centers, and k categories that serve as initial cluster centers, the nk clinical entities and their mapping results are assigned to the categories of the k initial cluster centers. In other words, the n clinical entities and their mapping results are divided into K categories.

[0067] Step S403: Use the average value of the clinical entities and their mapping results in each category as the second cluster center to determine the category of the second cluster center.

[0068] Step S404: Based on the n clinical entities and their mapping results, the k second cluster centers and the categories of the k second cluster centers, determine the category to which the n clinical entities and their mapping results belong.

[0069] Step S405: Determine the category of the third cluster center by using the average value of the clinical entities and their mapping results in each category as the third cluster center.

[0070] S406: Repeat steps S403 to S405. When the distance between each third cluster center and the corresponding second cluster center is not greater than the first preset value, determine the clinical entities in the category of each third cluster center and their mapping results.

[0071] S407: Determine the distance between each clinical entity and its mapping result in each category of each third cluster center and the corresponding third cluster center. If the distance between the clinical entity and its mapping result and the corresponding third cluster center is greater than the second preset value, remove the clinical entity and its mapping result.

[0072] S408: Based on the removed clinical entities and their mapping results, structured data is obtained.

[0073] From the above, it is easy to understand that by performing clustering calculations on each clinical entity and its mapping results, and based on the clustering calculation results corresponding to each clinical entity and its mapping results, clinical entities and their mapping results with clustering calculation results greater than a second preset value are removed, thereby obtaining structured data. The structured data may include clinical entities and the categories of data models corresponding to the clinical entities.

[0074] It is easy to understand that this application verifies structured data by performing clustering calculations on the clinical entities and their mapping results, and obtains verified structured data. That is, the verified structured data removes clinical entities and their mapping results whose clustering calculation results are greater than a preset value, and retains clinical entities and their mapping results whose clustering calculation results are not greater than a preset value. In other words, the verified structured data has better clustering, thereby making the consistency and accuracy of the verified structured data higher.

[0075] In summary, this application utilizes a deep learning model to identify clinical entities in electronic health record data, effectively improving the efficiency and accuracy of clinical entity recognition. An information extraction algorithm maps the identified clinical entities to a pre-defined data model, obtaining mapping results that adapt to electronic health record data of different formats or in different medical fields, thus resulting in highly consistent structured data.

[0076] In embodiments of this application, a deep learning model is used to identify clinical entities in electronic health record data, thereby obtaining the clinical entities. Specifically, the deep learning model is used to determine the probability of different tags corresponding to each character in the electronic health record data. Based on the probability of different tags corresponding to each character in the electronic health record data, the tag of each character in the electronic health record data is determined, thus determining the tag sequence of the electronic health record data. The clinical entities in the electronic health record data are then determined based on the tag sequence of the electronic health record data.

[0077] The deep learning model of this application is described in detail below.

[0078] In the embodiments of this application, a deep learning model is used to determine the probability of different tags corresponding to each character of the electronic health record data. Based on the probability of different tags corresponding to each character of the electronic health record data, the tag of each character of the electronic health record data is determined, thereby determining the tag sequence of the electronic health record data. Taking BIO annotation as an example, the different tags corresponding to each character include B, I, and O, where B and I together represent entities. B indicates the starting position of a named entity, I indicates the position inside the named entity, and O indicates an entity that is not an entity or is not of the target category. For example, "occasional colic in the lower right abdomen" is labeled as:

[0079] Table 1 Examples of BIO Labeling

[0080] As shown in Figure 4, in one embodiment of this application, the deep learning model may include an embedding layer 101 and a neural network layer 102, wherein the neural network layer 102 includes a feature extraction layer 1021 and a fully connected layer 1022.

[0081] The embedding layer 101 is used to convert each character of the electronic health record data into a character vector and a word vector to which each character belongs, and to concatenate the character vector and the word vector to obtain a concatenated vector.

[0082] Specifically, the embedding layer 101 is used to acquire the input electronic health record data, converting each character of the electronic health record data into a character vector and the word vector to which each character belongs. The method used can be word embeddings, such as word2vec, glovec, fasttext, etc., mapping each character to a real number vector, and these vectors are semantically related.

[0083] The feature extraction layer 1021 is used to determine the spliced ​​feature vector by extracting features from the spliced ​​vector.

[0084] In some embodiments, since medical texts contain a large number of specialized entities such as technical terms and proper nouns, contextual information is required to accurately extract these entities. Therefore, the neural network layer 102 of this application can adopt a bidirectional long short-term memory (BiLSTM) network. The BiLSTM network can fully consider the contextual semantic information of the electronic health record data, avoid the long-distance dependency problem of medical texts, and thus obtain richer information in the spliced ​​feature vectors, thereby improving the accuracy of the fully connected layer 1022 in determining the probability that each character of the electronic health record data corresponds to a different label.

[0085] The fully connected layer 1022 is used to obtain the probability of different tags corresponding to each character of the electronic health record data based on the spliced ​​feature vector.

[0086] Specifically, the fully connected layer 1022 maps the concatenated feature vectors to a probability distribution corresponding to each label through an activation function. Then, it normalizes the probability distribution of different labels corresponding to each character to generate the probability of different labels for each character. Based on the combination of labels with the highest probability for each character, the label sequence corresponding to the electronic health record data can be obtained. Based on the label sequence, the clinical entities identified from the electronic health record data can be obtained.

[0087] It's easy to understand that, based on the probability of each character in the electronic health record data corresponding to different labels determined by the deep learning model, the label with the highest probability value is used as the label for that character, thus determining the label sequence of the electronic health record data. This label sequence is the label for each character in each electronic health record data. Based on the label sequence, the clinical entities identified from the electronic health record data can be obtained. For example, as shown in Table 1, the probabilities of the different labels "B", "I", and "O" corresponding to the character "right" are 0.8, 0.12, and 0.08, respectively. Therefore, the label with the highest probability corresponding to the character "right" is "B". Similarly, the label sequence with the highest probability for the electronic health record data "occasional colic in the right lower abdomen" is "BIIOOBI". Therefore, this label sequence contains two clinical entities: the right lower abdomen corresponding to BII, and the colic corresponding to BI.

[0088] In the embodiments of this application, the BiLSTM network is a variant of the Long Short-Term Memory (LSTM) network, which runs two LSTMs simultaneously on the time series, one processing from front to back and the other from back to front.

[0089] Figure 5 shows a recurrent cell of an LSTM. A gating mechanism is used to update the information in the cell state c to control information transmission and avoid long-distance dependencies. Specifically, the forget gate uses a gating function to determine the cell state c of the previous node. t-1 For the current cell state c t The importance of c t-1 Forgot some of the information in the middle and incorporated it into c t In the state vector, the input gate corresponds to the current input information x. t Perform selective memorization and selectively add c t In the middle, the output gate determines the cell state c of the current node. t How much should be output to the current hidden state h? t LSTM uses the current input x tand the hidden state h from the previous moment t-1 As input to train the control signals of each gate, the hidden state h is updated. t And obtain the current output y t The calculation formulas are as shown in formulas (2) to (4), where W and b are the parameters that need to be learned. y t =o t ⊙tanh(c t (4)

[0090] As can be seen from the above, although neural network layer 102 can find the label with the highest probability corresponding to each character, in reality, directly determining the label of each character based on the probability of different labels corresponding to each character in the electronic health record data determined by neural network layer 102 is not ideal. This is because, although feature extraction layer 1021 can learn the dependencies between observed sequences through bidirectional settings, the probability outputs obtained by the normalization process of fully connected layer 1022 are independent of each other, and the outputs do not affect each other. Such a model cannot learn the transition dependencies between output labels (the probability transition matrix of the labels) and the constraints of sequence labeling, such as the sentence should begin with "B" or "O" instead of "I".

[0091] As shown in Figure 6, in another embodiment of this application, the deep learning model may include: an embedding layer 201, a neural network layer 202, and a constraint layer 203, wherein the neural network layer 202 includes a feature extraction layer 2021 and a fully connected layer 2022. The contents of the embedding layer 201 and the neural network layer 202 are the same as those of the embedding layer 101 and the neural network layer 102 described above, and will not be repeated here.

[0092] Constraint layer 203 is used to process the probabilities of different labels corresponding to each character by setting constraints, resulting in optimized probabilities of different labels corresponding to each character. The constraints are represented by the learned features, which characterize the relationship between the labels. As mentioned earlier, this allows us to obtain the optimized clinical entities identified from the electronic health record data.

[0093] Specifically, the constraints are the constraints for learning sequence labeling. The probability of different labels corresponding to each character output by neural network layer 202 is used as the input of constraint layer 103. By setting constraints, the maximum possible probability of the predicted sequence label that meets the sequence labeling transition constraints is output.

[0094] The constraint layer 203 can further improve the accuracy of the output results based on the learned constraints. For example: (1) Let B represent the starting boundary of an entity name. The label of "right" in "right lower abdomen" should be "B" and not "I". (2) The attributes of the labels corresponding to the same entity are the same. The label of "right lower abdomen" is "B-body I-body I-body", and the attribute is "body". It cannot be "B-body I-disease I-body" or others. The constraint layer 203 uses the learned features to represent the constraint relationship between the labels and uses the state feature function and the transition feature function to evaluate the label sequence. Assume that the given input sequence is: X = (x1, x2, ..., x n )

[0095] The corresponding output sequence is: Y = (y1, y2, ..., y3)

[0096] Define the scoring function as:

[0097] Where F(X, Y) represents the probability score that the entity label sequence corresponding to the input text sequence X is Y. This means that the i-th character is marked as label y. i The probability. N is the transition matrix. Indicates label y i The next label is y i+1 The probability of an entity's label is determined by the influence of context, external dictionary, and rules on the current entity label. Finding the highest probability score yields the optimal output label sequence.

[0098] As can be easily understood from the above, adding a constraint layer 203 to the neural network model allows the constraint layer 203 to process the probability of different labels corresponding to each character by setting constraints. It uses state feature functions and transition feature functions to evaluate the label sequence, obtaining optimized probabilities of different labels for each character. This ensures that the output label probability of each character fully considers the transition dependencies in the electronic health record data (i.e., the dependencies between preceding and following characters, the dependencies between sentences, etc.). Based on the optimized probabilities of different labels for each character, the label sequence of the electronic health record data is determined, resulting in higher accuracy of the output label sequence or clinical entities. This, in turn, improves the accuracy of the neural network model in recognizing clinical entities in the electronic health record data, thus ensuring the accuracy of the final structured data.

[0099] As shown in Figure 7, an electronic health record structured processing system is used to execute the above-described electronic health record structured processing method. The system includes: an acquisition module, a natural language processing module 1, an information extraction and mapping module 2, and a quality control module 3.

[0100] The acquisition module is used to acquire electronic health record data.

[0101] Natural Language Processing Module 1 is used to identify clinical entities in electronic health record data using a deep learning model to obtain clinical entities. The deep learning model is trained using an entity dataset corresponding to a medical professional dictionary.

[0102] The information extraction and mapping module 2 is used to map the clinical entity to a preset data model using an information extraction algorithm to obtain a mapping result. Based on the clinical entity and the mapping result, structured data is obtained, and the obtained structured data is transmitted to the quality control module 3. The mapping result includes the category of the data model to which the clinical entity belongs.

[0103] The quality control module 3 is used to verify the clinical entity and the mapping result through an anomaly detection algorithm to obtain verified structured data.

[0104] In other embodiments, the electronic health record structured processing system may further include: a data integration and storage module 4, an API and data access module 5, a user interface and interaction module 6, and a system maintenance and optimization module 7.

[0105] The data integration and storage module 4 integrates structured data verified by the quality control module 3 into the central database, enabling unified data management and efficient access. The database can employ a hybrid storage scheme combining relational and NoSQL databases to optimize data storage structure and improve query efficiency. This hybrid storage scheme includes a storage layer, a communication layer, and an application layer. The storage layer comprises SQL and NoSQL storage; SQL storage primarily handles structured data storage and processing, while NoSQL storage handles unstructured data storage and processing. The communication layer handles real-time synchronization and asynchronous transmission of data, ensuring eventual consistency across different database layers. The application layer implements business logic and external interfaces, interacting with the storage layer through a unified data access layer.

[0106] API and Data Access Module 5: This module is used to authenticate and authorize external requests, providing data services to those requests that pass authentication and authorization. For example, when healthcare professionals need to access patient data, they need to verify whether the doctor is the patient's attending physician. Similarly, when ordinary users need to access their own data, they need to verify whether they are the patient themselves. Otherwise, data access will be prohibited or restricted, ensuring data access security. It also provides data caching and load balancing to optimize performance.

[0107] The user interface and interaction module 6 is used to display structured data based on external requests and to review and analyze the requested structured data. For example, it displays analysis results through charts and reports, supports custom queries and visualizations, and enhances the user experience.

[0108] The system maintenance and optimization module 7 is used to collect the operation data and user feedback of the electronic health record structured processing system through log management and performance monitoring tools, and to adjust the configuration and optimize the algorithm of the electronic health record structured processing system.

[0109] It should be understood that the various embodiments of the present invention can be implemented in hardware, software, or a combination thereof. Software can be stored in volatile or non-volatile storage devices such as ROM (regardless of whether the data is erasable or rewritable), in memories such as RAM, memory chips, devices, or integrated circuits, or in storage media on which data can be optically or magnetically recorded and read by a machine (e.g., a computer), such as compressed optical discs (CDs), digital versatile optical discs (DVDs), magnetic disks, or magnetic tapes. Furthermore, the various embodiments of the present invention can be implemented in a computer or portable terminal having a controller and memory, and memory is an example of a non-transitory computer-readable storage medium suitable for storing one or more programs including commands for implementing the various embodiments of the present invention. Therefore, the present invention includes a program having code for implementing the means or methods defined in the claims, and a machine-readable non-transitory storage medium storing the program. The program can be electrically transferred via a medium such as communication signals transmitted via wired or wireless connections, and the program and its equivalents are included in the present invention.

[0110] The device can receive and store a program from a program provider connected via a cable or wireless connection. The program provider may include a program containing commands for implementing various embodiments of the invention, a memory for storing information required for various embodiments of the invention, a communication module for communicating with a mobile device via a wired or wireless connection, and a controller for automatically or upon request from the device to transmit the program to the device.

[0111] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for structuring electronic health records, characterized in that, The method includes: Obtain electronic health record data; A deep learning model is used to identify clinical entities in the electronic health record data to obtain clinical entities, wherein the deep learning model is trained on an entity dataset corresponding to a medical professional dictionary; The clinical entities are mapped to a preset data model using an information extraction algorithm to obtain mapping results. Structured data is obtained based on the clinical entities and the mapping results, wherein the mapping results include the category of the data model to which the clinical entities belong. The clinical entities and the mapping results are verified using an anomaly detection algorithm to obtain verified structured data.

2. The electronic health record structuring processing method according to claim 1, characterized in that, The method of using a deep learning model to identify clinical entities in the electronic health record data includes: Each character of the electronic health record data is converted into a character vector and the word vector to which each character belongs. The character vector and the word vector are then concatenated to obtain a concatenated vector. By extracting features from the concatenated vector, a concatenated feature vector is determined, and the probability of each character corresponding to a different label is obtained based on the concatenated feature vector. The tag sequence of the electronic health record data is determined based on the probability that each character corresponds to a different tag; Clinical entities in the electronic health record data are determined based on the tag sequence of the electronic health record data.

3. The electronic health record structuring processing method according to claim 1, characterized in that, The method of using a deep learning model to identify clinical entities in the electronic health record data includes: Each character of the electronic health record data is converted into a character vector and the word vector to which each character belongs. The character vector and the word vector are then concatenated to obtain a concatenated vector. By extracting features from the concatenated vector, a concatenated feature vector is determined, and the probability of each character corresponding to a different label is obtained based on the concatenated feature vector. By setting constraints, the probability of different labels corresponding to each character is processed to obtain the optimized probability of different labels corresponding to each character; The label sequence of electronic health record data is determined based on the probability of each character corresponding to a different label after optimization. Clinical entities in the electronic health record data are determined based on the tag sequence of the electronic health record data.

4. The electronic health record structuring processing method according to claim 1, characterized in that, The process of mapping the clinical entity to a preset data model using an information extraction algorithm to obtain the mapping result includes: The information extraction algorithm is used to extract entity attributes from the clinical entity to obtain the entity attributes of the clinical entity. Based on the preset data model category and the entity attributes of the clinical entity, the clinical entity is mapped to the corresponding category data model.

5. The electronic health record structuring processing method according to claim 4, characterized in that, The information extraction algorithm described above is used to extract entity attributes from the clinical entity, resulting in the following entity attributes: Shift-based dependency analysis is used to perform dependency relationship analysis on input clinical entities to identify core terms in electronic health record data. A recursive search is used to find all words that are parallel to the core word, forming a predicate list; For each predicate, find its subject and object based on the dependency relationship to form an information tuple; For an information tuple without a subject, find a parent tuple that is parallel to its predicate and has a subject. If the parent tuple exists, set the subject of the information tuple as the subject of the parent tuple. If no such parent tuple is found, do not process it. Return the extraction results to obtain the entity attributes of the clinical entity.

6. The electronic health record structuring processing method according to claim 4, characterized in that, The preset data model includes at least one of the following: disease-drug data model, traditional Chinese medicine diagnosis and treatment data model, and Western medicine diagnosis and treatment data model; Correspondingly, the categories of the preset data models include at least one of the following: the category of the disease-drug data model is drug research, the category of the traditional Chinese medicine diagnosis and treatment data model is traditional Chinese medicine diagnosis and treatment, and the category of the Western medicine diagnosis and treatment data model is Western medicine diagnosis and treatment.

7. The electronic health record structuring processing method according to claim 1, characterized in that, The verification of the clinical entity and the mapping result using an anomaly detection algorithm yields the following verified structured data: Clustering calculations are performed on the clinical entities and their mapping results. Based on the clustering calculation results corresponding to the clinical entities and their mapping results, clinical entities and their mapping results with clustering calculation results greater than a preset value are removed to obtain the validated structured data.

8. A structured processing system for electronic health records, characterized in that, include: The acquisition module is used to acquire electronic health record data; The natural language processing module is used to identify clinical entities in the electronic health record data using a deep learning model, thereby obtaining the clinical entities. The deep learning model utilizes an entity dataset corresponding to a medical professional dictionary. Training obtained; The information extraction and mapping module is used to map the clinical entity to a preset data model through an information extraction algorithm to obtain a mapping result, and to obtain structured data based on the clinical entity and the mapping result, wherein the mapping result includes the category of the data model to which the clinical entity belongs; The quality control module is used to verify the clinical entities and the mapping results through anomaly detection algorithms to obtain verified structured data.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the electronic health record structuring processing method according to any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the electronic health record structuring processing method according to any one of claims 1 to 7.

Images