Named entity recognition method and system for chief complaint data of coronary heart disease patient and medium

Abstract

The invention discloses a named entity identification method and system for coronary heart disease patient chief complaint data and a medium, and belongs to the field of medical information processing.The method comprises the steps that according to a preset named entity identification rule, information related to the coronary heart disease in a coronary heart disease patient chief complaint data set is mapped to a corresponding named entity category, and a named entity label set is generated, generating a clinical classification set corresponding to the named entity label set; taking the coronary heart disease patient chief complaint data set as input and the named entity label set as a label to train a coronary heart disease classification neural network model; taking the named entity label set as input and the clinical classification set as output, and training a chief complaint named entity classification network; and inputting to-be-identified coronary heart disease patient chief complaint data into the trained coronary heart disease classification neural network model, and outputting corresponding clinical classification data by the chief complaint named entity classification network. According to the method, valuable medical information can be effectively extracted from the chief complaint of the coronary heart disease patient, and support is provided for clinical decision and research.

Description

Technical Field

[0001] The present invention belongs to the field of medical information processing, and more specifically, relates to a method, system and medium for named entity recognition of chief complaint data of patients with coronary heart disease. Background Art

[0002] The prevalence and lethality of coronary heart disease highlight the urgency of rapid and accurate diagnosis. In the diagnosis of coronary heart disease, the chief complaint information of coronary heart disease patients plays a vital role, providing doctors with a basis for constructing the patient's initial symptoms and health status. Automated processing of the chief complaint information of coronary heart disease patients greatly accelerates the clinical decision-making process, which is particularly critical when dealing with high-risk cardiovascular events, and improves the overall quality and efficiency of medical services.

[0003] However, existing technologies face multiple challenges in automatically processing the chief complaint information of patients with coronary heart disease. On the one hand, the chief complaint information of patients with coronary heart disease mostly exists in the form of free text and lacks a standardized format, which increases the difficulty of data processing; natural language processing technology for unstructured text requires highly accurate and complex algorithms to effectively extract key information and convert it into useful medical data. On the other hand, medical texts contain rich professional terms and complex sentence structures, often involving multiple levels of medical knowledge. In the absence of sufficient medical background, it becomes particularly difficult to automatically process and understand this information.

[0004] Although a variety of technologies have been developed to automatically identify and process text data, these technologies often have difficulty accurately handling the specific needs of the medical field. Existing text analysis tools may not be able to accurately identify complex symptom descriptions related to coronary heart disease or correctly interpret the subtle differences in medical semantics. Traditional keyword search and manual annotation methods, although simple and direct, often lack flexibility and accuracy when processing complex clinical texts. Keyword search cannot fully understand the context and capture the subtle differences in semantics, while manual annotation is time-consuming and depends on the availability and consistency of professionals. Although primary automated recognition tools based on statistics and frequency have improved the processing speed of clinical complaints, they often perform poorly in terms of accuracy and adaptability, especially when faced with the complex complaints of patients with coronary heart disease. Therefore, it is of great significance to develop a method that can automatically identify the complaint data of patients with coronary heart disease. Summary of the invention

[0005] In view of the defects of the prior art and the need for improvement, the present invention provides a method, system and medium for named entity recognition of the chief complaint data of patients with coronary heart disease, which aims to solve the problems of low recognition accuracy and poor adaptability of the existing named entity recognition methods for the chief complaint data of patients with coronary heart disease.

[0006] To achieve the above-mentioned purpose, according to one aspect of the present invention, a method for named entity recognition for the chief complaint data of patients with coronary heart disease is provided, comprising a model training stage S1-S3 and a recognition stage S4: S1, obtaining the original chief complaint data set of patients with coronary heart disease; S2, according to a preset named entity recognition rule, mapping the information related to coronary heart disease in the chief complaint data set of patients with coronary heart disease to the corresponding named entity category to generate a named entity label set, and generating a clinical classification set corresponding to the named entity label set; S3, training a coronary heart disease classification neural network model with the chief complaint data set of patients with coronary heart disease as input and the named entity label set as labels; training a chief complaint named entity classification network with the named entity label set as input and the clinical classification set as output; wherein the output of the coronary heart disease classification neural network model is connected to the input of the chief complaint named entity classification network; S4, inputting the chief complaint data of patients with coronary heart disease to be identified into the trained coronary heart disease classification neural network model, and the chief complaint named entity classification network outputs corresponding clinical classification data.

[0007] Furthermore, the coronary heart disease classification neural network model includes a Macbert encoding unit, an attention model unit and a Span decoding unit connected in sequence; the Macbert encoding unit is used to convert the input coronary heart disease patient's chief complaint data into a feature vector; the Span decoding unit is used to mark the starting position and the ending position of the entity to predict the category and boundary of the entity in the feature vector.

[0008] Furthermore, when training the coronary heart disease classification neural network model, the loss function used includes the following three parts: a starting loss, which is used to calculate the error between the predicted entity starting position and the actual starting position; an ending loss, which is used to calculate the error between the predicted entity ending position and the actual ending position; and an entity classification loss, which is used to calculate the error between the predicted classification and the actual classification.

[0009] Furthermore, the main complaint named entity classification network includes a text preprocessing unit, a word embedding layer, a semantic extraction layer, a convolutional attention layer and an output classification layer connected in sequence; the word embedding layer adopts a pre-trained BERT model, which is used to segment the text output by the text preprocessing unit and convert it into a word meaning embedding vector, a position embedding vector and a sequence embedding vector of a text sequence.

[0010] Furthermore, when training the main complaint named entity classification network, the training is performed with the convergence of the cross entropy loss function of the main complaint named entity classification network as a goal.

[0011] Furthermore, the S1 specifically includes: collecting medical records of coronary heart disease, performing data cleaning on the medical records of coronary heart disease to delete duplicate medical records and medical records irrelevant to coronary heart disease, and obtaining an original data set of chief complaints of patients with coronary heart disease.

[0012] Furthermore, the clinical classification data includes entity classification, urgency, and response level.

[0013] According to another aspect of the present invention, a named entity recognition system for the chief complaint data of patients with coronary heart disease is provided, comprising: a processor; a memory storing a computer executable program, wherein when the program is executed by the processor, the processor executes the named entity recognition method for the chief complaint data of patients with coronary heart disease as described above.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method for named entity recognition for the chief complaint data of patients with coronary heart disease as described above is implemented.

[0015] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0016] (1) A named entity recognition method for the chief complaint data of patients with coronary heart disease is provided. Through automated and intelligent named entity recognition, key information can be quickly extracted from the chief complaint of patients with coronary heart disease, greatly shortening the diagnosis time. At the same time, accurate information processing reduces human errors and improves the accuracy of diagnosis.

[0017] Furthermore, the corresponding clinical classification data output by the chief complaint named entity classification network helps to identify early signs of coronary heart disease and potential health risk factors, systematically classifies and quantifies named entities, provides detailed and easy-to-understand data, and makes more accurate clinical decisions;

[0018] (2) A specific neural network model for coronary heart disease classification is provided, which adopts the Macbert-Span model. The model training uses three parts: starting loss, ending loss and entity classification loss. The combination of these three losses optimizes the model's ability to recognize entity boundaries and categories, thereby improving the overall named entity recognition performance;

[0019] (3) A specific main complaint named entity classification network is provided. This model can address the limitations of traditional models in dealing with data imbalance and multiple words with the same meaning. It uses the extensive context representation capabilities of the pre-trained model to perform data enhancement and improve the processing effect on minority class samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of a method for named entity recognition for the main complaint data of patients with coronary heart disease provided by an embodiment of the present invention;

[0021] Figure 2A diagram of the process of constructing a named entity tag set provided by an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of the architecture of a neural network model for coronary heart disease classification provided by an embodiment of the present invention;

[0023] Figure 4 A schematic diagram of the working principle of the main complaint named entity classification network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0025] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the accompanying drawings are used to distinguish similar objects but not necessarily to describe a specific order or sequence.

[0026] Embodiment 1

[0027] A named entity recognition method for the chief complaint data of patients with coronary heart disease. Figure 1-Figure 4 , a method for named entity recognition for the chief complaint data of patients with coronary heart disease in this embodiment is described in detail, and the method includes a model training stage (operation S1-operation S3) and a recognition stage (operation S4).

[0028] Operation S1, obtaining an original dataset of chief complaints of patients with coronary heart disease.

[0029] According to an embodiment of the present invention, operation S1 specifically includes: collecting medical records of coronary heart disease, performing data cleaning on the medical records of coronary heart disease to delete duplicate medical records and medical records irrelevant to coronary heart disease, and obtaining an original data set of chief complaints of patients with coronary heart disease.

[0030] Specifically, for example, medical record data is automatically extracted from the hospital's electronic health record system, including the patient's personal information, symptom description, diagnosis results, treatment history, and discharge status. Through an intelligent screening algorithm, the medical record data is preliminarily classified according to the discharge diagnosis, and the medical records diagnosed with coronary heart disease are automatically marked, and the medical record data of coronary heart disease is accurately extracted. At the same time, data cleaning is performed to delete duplicate medical records and eliminate medical records that are not related to coronary heart disease, such as irrelevant medical history and non-critical personal information. Data cleaning is automatically completed through predefined text processing scripts, which are specially designed to identify and process redundant patterns in medical texts. Finally, the classified and cleaned data is compiled into an unprocessed dataset of coronary heart disease patients' chief complaints, providing a basis for named entity recognition and subsequent in-depth analysis.

[0031] Operation S2, according to preset named entity recognition rules, maps the information related to coronary heart disease in the coronary heart disease patient complaint data set to corresponding named entity categories to generate a named entity label set, and generates a clinical classification set corresponding to the named entity label set.

[0032] The preset named entity recognition rules are designed based on the diagnosis and treatment experience of frontline clinicians and the clinical characteristics of coronary heart disease. The preset named entity recognition rules are implemented by designing an algorithm to automatically extract relevant information from the original coronary heart disease medical record text and map it to predefined named entity categories. To achieve this, the algorithm first uses natural language processing technology to segment words, and then applies BIO annotation to each word or phrase according to the preset named entity recognition rules to identify the named entities in the text.

[0033] Specifically, this algorithm first breaks down text data into individual words or phrases and labels them as "B" (beginning), "I" (internal), or "O" (external) according to their roles in the clinical context. For example, for the symptom description "chest pain for three weeks", "chest" may be labeled as "B-symptom", "pain" as "I-symptom", and the remaining irrelevant word "three weeks" is labeled as "O", such as Figure 2 shown.

[0034] In addition, in order to optimize the accuracy of annotation, multiple rounds of verification processes are set in this embodiment to review the results of automatic annotation to ensure the medical accuracy and clinical relevance of the annotation. After the above operations, a named entity label dataset of the main complaints of patients with coronary heart disease is finally generated.

[0035] In this embodiment, the clinical classification data includes entity classification, urgency, and response level.

[0036] Operation S3, using the coronary heart disease patient chief complaint dataset as input and the named entity label set as labels, trains the coronary heart disease classification neural network model; using the named entity label set as input and the clinical classification set as output, trains the chief complaint named entity classification network; wherein, the output of the coronary heart disease classification neural network model is connected to the input of the chief complaint named entity classification network.

[0037] According to an embodiment of the present invention, the coronary heart disease classification neural network model adopts the "Macbert-Span" named entity recognition model, which adopts a double sequence labeling strategy, including a Macbert encoding unit, an attention model unit, and a Span decoding unit connected in sequence, such as Figure 3 As shown in the figure. The Macbert encoding unit is used to convert the input coronary heart disease patient complaint data into a feature vector. The Macbert model uses deep learning technology to understand the complex medical terms and context in the text and converts the text into a Token word vector, which can well implement subsequent entity recognition. The Span decoding unit is used to mark the starting position (start_index) and ending position (end_index) of the entity to predict the category and boundary of the entity in the feature vector.

[0038] According to an embodiment of the present invention, when training a neural network model for coronary heart disease classification, the loss function used includes the following three parts: starting loss, which is used to calculate the error between the predicted entity starting position and the actual starting position; ending loss, which is used to calculate the error between the predicted entity ending position and the actual ending position; entity classification loss, which is used to calculate the error between the predicted classification and the actual classification. The combination of these three losses optimizes the model's ability to recognize entity boundaries and categories, thereby improving the overall named entity recognition performance; the four formulas shown below correspond to the losses of the three parts of starting loss, ending loss, and entity classification loss, and the fourth is the overall loss:

[0039] start_Loss=CE(Prestart,TTruestart)

[0040] end_Loss=CE(Preend,Trueend);

[0041] span_Loss=CE(Prestart,end,Truestart,end);

[0042] Loss=start_Loss+end_Loss+span_Loss.

[0043] In this embodiment, a neural network model (coronary heart disease classification neural network model) designed specifically for token-level classification is customized based on the BERT architecture to process more fine-grained text data. The model introduces the BERT model as the main feature extractor, combined with a top-level linear classifier, to convert text into an embedding vector and ultimately map it to a predefined named entity category.

[0044] During the model initialization phase, the " c hines e -robert a - www - e xt" pre-trained model and made necessary configuration adjustments to it to meet the recognition needs of coronary heart disease named entities. Afterwards, the model uses a dropout layer to prevent overfitting and enhance the model's generalization ability on new data. By pre-training on an open source medical nomenclature dataset, the model acquires preliminary medical language understanding capabilities. Afterwards, further training is performed on a specially constructed named entity label set. This training enables the model to accurately identify and classify entities related to coronary heart disease, such as symptoms, diseases, and other key information.

[0045] According to an embodiment of the present invention, the main complaint named entity classification network includes a text preprocessing unit, a word embedding layer, a semantic extraction layer, a convolutional attention layer and an output classification layer connected in sequence. The model can solve the limitations of traditional models in processing data imbalance and multiple words with the same meaning, and use the extensive context representation ability of the pre-trained model to perform data enhancement and improve the processing effect on minority class samples. The working principle of the main complaint named entity classification network is as follows: Figure 4 shown.

[0046] The text preprocessing unit performs basic text processing, including removing stop words, performing Chinese word segmentation, and removing noise, in order to construct a clear text input. The processed text is converted into a set of independent words and padded to a uniform length of 32 for subsequent processing.

[0047] The word embedding layer uses the pre-trained BERT model to segment the text output by the text preprocessing unit and convert it into word meaning embedding vectors, position embedding vectors, and sequence embedding vectors of the text sequence. These vectors together constitute a comprehensive representation of the input text, including word meaning, sentence position, and context information.

[0048] The semantic extraction layer uses a Transformer structure that includes a multi-head self-attention mechanism and a feedforward neural network. The design of this layer enables the model to dynamically capture the features of different parts of the input sequence, thereby better understanding and extracting the semantic information of the text.

[0049] The convolutional attention layer uses a convolutional neural network to further extract the features encoded from the Transformer structure. By using convolution kernels of different sizes (2, 3, 4 word window sizes), the model can capture the features of continuous phrases, strengthen the nonlinear representation through the relu activation function, and use the pooling operation to retain the most important features.

[0050] The output classification layer is concatenated through the output of multiple convolutional layers and then sent to a linear fully connected layer for the final classification of the text. This layer will calculate the probability of each classification and select the classification with the highest probability as the prediction result.

[0051] According to an embodiment of the present invention, when training the main complaint named entity classification network, the training is performed with the cross entropy loss function of the main complaint named entity classification network converging as the goal. The cross entropy loss function is used to measure the difference between the probability distribution predicted by the model and the probability distribution of the actual label, thereby optimizing the model parameters and improving the accuracy.

[0052] Operation S4, inputting the chief complaint data of the coronary heart disease patient to be identified into the trained coronary heart disease classification neural network model, and the chief complaint named entity classification network outputs the corresponding clinical classification data.

[0053] After the CHD classification neural network model identifies CHD-related named entities such as symptoms, diseases and other key information, the identified information is input into the chief complaint named entity classification network for subsequent classification and quantification.

[0054] The Chief Complaint Named Entity Classification Network includes a multi-layer neural network structure that is capable of processing embedding vectors extracted from the BERT model. The design of the network allows it to recognize and understand the clinical relevance of each entity, such as the severity of symptoms, treatment response, etc. For each named entity identified by the CHD Classification Neural Network Model, the network classifies it according to predefined clinical significance criteria. For example, symptoms such as angina pectoris and chest pain are classified and quantified in terms of urgency and the level of response required. In addition to classification, the network also quantifies entities to provide a numerical basis for subsequent clinical decisions.

[0055] The named entity recognition method for the chief complaint data of patients with coronary heart disease provided by the embodiment of the present invention combines medical expertise and advanced natural language processing technology, uses a deep learning model, learns the expression pattern of complex Chinese medical nomenclature from large-scale public medical data, improves the accuracy and efficiency of medical named entity recognition, and better captures the semantic complexity and context dependence in clinical texts; relies on the label data set of the chief complaint nomenclature of patients with coronary heart disease to improve the directionality of the naming recognition network; through automated and intelligent named entity recognition, key information can be quickly extracted from the patient's chief complaint, shortening the diagnosis time, while reducing human errors and improving the overall diagnostic accuracy of coronary heart disease. This method can structure a large amount of unstructured medical text data, providing rich data resources for big data medical research. In summary, this method, by combining advanced natural language processing technology and deep learning, not only improves the processing method of the chief complaint information of patients with coronary heart disease, brings two-way benefits to clinical and research, but also shows the advantages of migration application in other disease fields.

[0056] Embodiment 2

[0057] A named entity recognition system for the main complaint data of patients with coronary heart disease comprises: a processor; a memory storing a computer executable program, which, when executed by the processor, enables the processor to execute the above-mentioned named entity recognition method for the main complaint data of patients with coronary heart disease. The related technical scheme is the same as that of the first embodiment, and will not be described in detail here.

[0058] Embodiment 3

[0059] A computer-readable storage medium stores a computer program, and when the program is executed by a processor, the method for identifying named entities for the main complaint data of patients with coronary heart disease is implemented. The related technical solution is the same as that of the first embodiment, and will not be described in detail here.

[0060] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A named entity recognition method for the main complaint data of coronary heart disease patients, characterized in that: Including model training phase S1-S3 and recognition phase S4: S1, obtain the original dataset of patients’ complaints of coronary heart disease; S2, mapping the information related to coronary heart disease in the coronary heart disease patient complaint data set to corresponding named entity categories according to preset named entity recognition rules to generate a named entity label set, and generating a clinical classification set corresponding to the named entity label set; S3, using the coronary heart disease patient chief complaint dataset as input and the named entity label set as labels to train a coronary heart disease classification neural network model; using the named entity label set as input and the clinical classification set as output to train a chief complaint named entity classification network; wherein the output of the coronary heart disease classification neural network model is connected to the input of the chief complaint named entity classification network; S4, inputting the chief complaint data of the coronary heart disease patient to be identified into the trained coronary heart disease classification neural network model, and the chief complaint named entity classification network outputs corresponding clinical classification data.

2. The method for named entity recognition for the main complaint data of patients with coronary heart disease according to claim 1, characterized in that: The coronary heart disease classification neural network model includes a Macbert encoding unit, an attention model unit and a Span decoding unit connected in sequence; The Macbert encoding unit is used to convert the input coronary heart disease patient's main complaint data into a feature vector; The Span decoding unit is used to mark the starting position and the ending position of the entity to predict the category and boundary of the entity in the feature vector.

3. The method for named entity recognition for the main complaint data of patients with coronary heart disease according to claim 1 or 2, characterized in that: When training the coronary heart disease classification neural network model, the loss function used includes the following three parts: Starting loss, which is used to calculate the error between the predicted starting position of the entity and the true starting position; Termination loss, which is used to calculate the error between the predicted entity termination position and the actual termination position; Entity classification loss is used to calculate the error between the predicted classification and the true classification.

4. The method for named entity recognition for the main complaint data of patients with coronary heart disease according to claim 1, characterized in that: The main complaint named entity classification network includes a text preprocessing unit, a word embedding layer, a semantic extraction layer, a convolutional attention layer and an output classification layer connected in sequence; The word embedding layer adopts a pre-trained BERT model, which is used to segment the text output by the text preprocessing unit and convert it into a word meaning embedding vector, a position embedding vector and a sequence embedding vector of a text sequence.

5. The method for named entity recognition for the main complaint data of patients with coronary heart disease according to claim 1 or 4, characterized in that: When training the main complaint named entity classification network, the training is performed with the convergence of the cross entropy loss function of the main complaint named entity classification network as a goal.

6. The method for named entity recognition for the main complaint data of coronary heart disease patients according to claim 1, characterized in that: The S1 specifically includes: Medical records of coronary heart disease are collected, and data cleaning is performed on the medical records of coronary heart disease to delete duplicate medical records and medical records irrelevant to coronary heart disease, so as to obtain an original data set of chief complaints of patients with coronary heart disease.

7. The method for named entity recognition for the main complaint data of coronary heart disease patients according to claim 1, characterized in that: The clinical classification data includes entity classification, urgency, and response level.

8. A named entity recognition system for the main complaint data of patients with coronary heart disease, characterized in that: include: processor; A memory storing a computer executable program, wherein when the program is executed by the processor, the processor executes the named entity recognition method for the main complaint data of coronary heart disease patients according to any one of claims 1 to 7.

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 for named entity recognition for the chief complaint data of patients with coronary heart disease according to any one of claims 1 to 7 is implemented.

Images