Infectious disease data intelligent coding method, device, computer equipment and storage medium

Abstract

Embodiments of the present invention disclose a method, apparatus, computer device, and storage medium for intelligent encoding of infectious disease data. The method comprises: obtaining medical text data related to infectious diseases; extracting local features from the medical text data; constructing a clinical medical knowledge graph and modeling the encoding relationship to obtain a modeling result; fusing the local features with the modeling result to obtain a feature fusion result; inputting the feature fusion result into a coding decision model for encoding to obtain an ICD coding result; and outputting the ICD coding result. Implementing the method of the embodiment of the present invention can improve the feature extraction and coding prediction process of infectious disease data, thereby enhancing the accuracy and efficiency of coding.

Description

Technical Field

[0001] The present invention relates to a medical information processing method, and more specifically to an infectious disease data intelligent coding method, device, computer equipment and storage medium. Background Art

[0002] In modern hospital management, accurate coding of infectious diseases is crucial for disease monitoring, treatment analysis, and medical insurance settlement. With the widespread adoption of electronic medical record systems, a vast amount of unstructured medical data has emerged, including medical records, laboratory results, and clinical diagnostic information. However, manual coding of this information is not only time-consuming but also prone to errors, leading to an increasingly urgent need for automated coding.

[0003] Traditional ICD (International Classification of Diseases) coding methods rely on manual judgment, influenced by physician experience and the quality of data entry, which can lead to inaccurate or missing codes. Although existing automatic coding technologies have made progress, they still face challenges in feature extraction and context understanding, and modeling coding relationships is often imprecise. While current technologies, such as multi-scale convolutional neural networks and graph convolutional networks, can process text features and coding relationships, they are limited in their ability to process long texts and consume a lot of computing resources.

[0004] Therefore, it is necessary to design a new method to improve the feature extraction and coding prediction process of infectious disease data to enhance the accuracy and efficiency of coding. Summary of the Invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method, device, computer equipment and storage medium for intelligent encoding of infectious disease data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent encoding method for infectious disease data, comprising:

[0007] Obtain medical text data related to infectious diseases;

[0008] Extracting local features from the medical text data;

[0009] Construct a clinical medical knowledge graph and model the coding relationship to obtain modeling results;

[0010] Fusing the local features with the modeling results to obtain a feature fusion result;

[0011] Inputting the feature fusion result into the coding decision model for coding to obtain an ICD coding result;

[0012] Output the ICD encoding result.

[0013] A further technical solution is: extracting local features from the medical text data includes:

[0014] Performing text decomposition and embedding on the medical text data to obtain an embedding matrix;

[0015] Performing a convolution operation on the embedding matrix to obtain a convolution result;

[0016] Performing a maximum pooling layer processing on the convolution result to obtain a pooling result;

[0017] The pooling results are subjected to feature map splicing and full connection processing to obtain local features.

[0018] A further technical solution is: constructing a clinical medical knowledge graph and modeling the coding relationship to obtain a modeling result, including:

[0019] Construct a clinical medical knowledge graph and use graph neural network to model the encoding relationship to obtain the modeling results.

[0020] A further technical solution is: constructing a clinical medical knowledge graph and using a graph neural network to model the coding relationship to obtain a modeling result, including:

[0021] Construct a clinical medical knowledge graph;

[0022] Performing multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation;

[0023] Extracting global features using a pooling operation on the node embedding representation;

[0024] Calculating similarities between nodes based on the global feature representation to determine dependencies between the nodes;

[0025] The node embedding representations and the dependencies between nodes are combined to obtain a modeling result.

[0026] A further technical solution is: performing multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation, including:

[0027] Aggregating information of neighboring nodes in the clinical medical knowledge graph and calculating a weighted sum of features of each node;

[0028] Apply the weight matrix and bias term of the graph convolution layer and use the ReLU activation function for nonlinear transformation to obtain the node embedding representation.

[0029] A further technical solution is: fusing the local features with the modeling results to obtain a feature fusion result, including:

[0030] Concatenating the local features with the modeling results to obtain a high-dimensional feature vector;

[0031] Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a dimensionality reduction result;

[0032] The dimensionality reduction result is input into the ensemble learning model for feature fusion to obtain a feature fusion result.

[0033] A further technical solution is: inputting the feature fusion result into a coding decision model for coding to obtain an ICD coding result, including:

[0034] Inputting the feature fusion result into the coding decision model for coding, and using fuzzy logic to determine the ICD coding result;

[0035] The coding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

[0036] The present invention also provides an intelligent encoding device for infectious disease data, comprising:

[0037] A data acquisition unit, used for acquiring medical text data related to infectious diseases;

[0038] A local feature extraction unit, configured to extract local features from the medical text data;

[0039] Modeling unit, used to construct clinical medical knowledge graph and model coding relationships to obtain modeling results;

[0040] A fusion unit, configured to fuse the local features with the modeling results to obtain a feature fusion result;

[0041] An encoding unit, configured to input the feature fusion result into an encoding decision model for encoding to obtain an ICD encoding result;

[0042] An output unit is used to output the ICD coding result.

[0043] The present invention further provides a computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0044] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0045] The beneficial effects of the present invention compared with the existing technology are: the present invention obtains medical text data related to infectious diseases and extracts local features, constructs a clinical medical knowledge graph, models the coding relationship, and fuses the local features with the modeling results. The fusion results are input into the coding decision model, and the ICD coding results are generated and output, thereby improving the feature extraction and coding prediction process of infectious disease data to enhance the accuracy and efficiency of coding.

[0046] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A schematic diagram of an application scenario of the method for intelligent encoding of infectious disease data provided by an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of a flow chart of an intelligent encoding method for infectious disease data provided by an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of a sub-process of the method for intelligent encoding of infectious disease data provided by an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of a sub-process of the method for intelligent encoding of infectious disease data provided by an embodiment of the present invention;

[0052] Figure 5 A schematic diagram of a sub-process of the method for intelligent encoding of infectious disease data provided by an embodiment of the present invention;

[0053] Figure 6 A schematic diagram of a sub-process of the method for intelligent encoding of infectious disease data provided by an embodiment of the present invention;

[0054] Figure 7 A schematic block diagram of an intelligent encoding device for infectious disease data provided by an embodiment of the present invention;

[0055] Figure 8 A schematic block diagram of a local feature extraction unit of the infectious disease data intelligent encoding device provided by an embodiment of the present invention;

[0056] Figure 9 A schematic block diagram of a modeling unit of the infectious disease data intelligent encoding device provided by an embodiment of the present invention;

[0057] Figure 10 A schematic block diagram of a multi-layer graph convolution subunit of the infectious disease data intelligent encoding device provided by an embodiment of the present invention;

[0058] Figure 11 A schematic block diagram of a fusion unit of the infectious disease data intelligent encoding device provided by an embodiment of the present invention;

[0059] Figure 12 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0062] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0063] It should be further understood that the term "and / or" used in the present description and appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0064] See also Figure 1 and Figure 2 , Figure 1 Schematic diagram of an application scenario of the intelligent encoding method for infectious disease data provided by an embodiment of the present invention. Figure 2A schematic flow chart of the method for intelligently encoding infectious disease data provided by an embodiment of the present invention. The method for intelligently encoding infectious disease data is applied to a server. The server interacts with a terminal to obtain medical text data from an electronic medical record system; a deep convolutional neural network (CNN) is used to extract text features to capture important information in disease descriptions; a clinical medical knowledge graph is constructed, and a graph neural network (GNN) is applied to process the complex relationships between disease codes to achieve a deeper semantic understanding; an ensemble learning method is used to fuse the features extracted by CNN and GNN to improve the performance of the model; reinforcement learning and meta-learning are used to optimize model training to improve its adaptability and effectiveness; and a fuzzy logic system and a dynamic threshold adjustment strategy are used to make coding decisions to ensure that the final generated ICD code is accurate. This method aims to improve the accuracy of disease coding while being able to effectively process large amounts of complex medical data, thereby improving the efficiency of medical services.

[0065] Figure 2 FIG. 1 is a flow chart of the method for intelligently encoding infectious disease data provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S160.

[0066] S110. Obtain medical text data related to infectious diseases.

[0067] In this embodiment, the medical text data includes records related to infectious diseases, such as medical history descriptions, laboratory test results, and doctors' diagnoses.

[0068] Specifically, medical text data, primarily including medical record descriptions, laboratory test results, and physicians' diagnostic information, is received from the hospital's electronic medical record system. The system extracts the patient's symptoms, medical history, and physical signs from the medical record to identify information related to infectious diseases. These descriptions help understand the patient's specific condition. Laboratory test results (such as blood tests and urine analyses) are obtained, which help confirm the presence and severity of infectious diseases. Furthermore, the system extracts diagnostic conclusions about infectious diseases from physicians' diagnostic reports, including preliminary diagnoses, etiology analysis, and treatment recommendations.

[0069] Ensure that medical text data is stored in structured or unstructured formats and undergoes preprocessing, including noise removal, word segmentation, and entity recognition, to ensure data quality.

[0070] Data standardization: De-noise and format standardize the received data to unify the data format for subsequent processing.

[0071] Word segmentation and named entity recognition: Perform word segmentation on medical texts and use named entity recognition (NER) technology to extract entities related to infectious diseases, such as disease names, symptoms, and detection indicators.

[0072] Labeling and model training: Label the text data of each record with the corresponding ICD code to provide label data for subsequent model training and help improve the accuracy of disease coding.

[0073] Through these steps, key information can be extracted from medical texts more effectively to provide support for subsequent analysis and decision-making.

[0074] S120: Extract local features from the medical text data.

[0075] In this example, local features refer to small snippets or phrases within the text, such as specific symptoms, disease names, and experimental results. By sliding the convolution kernel over the text, the CNN can capture these local features and identify important keywords and patterns. For example, in the sentence "The patient developed a high fever and cough," local features might include "high fever" and "cough."

[0076] Contextual representation refers to understanding local features by taking into account their position and relationship within the overall text. Through multi-layer convolution and pooling operations, CNNs can integrate the connections between different local features, forming a more comprehensive understanding of the text. For example, the aforementioned "high fever" and "cough" combined with other patient descriptions (such as medical history and treatment response) can help the model better understand the entire case.

[0077] Combining these two, CNN can effectively extract key information from medical texts and provide rich representations to support subsequent analysis or decision-making.

[0078] In one embodiment, see Figure 3 , the above-mentioned step S120 may include steps S121~S124.

[0079] S121. Perform text decomposition and embedding on the medical text data to obtain an embedding matrix.

[0080] In this embodiment, the embedding matrix refers to a matrix obtained after text decomposition and embedding of medical text data.

[0081] Specifically, medical text is segmented into words or character sequences. This can be done using existing word segmentation tools (such as Jieba Word Segmenter or SpaCy) or custom word segmenters. For example, the sentence "patients need regular checkups" can be broken down into "patient," "need," "regularly," and "checkup."

[0082] Convert the segmented text into an embedding matrix. Use word embedding techniques such as Word2Vec, GloVe, or pre-trained models (such as BERT). The embedding matrix maps each word into a vector of fixed dimensions, typically tens to hundreds of dimensions. These vectors retain the semantic information of the word.

[0083] Word embedding technology can capture the semantic relationship between words. Words with similar meanings are closer in the vector space, making the model more accurate in understanding the context. It converts high-dimensional sparse representations into low-dimensional dense representations, reducing computational complexity and improving training efficiency. By using pre-trained models, existing knowledge can be directly utilized to accelerate the model's learning process and improve performance.

[0084] S122: Perform a convolution operation on the embedding matrix to obtain a convolution result.

[0085] In this embodiment, the convolution result refers to the result obtained after the convolution operation is performed on the embedded matrix.

[0086] Specifically, a convolution operation is performed by sliding a convolution kernel (filter) over the embedding matrix. Each convolution kernel is responsible for extracting a specific type of local feature; for example, a 3-gram convolution kernel can detect consecutive combinations of three words. A weighted sum of the local regions is calculated to obtain a feature map.

[0087] Perform a nonlinear transformation on the features after the convolution operation, usually using ReLU (Rectified Linear Unit) as the activation function. This operation can introduce nonlinear factors and increase the expressive power of the model.

[0088] Convolution operations can effectively extract important local features in text, such as keywords and phrases, helping to capture contextual information. Activation functions such as ReLU enable the model to learn more complex patterns, improve prediction accuracy, and avoid the limitations of linear models.

[0089] S123. Perform maximum pooling layer processing on the convolution result to obtain a pooling result.

[0090] In this embodiment, the pooling result refers to the result obtained by performing maximum pooling layer processing on the convolution result.

[0091] Specifically, after the convolution operation, a max pooling layer is used to select local maxima from the convolutional feature map. This process helps reduce the dimensionality of the feature map while retaining important information. For example, if the feature map is two-dimensional, it can be reduced by selecting the maximum value in each 2x2 region.

[0092] In addition to maximum pooling, average pooling can also be applied to calculate the average value from each local area to extract smoother features.

[0093] Max pooling can significantly reduce feature dimensions, lower computational complexity, help control model complexity, and reduce the risk of overfitting. The pooling operation enhances the model's robustness to small changes in input, making it more stable when facing text in different formats.

[0094] S124: performing feature map splicing and full connection processing on the pooling results to obtain local features.

[0095] In this embodiment, multiple convolution kernels are used to generate different feature maps, and these feature maps are spliced ​​to merge the features extracted by different convolution kernels. Through splicing, the features extracted by different convolution kernels are combined to form a richer feature representation.

[0096] The pooled feature map is flattened into a one-dimensional vector and then further processed through a fully connected layer to map the features into a high-dimensional space.

[0097] Through splicing, the model can comprehensively consider multiple local features and improve the overall representation ability; the fully connected layer can convert low-dimensional features into high-dimensional representations, enabling the model to learn more complex decision boundaries, thereby improving the accuracy of classification or regression tasks.

[0098] Through the above process, the resulting feature map is flattened into a one-dimensional vector. This vector serves as a high-dimensional feature representation of the medical text and can be used for subsequent classification or regression tasks.

[0099] The flattened feature vector provides a unified input format, making it easy to combine with other machine learning algorithms and support subsequent analysis and decision-making. The final feature vector effectively condenses the important information in the text, facilitates subsequent processing and in-depth analysis, and improves the efficiency and effectiveness of medical text processing.

[0100] Through these steps, deep convolutional neural networks can effectively extract rich features from medical text, improving the model's ability to understand and analyze text content. These features not only preserve semantic information but also enhance the model's expressive power through multi-level processing, providing strong support for medical analysis and decision-making.

[0101] S130. Construct a clinical medical knowledge graph and model the coding relationship to obtain a modeling result.

[0102] In this embodiment, the modeling results refer to the node representations and relationship information obtained by processing the clinical medical knowledge graph through a graph neural network (GNN). These results can be used to implement various applications, such as:

[0103] Node embedding representation: A vector representation of each node (such as disease, symptom, drug, etc.) that captures its characteristics and relationships.

[0104] Similarity calculation: The similarity between nodes is used to determine the correlation or similarity between different medical entities, which is very important for disease diagnosis and treatment plan recommendation.

[0105] Dependency: Revealing the dependencies between different medical concepts helps understand their roles in clinical decision making.

[0106] Knowledge reasoning: Utilizes the relationships between nodes for reasoning, assisting doctors in making more accurate diagnoses and treatment recommendations.

[0107] Decision support: Provide data support for clinical decision-making systems to improve the efficiency and accuracy of medical services.

[0108] These modeling results can help medical staff better understand the complex medical knowledge system and thus improve patient treatment outcomes.

[0109] Specifically, a clinical medical knowledge graph is constructed, and a graph neural network is used to model the encoding relationship to obtain the modeling results.

[0110] Graph neural networks (GNNs) are used to process graph-structured data to capture the complex relationships and dependencies between disease codes.

[0111] In one embodiment, see Figure 4 The above-mentioned step S130 may include steps S131~S135.

[0112] S131. Construct a clinical medicine knowledge graph.

[0113] In this embodiment, in the knowledge graph, the main nodes include:

[0114] Disease coding node: a code that uniquely identifies each disease;

[0115] Disease name node: describes the name of a specific disease;

[0116] Symptom nodes: various symptoms associated with the disease.

[0117] Each node is represented by a feature vector. For example, the embedding vector of a disease code can contain various attribute information of the disease (such as severity, transmission route, etc.).

[0118] Definition of relations and edges:

[0119] Nodes are connected by edges, which represent the relationship between nodes:

[0120] Parent-child relationship: Disease code nodes can be represented by a hierarchical relationship. For example, a major disease category can serve as the parent node of a subcategory disease.

[0121] Association strength: The edge weights between disease nodes and symptom nodes can be set based on statistical data or clinical research results to reflect the association strength between symptoms and diseases.

[0122] There are two ways to initialize the embedding representation of each node:

[0123] Random initialization: Generate random initial embedding vectors.

[0124] Based on pre-trained models: Use existing medical text data for pre-training to obtain more effective embedding representations and improve model performance.

[0125] S132. Perform multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation.

[0126] In one embodiment, see Figure 5 , the above-mentioned step S132 may include steps S1321~S1322.

[0127] S1321. Aggregate information of neighboring nodes in the clinical medical knowledge graph and calculate the weighted sum of features of each node;

[0128] S1322. Apply the weight matrix and bias term of the graph convolution layer and use the ReLU activation function for nonlinear transformation to obtain the node embedding representation.

[0129] In this embodiment, the embedding representation of each node is initialized , can be randomly initialized or embedded based on a pre-trained model. For each node Performing graph convolution operations includes the following steps:

[0130] By aggregating information from neighboring nodes, the weighted sum of the neighboring node features of each node is calculated using the following formula: ;in, is the normalization coefficient, is the weight matrix of the current layer, is the bias term.

[0131] Use the aggregation results to perform nonlinear transformation on the node representation: ;here, Usually it is the ReLU activation function to enhance the expressiveness of the model.

[0132] S133. Extract global features from the node embedding representation using a pooling operation.

[0133] In this embodiment, after completing multiple layers of graph convolution, a pooling operation is applied to extract global features.

[0134] Specifically, key features are extracted from the neighbors of each node to reduce dimensionality while retaining important information. This process helps control overfitting and improve the efficiency of the model.

[0135] S134. Calculate the similarity between nodes based on the global feature representation to determine the dependency relationship between the nodes.

[0136] In this embodiment, the final node embedding representation is obtained Finally, the important relationships are identified by calculating the similarity between nodes. The specific methods include:

[0137] Inner Product: Calculate the inner product between node embedding vectors to obtain a similarity score.

[0138] Distance metric: Use methods such as Euclidean distance or cosine similarity.

[0139] S135 , combining the node embedding representation and the dependencies between nodes to obtain a modeling result.

[0140] Specifically, the role of node embedded representation is,

[0141] Node embeddings are vector representations generated by computing the features and neighbor relationships of each node in a knowledge graph using a graph neural network (GNN). These embeddings contain the properties of the node itself and its relationships with other nodes. Their specific functions are as follows:

[0142] 1. Providing high-dimensional feature representations of nodes. In a knowledge graph, node embeddings convert the complex characteristics of each node into a high-dimensional vector. These vectors capture the semantic information of the node and include the relationships between them. The embedding representation of disease coding nodes is high-dimensional and can include multi-dimensional feature information, such as the common symptoms, course of disease, and treatment methods of a specific disease. This high-dimensional embedding enables the system to perform more accurate matching during the encoding process.

[0143] Second, it enriches contextual information and enhances predictive capabilities. This embedding can represent the interdependencies and influences between nodes, capturing complex semantic relationships. For example, a disease code node can capture the strength of its association with a symptom through its neighbor relationship with the symptom node. If multiple similar disease code nodes share similar symptom information, their embedding representations will become similar, helping the system identify the connection between these diseases during encoding.

[0144] Third, it is used for feature fusion, combining the outputs of other models. In the subsequent feature fusion process, the node embedding representation will be combined with the medical text features extracted by the CNN. This multimodal feature fusion can provide the system with more comprehensive disease coding decision support.

[0145] Combining medical text features: Features extracted through CNNs primarily focus on local patterns and linguistic information within medical text. Node embeddings, on the other hand, better reflect the context and structured relationships within the knowledge graph. Therefore, combining these two approaches can fully leverage the detailed information in the text and the global information in the knowledge graph, achieving a complementary effect.

[0146] The role of inter-node dependencies is as follows:

[0147] The dependencies between nodes are calculated in the knowledge graph using GNN. They are a pattern that reflects the mutual influence or association between nodes (such as disease codes, symptoms, etc.). The role of these dependencies in coding decisions is as follows:

[0148] Identify complex disease-symptom associations:

[0149] In clinical practice, certain diseases have strong dependencies on specific symptoms, and these dependencies are reflected through edge weights and feature propagation in GNNs. In the decision-making process of disease coding, the system can use this dependency to improve accuracy.

[0150] For example, if the relationship between a specific symptom node and multiple disease coding nodes is very strong (such as a certain symptom is a common manifestation of multiple diseases), the system can further confirm the rationality of the coding selection through these dependencies.

[0151] Furthermore, some diseases may have parent-child relationships. For example, the dependency between a sub-category of disease (e.g., an infection caused by a specific virus) and its superordinate disease (e.g., a generalized viral infection) can also be represented using this dependency model. These parent-child relationships help find the best match among similar codes, reducing errors.

[0152] Support similarity calculation in encoding decisions:

[0153] Based on the dependencies between nodes, the system can calculate node similarity to identify nodes with high correlation. This similarity calculation method helps capture the interactions and similarities between different nodes, and is particularly effective when screening between similar disease codes.

[0154] Inner product or distance metric: Using inner product or distance metric, the system can calculate the similarity between two nodes (such as disease code nodes or symptom nodes). If the similarity between two code nodes is high, the system can determine that they may have similar medical semantics.

[0155] Decision support: Nodes with high similarity may share the same symptoms or clinical manifestations. The system can leverage this dependency to further differentiate or cluster similar disease codes. This information can assist the system in making more refined disease code selections when making coding decisions.

[0156] Optimize the hierarchical structure processing of disease coding:

[0157] Disease codes often have a hierarchical structure, such as parent-child relationships or different levels of classification. Dependencies between nodes can help the system better handle this hierarchical structure during coding, ensuring that the selected code conforms to a standardized classification system.

[0158] Parent-child code selection: In some complex cases, diseases may fall into multiple categories. The system can use the dependencies between disease code nodes to make appropriate choices between parent-child codes. For example, a specific infectious disease may be a subcategory of a higher-level category (such as "infectious diseases"). This parent-child relationship allows the system to more accurately select the appropriate code.

[0159] Specific application of node embedding representation and dependency in feature fusion:

[0160] The specific application of feature fusion: The system combines the node embedding representation with the text features extracted by CNN to form a multi-dimensional feature vector. The key to this step is how to effectively integrate features from different sources.

[0161] Ensemble learning: The system uses ensemble learning methods (such as random forests) to improve model robustness by fusing node embeddings and text features. Ensemble learning methods combine different features and integrate multiple models to reduce the bias of a single model and enhance prediction stability.

[0162] Application in coding decisions:

[0163] In the final encoding decision process, the system uses node similarity calculation and dependency relationships to optimize the encoding selection. By handling uncertainty through fuzzy logic, the system can dynamically adjust the threshold, making the encoding selection more flexible and accurate.

[0164] Uncertainty handling: Based on node embeddings and similarity, the system can determine the prediction probabilities of different encodings. If certain encodings have high similarity, the system will prefer them and dynamically adjust based on the corresponding threshold.

[0165] Dependency-assisted decision-making: Dependencies between nodes also provide the system with rich auxiliary information. For example, between symptom nodes and disease code nodes, the system can use these dependencies to make more accurate coding decisions, ensuring that the selected code closely matches the patient's symptoms.

[0166] In this example, the final output is an embedded representation of each node, which is used for subsequent feature fusion and coding decisions. This embedded information is rich in relational features, significantly enhancing the understanding and accuracy of intelligent coding of infectious diseases.

[0167] Through graph structures, different types of medical information can be effectively integrated to capture complex relationships and dependencies; graph convolutional networks can learn rich node representations through multi-level aggregation and nonlinear transformations, thereby improving model performance; improved node similarity calculations provide strong support for disease diagnosis and treatment plans, helping clinicians make more accurate decisions; the combination of knowledge graphs and graph convolutional networks has good scalability and can adapt to the increasing amount of new data and new knowledge.

[0168] S140: Fusing the local features with the modeling results to obtain a feature fusion result.

[0169] In this embodiment, the feature fusion result is a comprehensive feature expression obtained by combining local features with modeling results. The goal of this process is to combine local features and global relationship information to improve the accuracy of subsequent encoding decisions.

[0170] In one embodiment, see Figure 6 , the above-mentioned step S140 may include steps S141~S143.

[0171] S141. Concatenate the local features with the modeling results to obtain a high-dimensional feature vector.

[0172] In this embodiment, the high-dimensional feature vector refers to a vector obtained by concatenating the local feature and the modeling result.

[0173] The feature vectors from the CNN and GNN are concatenated into a higher-dimensional feature vector. This concatenation effectively fuses the diverse information from the image and graph data. This integration ensures that the model fully utilizes rich contextual information, resulting in a more comprehensive feature representation.

[0174] S142: Perform dimensionality reduction processing on the high-dimensional feature vector to obtain a dimensionality reduction result.

[0175] In this embodiment, the dimensionality reduction result refers to the result obtained after the dimensionality reduction of the high-dimensional feature vector.

[0176] To reduce computational complexity and prevent overfitting, the concatenated high-dimensional feature vectors are then subjected to dimensionality reduction. Common dimensionality reduction techniques include principal component analysis (PCA) and linear discriminant analysis (LDA). These methods can reduce feature dimensionality while retaining important feature information. This step not only improves computational efficiency but also helps the model focus on the most discriminative features, avoiding interference from redundant information.

[0177] S143: Input the dimension reduction result into an integrated learning model for feature fusion to obtain a feature fusion result.

[0178] In this embodiment, the fused feature vector is then fed into an ensemble learning model, such as a random forest or gradient boosted tree. These ensemble learning methods significantly improve overall accuracy and stability by combining the predictions of multiple base learners. During this process, each base learner is trained using the same feature set, and the combined predictions are ultimately generated through weighted voting or averaging.

[0179] The resulting fusion features serve as input to the coding decision model. This feature set provides comprehensive information support, significantly enhancing the accuracy and performance of intelligent coding for infectious diseases.

[0180] Combining the features of CNN and GNN enables the model to obtain information from multiple data types and improve the diversity of information; through dimensionality reduction, redundant features are removed, the complexity of the model is reduced, and the model's generalization ability on new data is stronger; taking advantage of the advantages of ensemble learning models, the accuracy and reliability of predictions are improved through the combination of multiple base learners; this feature fusion method can be used for a variety of intelligent decision-making systems, especially in the fields of medicine, finance, etc., and has important application value.

[0181] By combining the above steps and benefits, this feature fusion solution can effectively improve the performance of the model and meet the needs of actual applications.

[0182] Ensemble learning methods (such as random forests) improve the accuracy and stability of predictions by fusing multiple models rather than a single multi-layer perceptron.

[0183] S150: Input the feature fusion result into a coding decision model for coding to obtain an ICD coding result.

[0184] In this example, ICD codes refer to the International Classification of Diseases (ICD) system used to identify diseases and health conditions. ICD codes are used to standardize medical records, statistics, and research, helping doctors, researchers, and public health agencies accurately record and analyze disease data.

[0185] In practice, ICD coding results are typically generated by analyzing a patient's clinical data, symptoms, diagnosis, and other information and mapping them to corresponding codes in the ICD system. This process can leverage machine learning models to improve coding accuracy and efficiency.

[0186] Specifically, the feature fusion result is input into the coding decision model for coding, and fuzzy logic is used to determine the ICD coding result;

[0187] The coding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

[0188] A reinforcement learning framework is applied to the coding decision model to optimize the coding decision process. A reward mechanism dynamically adjusts model parameters to improve adaptability to different cases. Furthermore, a meta-learning strategy is introduced to enable the model to quickly adapt to new disease coding tasks.

[0189] The fused features are input into the coding decision model. A fuzzy logic system is used to address the uncertainty of the predicted results, and a dynamic threshold adjustment strategy is used to optimize the final code selection. The predicted probability of each ICD code is compared with the adjusted threshold, and the ICD code that meets the requirements is selected as the final disease code output, completing the code assignment.

[0190] Specifically, models for encoding decisions typically use a multi-layered neural network structure, which can effectively process complex input data and enable the model to capture and express various feature relationships in the data.

[0191] Input: The input of the model is the fused feature vector, which integrates various disease-related information, such as the patient's medical history, examination results, etc.

[0192] Output: The output of the model is the corresponding disease code. The predicted probability of each code reflects the degree of match between the code and the current input features.

[0193] By introducing reinforcement learning algorithms (such as Q-learning or deep Q networks), the model can select the optimal action (a) based on the current state (s) in each prediction and update the Q value based on the reward (r). The specific update formula is: ; where α is the learning rate, γ is the discount factor, For the next state.

[0194] This dynamic updating mechanism enables the model to gradually improve its decision-making strategy, thereby improving the accuracy of encoding.

[0195] Furthermore, the introduction of a meta-learning strategy enables the model to quickly adapt to new disease coding tasks. By training on multiple related tasks, the model can learn shared knowledge, thereby speeding up its response to new situations.

[0196] When encountering new tasks, the model can quickly adjust by leveraging previously learned knowledge without having to start training from scratch. Even when data is scarce, the model can effectively learn and infer through existing experience.

[0197] During model training, it is crucial to regularly evaluate performance. Techniques such as cross-validation and holdout can be used to ensure the generalization ability of the model and avoid overfitting.

[0198] Cross-validation: Divide the dataset into multiple subsets and evaluate the performance of the model by alternating between training and testing on different subsets.

[0199] Hyperparameter adjustment: Based on the evaluation results, timely adjust the model's hyperparameters (such as learning rate, number of hidden layers, etc.) to optimize the overall performance of the model.

[0200] Through the above methods, the application of reinforcement learning framework in disease coding brings multiple benefits:

[0201] Improve coding accuracy: Dynamically optimized decision-making processes can effectively reduce coding errors and improve the quality and reliability of medical documents.

[0202] Enhanced model adaptability: The meta-learning strategy enables the model to quickly adapt to new encoding tasks, reducing the dependence on large amounts of labeled data.

[0203] Stronger generalization: Regular performance evaluation and hyperparameter tuning ensure that the model maintains good performance in different situations, facilitating implementation in real-world clinical settings.

[0204] Support clinical decision-making: Accurate disease coding not only helps with patient management, but also provides reliable data support for clinical decision-making and improves medical efficiency.

[0205] In summary, combining the reinforcement learning framework with the meta-learning strategy can significantly optimize the disease coding process and bring more efficient and accurate solutions to the medical industry.

[0206] In this example, during the disease coding process, the model outputs a series of predicted probability values ​​that indicate how well each possible International Classification of Diseases (ICD) code matches the input features. This allows the system to provide a clear coding choice for each input, enhancing its interpretability and accuracy.

[0207] The model first generates unnormalized scores based on the input features and then normalizes them using the softmax function. The softmax function converts these scores into probability values, as shown in the following formula: ,in, is the unnormalized score for the corresponding code, and N is the number of possible codes. In this way, the output probability values ​​can be easily understood and interpreted, allowing medical personnel to more intuitively assess the likelihood of each code.

[0208] To address the uncertainty in the prediction results, the system uses a fuzzy logic system. Based on dynamic threshold setting, the system compares the predicted probability value with a pre-set threshold to determine the final coding selection.

[0209] Dynamic threshold setting: This threshold is not fixed but continuously adjusted based on historical data and real-time feedback. This allows the system to better adapt to the characteristics and needs of different cases, ensuring appropriate decisions even in the face of high uncertainty.

[0210] The specific optimization process includes the following steps:

[0211] By analyzing historical case data, we can identify which combinations of features are associated with successful coding outcomes. This process helps us understand which factors have the greatest impact on coding outcomes.

[0212] Feedback on encoding results is collected in real time and used to continuously update and adjust the dynamic thresholds. This feedback mechanism ensures that the system can learn and improve its decision-making process.

[0213] Based on the analysis results and feedback information, the threshold used to judge the coding selection is dynamically adjusted to improve the accuracy of the decision.

[0214] Based on the adjusted dynamic threshold, the system selects the ICD code that meets the requirements from the predicted probability value as the final disease code output.

[0215] This approach brings multiple benefits to the disease coding process:

[0216] Improved accuracy: By normalizing probability values ​​and introducing fuzzy logic, the system can make more accurate coding choices when faced with complex and uncertain situations.

[0217] Enhanced explainability: Probability-based outputs enable medical professionals to understand the rationale behind each coding choice, thereby increasing trust.

[0218] Strong adaptability: The setting of dynamic thresholds allows the system to be flexibly adjusted according to the characteristics of different cases, enhancing the adaptability of the system.

[0219] Continuous optimization: Through feedback mechanisms and historical data analysis, the system is able to continuously optimize its decision-making process and improve overall performance.

[0220] Support clinical decision-making: The resulting disease codes will be used for subsequent clinical decision support and data analysis to ensure that the patient's condition is accurately reflected and provide a good data foundation for medical treatment.

[0221] In short, by combining probability output, fuzzy logic system and dynamic threshold adjustment, the disease coding process has become more intelligent and efficient, effectively improving the quality and reliability of coding.

[0222] S160: Output the ICD coding result.

[0223] In this embodiment, the final ICD coding results are managed and stored, including data storage structure and clinical support functions.

[0224] The above-mentioned intelligent coding method for infectious disease data obtains medical text data related to infectious diseases and extracts local features, constructs a clinical medical knowledge graph, models the coding relationship, and fuses the local features with the modeling results. The fusion results are input into the coding decision model to generate and output the ICD coding results, thereby improving the feature extraction and coding prediction process of infectious disease data to enhance the accuracy and efficiency of coding.

[0225] Figure 7 FIG. 3 is a schematic block diagram of an intelligent encoding device 300 for infectious disease data provided by an embodiment of the present invention. Figure 7 As shown, corresponding to the above infectious disease data intelligent encoding method, the present invention also provides an infectious disease data intelligent encoding device 300. The infectious disease data intelligent encoding device 300 includes a unit for executing the above infectious disease data intelligent encoding method, and the device can be configured in a server. Figure 7 The infectious disease data intelligent encoding device 300 includes a data acquisition unit 301, a local feature extraction unit 302, a modeling unit 303, a fusion unit 304, an encoding unit 305 and an output unit 306.

[0226] The data acquisition unit 301 is used to acquire medical text data related to infectious diseases; the local feature extraction unit 302 is used to extract local features from the medical text data; the modeling unit 303 is used to construct a clinical medical knowledge graph and model the coding relationship to obtain a modeling result; the fusion unit 304 is used to fuse the local features with the modeling result to obtain a feature fusion result; the encoding unit 305 is used to input the feature fusion result into the coding decision model for encoding to obtain an ICD coding result; the output unit 306 is used to output the ICD coding result.

[0227] In one embodiment, if Figure 8 As shown, the local feature extraction unit 302 includes an embedding subunit 3021, a convolution subunit 3022, a maximum pooling subunit 3023 and a splicing fully connected subunit 3024.

[0228] The embedding subunit 3021 is used to perform text decomposition and embedding on the medical text data to obtain an embedding matrix; the convolution subunit 3022 is used to perform a convolution operation on the embedding matrix to obtain a convolution result; the maximum pooling subunit 3023 is used to perform a maximum pooling layer processing on the convolution result to obtain a pooling result; the splicing fully connected subunit 3024 is used to perform feature map splicing and full connection processing on the pooling result to obtain local features.

[0229] In one embodiment, the modeling unit 303 is used to construct a clinical medical knowledge graph and use a graph neural network to model the coding relationship to obtain a modeling result.

[0230] In one embodiment, if Figure 9 As shown, the modeling unit 303 includes a graph construction subunit 3031, a multi-layer graph convolution subunit 3032, a pooling operation subunit 3033, a similarity calculation subunit 3034 and a combination subunit 3035.

[0231] The graph construction subunit 3031 is used to construct a clinical medical knowledge graph; the multi-layer graph convolution subunit 3032 is used to perform multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation; the pooling operation subunit 3033 is used to extract global features from the node embedding representation using a pooling operation; the similarity calculation subunit 3034 is used to calculate the similarity between nodes based on the global feature representation to determine the dependency relationship between nodes; the combination subunit 3035 is used to combine the node embedding representation and the dependency relationship between nodes to obtain a modeling result.

[0232] In one embodiment, if Figure 10As shown, the multi-layer graph convolution subunit 3032 includes a weighted sum module 30321 and a transformation module 30322.

[0233] The weighted sum module 30321 is used to aggregate the information of neighbor nodes in the clinical medical knowledge graph and calculate the weighted sum of the features of each node; the transformation module 30322 is used to apply the weight matrix and bias term of the graph convolution layer and use the ReLU activation function for nonlinear transformation to obtain the node embedding representation.

[0234] In one embodiment, if Figure 11 As shown, the fusion unit 304 includes a splicing subunit 3041 , a dimensionality reduction subunit 3042 and a feature fusion subunit 3043 .

[0235] The splicing subunit 3041 is used to splice the local features with the modeling results to obtain a high-dimensional feature vector; the dimensionality reduction subunit 3042 is used to perform dimensionality reduction processing on the high-dimensional feature vector to obtain a dimensionality reduction result; the feature fusion subunit 3043 is used to input the dimensionality reduction result into the integrated learning model for feature fusion to obtain a feature fusion result.

[0236] In one embodiment, the encoding unit 305 is used to input the feature fusion result into the encoding decision model for encoding, and use fuzzy logic to determine the ICD encoding result; wherein, the encoding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

[0237] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned infectious disease data intelligent encoding device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and brevity of the description, it will not be repeated here.

[0238] The above-mentioned infectious disease data intelligent encoding device 300 can be implemented in the form of a computer program. The computer program can be used in Figure 12 Runs on the computer equipment shown.

[0239] See also Figure 12 , Figure 12 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0240] See Figure 12 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0241] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, can enable the processor 502 to execute a method for intelligently encoding infectious disease data.

[0242] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0243] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an intelligent encoding method for infectious disease data.

[0244] The network interface 505 is used to communicate with other devices through the network. Figure 12 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0245] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0246] Obtain medical text data related to infectious diseases; extract local features from the medical text data; construct a clinical medical knowledge graph and model the coding relationship to obtain a modeling result; fuse the local features with the modeling result to obtain a feature fusion result; input the feature fusion result into a coding decision model for coding to obtain an ICD coding result; and output the ICD coding result.

[0247] In one embodiment, when the processor 502 implements the step of extracting local features from the medical text data, it specifically implements the following steps:

[0248] The medical text data is subjected to text decomposition and embedding to obtain an embedding matrix; a convolution operation is performed on the embedding matrix to obtain a convolution result; a maximum pooling layer is performed on the convolution result to obtain a pooling result; and feature map splicing and full connection processing are performed on the pooling result to obtain local features.

[0249] In one embodiment, when the processor 502 implements the step of constructing a clinical medical knowledge graph and modeling the coding relationship to obtain a modeling result, the processor 502 specifically implements the following steps:

[0250] Construct a clinical medical knowledge graph and use graph neural network to model the encoding relationship to obtain the modeling results.

[0251] In one embodiment, when implementing the step of constructing a clinical medical knowledge graph and using a graph neural network to model the coding relationship to obtain a modeling result, the processor 502 specifically implements the following steps:

[0252] Construct a clinical medical knowledge graph; perform multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation; use a pooling operation to extract global features from the node embedding representation; calculate the similarity between nodes based on the global feature representation to determine the dependency relationship between nodes; combine the node embedding representation and the dependency relationship between nodes to obtain a modeling result.

[0253] In one embodiment, when the processor 502 performs multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation, the processor 502 specifically implements the following steps:

[0254] The information of neighbor nodes in the clinical medical knowledge graph is aggregated, and the weighted sum of the features of each node is calculated; the weight matrix and bias term of the graph convolution layer are applied, and a nonlinear transformation is performed using the ReLU activation function to obtain a node embedding representation.

[0255] In one embodiment, when the processor 502 implements the step of fusing the local features with the modeling results to obtain a feature fusion result, the processor 502 specifically implements the following steps:

[0256] The local features are spliced ​​with the modeling results to obtain a high-dimensional feature vector; the high-dimensional feature vector is subjected to dimensionality reduction processing to obtain a dimensionality reduction result; the dimensionality reduction result is input into an integrated learning model for feature fusion to obtain a feature fusion result.

[0257] In one embodiment, when the processor 502 implements the step of inputting the feature fusion result into the coding decision model for coding to obtain the ICD coding result, the processor 502 specifically implements the following steps:

[0258] Inputting the feature fusion result into the coding decision model for coding, and using fuzzy logic to determine the ICD coding result;

[0259] The coding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

[0260] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0261] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0262] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0263] Obtain medical text data related to infectious diseases; extract local features from the medical text data; construct a clinical medical knowledge graph and model the coding relationship to obtain a modeling result; fuse the local features with the modeling result to obtain a feature fusion result; input the feature fusion result into a coding decision model for coding to obtain an ICD coding result; and output the ICD coding result.

[0264] In one embodiment, when the processor executes the computer program to implement the step of extracting local features from the medical text data, the processor specifically implements the following steps:

[0265] The medical text data is subjected to text decomposition and embedding to obtain an embedding matrix; a convolution operation is performed on the embedding matrix to obtain a convolution result; a maximum pooling layer is performed on the convolution result to obtain a pooling result; and feature map splicing and full connection processing are performed on the pooling result to obtain local features.

[0266] In one embodiment, when the processor executes the computer program to implement the step of constructing a clinical medical knowledge graph and modeling the coding relationship to obtain a modeling result, the processor specifically implements the following steps:

[0267] Construct a clinical medical knowledge graph and use graph neural network to model the encoding relationship to obtain the modeling results.

[0268] In one embodiment, when the processor executes the computer program to implement the step of constructing a clinical medical knowledge graph and using a graph neural network to model the coding relationship to obtain a modeling result, the processor specifically implements the following steps:

[0269] Construct a clinical medical knowledge graph; perform multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation; use a pooling operation to extract global features from the node embedding representation; calculate the similarity between nodes based on the global feature representation to determine the dependency relationship between nodes; combine the node embedding representation and the dependency relationship between nodes to obtain a modeling result.

[0270] In one embodiment, when the processor executes the computer program to implement the step of performing multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation, the processor specifically implements the following steps:

[0271] The information of neighbor nodes in the clinical medical knowledge graph is aggregated, and the weighted sum of the features of each node is calculated; the weight matrix and bias term of the graph convolution layer are applied, and a nonlinear transformation is performed using the ReLU activation function to obtain a node embedding representation.

[0272] In one embodiment, when the processor executes the computer program to implement the step of fusing the local features with the modeling results to obtain a feature fusion result, the processor specifically implements the following steps:

[0273] The local features are spliced ​​with the modeling results to obtain a high-dimensional feature vector; the high-dimensional feature vector is subjected to dimensionality reduction processing to obtain a dimensionality reduction result; the dimensionality reduction result is input into an integrated learning model for feature fusion to obtain a feature fusion result.

[0274] In one embodiment, when the processor executes the computer program to implement the step of inputting the feature fusion result into the coding decision model for coding to obtain the ICD coding result, the processor specifically implements the following steps:

[0275] Inputting the feature fusion result into the coding decision model for coding, and using fuzzy logic to determine the ICD coding result;

[0276] The coding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

[0277] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0278] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0279] In the several embodiments provided herein, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementation may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0280] The steps in the methods of the embodiments of the present invention may be adjusted in order, combined, or deleted as needed. The units in the devices of the embodiments of the present invention may be combined, divided, or deleted as needed. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0281] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, terminal, or network device) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0282] The foregoing description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art will readily conceive of various equivalent modifications or substitutions within the technical scope disclosed herein, and such modifications or substitutions are intended to be encompassed within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An intelligent coding method for infectious disease data, characterized in that: include: Obtain medical text data related to infectious diseases; Extracting local features from the medical text data; Construct a clinical medical knowledge graph and model the coding relationship to obtain modeling results; Fusing the local features with the modeling results to obtain a feature fusion result; The feature fusion result is input into the coding decision model for coding to obtain the ICD coding result; wherein, the coding decision model is constructed based on the reinforcement learning framework, and can select the optimal action according to the current state in each prediction, and dynamically adjust the model parameters according to the rewards obtained, thereby improving the adaptability to different cases; at the same time, a meta-learning strategy is introduced to enable the model to quickly adapt to the task of new disease coding and reduce the dependence on large amounts of labeled data; a fuzzy logic system is used to handle the uncertainty of the predicted results, and a dynamic threshold adjustment strategy is used to optimize the final coding selection; based on the comparison of the predicted probability value of each ICD code with the adjusted threshold, the ICD code that meets the requirements is selected as the final disease coding output to complete the coding allocation; Outputting the ICD coding result; The construction of the clinical medical knowledge graph and modeling of the coding relationship to obtain the modeling results include: Construct a clinical medical knowledge graph and use graph neural networks to model the coding relationship to obtain modeling results; The construction of the clinical medical knowledge graph uses a graph neural network to model the coding relationship to obtain a modeling result, including: Construct a clinical medical knowledge graph; Performing multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation; Extracting global features using a pooling operation on the node embedding representation; Calculating similarities between nodes based on the global feature representation to determine dependencies between the nodes; Combining the node embedding representation and the dependencies between nodes to obtain a modeling result; The performing multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation includes: Aggregating information of neighboring nodes in the clinical medical knowledge graph and calculating a weighted sum of features of each node; Apply the weight matrix and bias term of the graph convolution layer and use the ReLU activation function for nonlinear transformation to obtain the node embedding representation; The extracting of local features from the medical text data includes: Performing text decomposition and embedding on the medical text data to obtain an embedding matrix; Performing a convolution operation on the embedding matrix to obtain a convolution result; Performing a maximum pooling layer processing on the convolution result to obtain a pooling result; Performing feature map splicing and full connection processing on the pooling results to obtain local features; The fusing of the local features with the modeling results to obtain a feature fusion result includes: Concatenating the local features with the modeling results to obtain a high-dimensional feature vector; Performing dimensionality reduction processing on the high-dimensional feature vector to obtain a dimensionality reduction result; The dimensionality reduction result is input into the ensemble learning model for feature fusion to obtain a feature fusion result.

2. The method for intelligently encoding infectious disease data according to claim 1, characterized in that: Inputting the feature fusion result into the coding decision model for coding to obtain an ICD coding result includes: Inputting the feature fusion result into the coding decision model for coding, and using fuzzy logic to determine the ICD coding result; The coding decision model is a multi-level neural network model constructed using a reinforcement learning framework.

3. Intelligent encoding device for infectious disease data, characterized in that: include: A data acquisition unit, used for acquiring medical text data related to infectious diseases; A local feature extraction unit, configured to extract local features from the medical text data; Modeling unit, used to construct clinical medical knowledge graph and model coding relationships to obtain modeling results; A fusion unit, configured to fuse the local features with the modeling results to obtain a feature fusion result; An encoding unit is configured to input the feature fusion result into a coding decision model for encoding to obtain an ICD coding result; wherein the coding decision model is constructed based on a reinforcement learning framework, and is capable of selecting the optimal action based on the current state in each prediction, and dynamically adjusting model parameters based on the rewards obtained, thereby improving adaptability to different cases; at the same time, a meta-learning strategy is introduced to enable the model to quickly adapt to new disease coding tasks and reduce dependence on large amounts of labeled data; a fuzzy logic system is used to handle the uncertainty of the predicted results, and a dynamic threshold adjustment strategy is used to optimize the final coding selection; based on the comparison of the predicted probability value of each ICD code with the adjusted threshold, the ICD code that meets the requirements is selected as the final disease coding output to complete the coding allocation; An output unit, configured to output the ICD coding result; The modeling unit is used to construct a clinical medical knowledge graph and use a graph neural network to model the coding relationship to obtain a modeling result; The modeling unit includes a graph construction subunit, a multi-layer graph convolution subunit, a pooling operation subunit, a similarity calculation subunit and a combination subunit; A graph construction subunit is used to construct a clinical medical knowledge graph; a multi-layer graph convolution subunit is used to perform multi-layer graph convolution processing on the clinical medical knowledge graph to obtain a node embedding representation; a pooling operation subunit is used to extract global features from the node embedding representation using a pooling operation; a similarity calculation subunit is used to calculate the similarity between nodes based on the global feature representation to determine the dependency relationship between nodes; a combination subunit is used to combine the node embedding representation and the dependency relationship between nodes to obtain a modeling result; The multi-layer graph convolution subunit includes a weighted sum module and a transformation module; A weighted sum module is used to aggregate information of neighboring nodes in the clinical medical knowledge graph and calculate the weighted sum of the features of each node; a transformation module is used to apply the weight matrix and bias term of the graph convolution layer and perform nonlinear transformation using the ReLU activation function to obtain a node embedding representation; The multi-layer graph convolution subunit includes a weighted sum module and a transformation module; the weighted sum module is used to aggregate the information of neighbor nodes in the clinical medical knowledge graph and calculate the weighted sum of the features of each node; the transformation module is used to apply the weight matrix and bias term of the graph convolution layer and use the ReLU activation function to perform nonlinear transformation to obtain a node embedding representation; the fusion unit includes a splicing subunit, a dimensionality reduction subunit and a feature fusion subunit; the splicing subunit is used to splice the local features with the modeling results to obtain a high-dimensional feature vector; the dimensionality reduction subunit is used to perform dimensionality reduction processing on the high-dimensional feature vector to obtain a dimensionality reduction result; the feature fusion subunit is used to input the dimensionality reduction result into the integrated learning model for feature fusion to obtain a feature fusion result.

4. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 2 when executing the computer program.

5. A storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

Images