An automatic ICD coding method and system based on fusion of multiple label information

Abstract

The present invention belongs to the field of medical data processing technology, and discloses an automatic ICD coding method and system based on the fusion of multiple label information, including feature extraction of clinical text and ICD label description to obtain clinical text representation and label representation; interactive fusion of the two to obtain label-related context representation, extracting co-occurrence representation of label representation, and fusion of it with ICD label description and label-related context representation to obtain enhanced label representation; interactive fusion of enhanced label representation and clinical text representation to obtain document representation corresponding to each label, and predicting ICD coding based on document representation; defining loss function and training the model. The present invention integrates information from three aspects: context, original semantics of labels and label correlation, provides richer input features for automatic ICD coding, improves the comprehensiveness and accuracy of coding, and provides important support for clinical decision-making and medical resource allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data processing, and in particular to an automatic ICD encoding method and system based on the fusion of multiple tag information. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the modern medical system, the International Classification of Diseases (ICD) is an important tool for recording and counting diseases, injuries and the causes of death caused by them. With the increasing complexity of global medical services, the demand for ICD coding is growing. It is not only the basis of medical statistics, but also an important basis for medical insurance reimbursement, health management and resource allocation. With the popularization of electronic health record (EHR) systems, how to quickly and accurately perform ICD coding has become a major challenge facing the medical industry. Traditional ICD coding methods mainly rely on manual coding, which is often time-consuming and error-prone. Manual coders need to assign appropriate ICD codes based on the patient's clinical text according to strict standards and specifications. Due to the complexity of medical texts, coders may miss key information when making judgments, resulting in inaccurate coding. In addition, due to the different experience and knowledge background of each coder, inconsistent coding may occur, which directly affects the accuracy and comparability of medical statistics.

[0004] With the rapid development of computer technology, automatic ICD coding methods have gradually attracted attention. Current automatic ICD coding techniques mainly regard this task as a multi-label text classification problem, aiming to automatically identify and assign multiple ICD codes from patients' electronic health records. Although there have been some preliminary studies and applications, existing methods still have many shortcomings. For example, many existing coding systems only focus on brief descriptions in the text, while ignoring the richer information contained in the medical text. This lack of information may lead to incomplete or inaccurate coding and fail to truly reflect the patient's health status. In addition, medical texts often use complex terminology and professional expressions, which makes natural language processing technology face challenges in identifying and parsing this information. Although existing methods based on label attention mechanisms have improved the accuracy of coding to a certain extent, due to their reliance on information extraction, they may still be insufficient to cope with the complexity of medical texts. Therefore, improving the automation level of ICD coding and ensuring the accuracy of coding have become a hot topic in current research. In order to solve these problems, many researchers have begun to explore the application of deep learning technology to ICD coding, especially by building more complex models to capture the relationship between text and labels. However, these methods still need to be improved in the depth and breadth of information fusion in order to more comprehensively understand the information in medical texts.

[0005] In summary, although automatic ICD coding technology has developed rapidly in recent years, its accuracy and application scope still need to be further improved to meet the growing demand for medical services and industry standards. Therefore, a new method is urgently needed to provide a more comprehensive and accurate automatic ICD coding solution. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes an automatic ICD encoding method and system based on the fusion of multiple label information, designs a multiple label information fusion module to enhance label representation learning, and integrates information from three aspects: context, label original semantics and label correlation. It provides richer input features for automatic ICD encoding, improves the comprehensiveness and accuracy of encoding, and provides important support for clinical decision-making and medical resource allocation.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides an automatic ICD encoding method based on fusion of multiple tag information, comprising the following steps:

[0009] Feature extraction is performed on the acquired clinical text and ICD label descriptions to obtain clinical text representation and label representation;

[0010] Interactively fuse the label representation with the clinical text representation to obtain the label-related context representation, extract the co-occurrence representation of the label representation, and fuse it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation;

[0011] Interactively fuse the enhanced label representation with the clinical text representation to obtain the document representation corresponding to each label, and predict the ICD code based on the document representation corresponding to each label;

[0012] Define the loss function, optimize the model parameters, and train the automatic ICD coding model.

[0013] As an optional implementation, the label representation and the clinical text representation are interactively integrated, specifically:

[0014] The label representation is used as the query sequence, and the clinical text representation is used as the key and value sequence. The Transformer decoder including the self-attention module, the cross-attention module and the feedforward neural network is used to obtain the label-related context representation.

[0015] As an optional implementation, the co-occurrence representation of the tag representation is extracted, specifically:

[0016] A multi-layer graph convolutional neural network is used to model the co-occurrence relationship between tags into a graph. The tag features are used as nodes in the graph structure, and the adjacency matrix is ​​used as edges in the graph structure. The tag features are enhanced by propagating messages between nodes. The deep connection between each tag is learned from the statistical relationship of tag co-occurrence, and the co-occurrence representation of the tag representation is output.

[0017] As an optional implementation, the self-attention module and the cross-attention module both use a multi-head scaled dot product attention mechanism. The multi-head attention can capture multiple perspective features and help the model learn richer features.

[0018] As an optional implementation, the clinical text is extracted from electronic health records and undergoes preliminary data cleaning and preprocessing to remove irrelevant information and correct format errors to ensure the quality and consistency of the input data.

[0019] As an optional implementation, the loss function of the automatic ICD coding model is defined as:

[0020]

[0021] in For the tags, For the The true value of the label, For the The document features corresponding to the labels are linearly projected to obtain the prediction results. A collection of tags.

[0022] In a second aspect, the present invention provides an automatic ICD coding system based on the fusion of multiple tag information, comprising:

[0023] The feature extraction module is configured to: extract features from the acquired clinical text and ICD label descriptions respectively to obtain clinical text representation and label representation;

[0024] The information enhancement module is configured to: interactively fuse the label representation with the clinical text representation to obtain the label-related context representation, extract the co-occurrence representation of the label representation, and fuse it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation;

[0025] The ICD coding module is configured to: interactively fuse the enhanced label representation with the clinical text representation, obtain the document representation corresponding to each label, and predict the ICD code based on the document representation corresponding to each label;

[0026] The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic ICD coding model.

[0027] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0028] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0029] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) This paper proposes an automatic ICD encoding method and system based on the fusion of multiple label information, and develops a Transformer-based architecture for automatic ICD encoding. As an effective means of fusing text and labels and a way to query whether multiple labels exist, it improves the limitations of traditional encoding methods, making the model perform well in automatic encoding tasks and adapt to the needs of multi-label classification.

[0032] (2) The present invention proposes an automatic ICD coding method and system based on the fusion of multiple label information, and proposes a multiple label information fusion module that integrates contextualized label features, label original semantics, and label correlation to enhance the representation learning of ICD coding. This method effectively improves the comprehensive understanding of the patient's health status and provides richer input features for automatic ICD coding.

[0033] (3) The present invention proposes an automatic ICD coding method and system based on the fusion of multiple label information. The proposed multi-label information fusion method not only reflects the patient's main disease, but also records and takes into account complications and other related health problems at the same time, greatly improving the comprehensiveness and accuracy of the coding, and providing important support for clinical decision-making and medical resource allocation.

[0034] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0036] Figure 1 A flowchart of an automatic ICD encoding method based on fusion of multiple tag information provided in Example 1 of the present invention;

[0037] Figure 2 A framework diagram of the automatic ICD encoding method based on the fusion of multiple tag information provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0042] Example 1

[0043] like Figure 1-2 As shown, this embodiment provides an automatic ICD encoding method based on the fusion of multiple tag information, including the following steps:

[0044] S1 extracts features from the acquired clinical text and ICD label descriptions to obtain clinical text representation and label representation;

[0045] S2 interactively fuses the label representation with the clinical text representation to obtain the label-related context representation, extracts the co-occurrence representation of the label representation, and fuses it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation;

[0046] S3 interactively fuses the enhanced label representation with the clinical text representation to obtain the document representation corresponding to each label, and predicts the ICD code based on the document representation corresponding to each label;

[0047] S4 defines the loss function, optimizes the model parameters, and trains the automatic ICD coding model.

[0048] The goal of the automatic ICD coding task is to obtain the ICD coding label set from Select a specific set of codes to assign to the discharge summary , that is, for each label predict ,in and , representing the number of ICD codes and the number of words in the clinical text, respectively. , represents the set of discharge summaries. Our proposed architecture is as follows Figure 1 As shown in Figure 1, the model consists of a text encoder for extracting clinical text and label description features, a label information enhancer for aggregating context, label co-occurrence relations, and label descriptions, a Transformer decoder layer for generating label-specific representations, and a classifier for predicting the final ICD code. The following will introduce the proposed framework and the key modules in the framework in detail.

[0049] The clinical texts in this disclosure are extracted from electronic health records and undergo preliminary data cleaning and preprocessing to remove irrelevant information and correct format errors to ensure the quality and consistency of the input data.

[0050] We use Bi-LSTM layers to encode clinical text and ICD descriptions. Specifically, our model encodes clinical text consisting of a series of words As input, clinical text representation is obtained through a text encoder consisting of a pre-trained word embedding layer and an m-layer Bi-LSTM ∈ ,in Represents the dimensions of the last layer of Bi-LSTM:

[0051] (1)

[0052] Similarly, for tags ∈ C Corresponding descriptions, we obtain label embeddings of the same dimension through the same text encoder ∈ ,in Represents the number of words in the tag description, and then obtains the representation through maximum pooling ∈ :

[0053] (2)

[0054] Therefore, the label feature set can be expressed as .

[0055] By combining contextual information, clinical text and the relationship between them, a high-dimensional feature vector is generated. These features not only contain disease-related information, but also include individual differences and medical backgrounds of patients, thereby providing more comprehensive data support for subsequent model training. The cross-attention module built into the Transformer decoder can be used as a feature selector to obtain contextual features related to the label. At the same time, multi-head attention can capture multiple perspective features to help the model learn richer features. Our framework uses the Transformer decoder to replace the label-wise attention mechanism commonly used in previous ICD encoding work to obtain label-specific features, and use it as a mechanism to enhance label information. The attention layers in this disclosure all use a multi-head scaled dot product attention mechanism:

[0056] (3)

[0057] (4)

[0058] (5)

[0059] in, represents the dimension of the attention layer, Indicates The result of the attention head is , , ∈ Represents the parameters of the projection layer.

[0060] After obtaining the features of the clinical text and label description in the first stage, we use the label features as the query ∈ , Clinical Text Features H ∈ As keys and values, we pass through the first layer of standard Transforme decoder layers containing self-attention modules, criss-cross attention modules, and feed-forward neural networks to obtain contextual features related to the labels:

[0061] (6)

[0062] (7)

[0063] (8)

[0064] We take the output of the first Transformer decoder layer ∈ is considered as contextual information related to the label, where ∈ Yes Tags Related features, and They are the results of the self-attention module and the results of the cross-attention module, respectively.

[0065] Residual connections and layer normalization are used between each module in the decoder. Since we do not need to use a masking mechanism, the attention calculation process can be parallelized to improve the calculation speed.

[0066] Considering that when using the attention mechanism to fuse label and text features, more informative queries can help us obtain more features from the context that are helpful for classification, we carefully designed a label information enhancement module to enhance the label embedding as a query with the help of contextual information, label co-occurrence, and label original semantics.

[0067] Inspired by the idea of ​​prototype networks in meta-learning, we design a label prototype to preserve the information specific to each category from the sample. , we regard it as containing information related to a specific label from the context. , calculate the average value of all samples in a mini-batch and continuously update it during the training process. However, since each sample only corresponds to a part of the labels, if the sample does not contain the label , directly average all samples with labels C The relevant features will introduce noise that is irrelevant to the label because the category is not included in the context During training, we only select the label The tag-specific context features are calculated.

[0068] For a sample in a mini-batch, we calculate the category as follows The average value of:

[0069] (9)

[0070] in, ∈ , Representation sample Tags included in Related features, Indicates that a batch contains labels The number of samples.

[0071] Contains data from different samples and labels Then we iterate and update as follows:

[0072] (10)

[0073] (11)

[0074] Indicates the result calculated in the previous batch. Indicates that the previous calculation process included labels The number of samples. Finally, the label feature containing the sample context information is obtained , after each update, The value of is saved offline for the next calculation and the final inference stage, so the true label information of the sample is not required in the inference stage.

[0075] In clinical practice, it is common for other diseases to coexist with the primary disease or for the primary disease to cause other diseases. Therefore, disease codes also co-occur, and label dependency can also enhance the semantic information of labels.

[0076] To model label co-occurrence, we construct a The label co-occurrence matrix ,in Indicates that when the tag When it appears, the label The conditional probability of occurrence is calculated, and a two-layer graph convolutional neural network (GCN) is used to model the co-occurrence relationship between disease codes into a graph. The label co-occurrence matrix is ​​used as the adjacency matrix of the graph. GCN enhances label features by propagating messages between nodes, and can learn the deep connections between labels from the statistical relationship of label co-occurrence.

[0077] In a multi-layer GCN, each layer is based on label features and adjacency matrix As input, the former represents the nodes in the graph structure, the latter is the edge, and outputs the enhanced label features. Each layer can be represented as the following calculation process:

[0078] (12)

[0079] in, Indicates The node characteristics of the layer, , Initialized as the co-occurrence matrix, is the weight of each layer. In order to make the label feature dimensions the same, the dimension represented by the last layer node is .

[0080] We use the output of the last layer as the label feature containing co-occurrence information , among which Rows represent labels The correlation characteristics are expressed as .

[0081] So far we have obtained label representations containing two different types of information. , respectively, are label representations that are fused with contextual information through the cross-attention module , after GCN learns the label representation of label correlation They contain different aspects of the label features, and we aggregate them with the original label semantic features to obtain information-enhanced label features: (13)

[0082] in, ∈ represents the weight of the aggregation operation, and the label feature set after information enhancement is expressed as .

[0083] A transformer-based deep learning architecture is used to train with a large amount of labeled data. In this process, methods such as cross-validation and hyperparameter adjustment are used to optimize the performance of the model and improve the ability to parse clinical texts. After enhancing the label information, our goal is to use labels with richer information to represent To identify whether a part of the document is related to a certain label and assign the corresponding label to the document. We use the second layer of Transformer decoder to obtain document representations specific to each label, which is equivalent to enhancing the query information between two layers when using multiple layers of Transformer decoders. The second layer represents the enhanced label. As a query, the document represents As the key and value are taken as input, after the same decoding process as equations (6), (7), and (8), we will get the document features corresponding to each label .

[0084] In order to perform multi-label classification, we regard the prediction of each label as a binary classification task, and obtain the predicted value by linearly projecting the document features corresponding to each label:

[0085] (14)

[0086] in, ,and For a specific label in the linear layer We use multi-label binary cross entropy to train the model:

[0087] (15)

[0088] in, For the tags, For the The true value of the label, For the The document features corresponding to the labels are linearly projected to obtain the prediction results. A collection of tags.

[0089] The performance of the model is evaluated using a variety of evaluation indicators (such as accuracy, recall, and F1 score). Through continuous iteration and optimization, the model is ensured to be able to encode efficiently and accurately in practical applications. This method is widely applicable to the electronic health record systems of various medical institutions, especially when faced with large amounts of clinical data, and can significantly improve the efficiency and accuracy of automatic ICD coding. In addition, this technology can also be used in many fields such as medical research, health management, and insurance reimbursement, and has good market prospects.

[0090] Example 2

[0091] This embodiment provides an automatic ICD encoding system based on the fusion of multiple tag information, including:

[0092] The feature extraction module is configured to: extract features from the acquired clinical text and ICD label descriptions respectively to obtain clinical text representation and label representation;

[0093] The information enhancement module is configured to: interactively fuse the label representation with the clinical text representation to obtain the label-related context representation, extract the co-occurrence representation of the label representation, and fuse it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation;

[0094] The ICD coding module is configured to: interactively fuse the enhanced label representation with the clinical text representation, obtain the document representation corresponding to each label, and predict the ICD code based on the document representation corresponding to each label;

[0095] The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic ICD coding model.

[0096] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0097] In further embodiments, there is also provided:

[0098] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0099] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf 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 the processor may also be any conventional processor, etc.

[0100] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0101] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0102] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0103] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0104] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0105] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0106] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0107] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. 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 this application.

[0108] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. An automatic ICD encoding method based on the fusion of multiple tag information, characterized in that: The following steps are involved: Feature extraction is performed on the acquired clinical text and ICD label descriptions to obtain clinical text representation and label representation; Interactively fuse the label representation with the clinical text representation to obtain the label-related context representation, extract the co-occurrence representation of the label representation, and fuse it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation; Interactively fuse the enhanced label representation with the clinical text representation to obtain the document representation corresponding to each label, and predict the ICD code based on the document representation corresponding to each label; Define loss function, optimize model parameters, and train the automatic ICD coding model; The label representation is used as the query sequence, the clinical text representation is used as the key and value sequence, and the Transformer decoder including the self-attention module, the cross-attention module and the feedforward neural network is used to obtain the label-related context representation; Extract the co-occurrence representation of the tag representation, specifically: A multi-layer graph convolutional neural network is used to model the co-occurrence relationship between tags into a graph. The tag features are used as nodes in the graph structure, and the adjacency matrix is ​​used as the edges in the graph structure. The tag features are enhanced by propagating messages between nodes. The deep connection between each tag is learned from the statistical relationship of tag co-occurrence, and the co-occurrence representation of the tag representation is output. The self-attention module and the cross-attention module both use a multi-head scaled dot product attention mechanism. The multi-head attention can capture multiple perspective features and help the model learn richer features.

2. The automatic ICD encoding method based on fusion of multiple tag information as claimed in claim 1, characterized in that: The clinical texts were extracted from electronic health records and underwent preliminary data cleaning and preprocessing to remove irrelevant information and correct format errors to ensure the quality and consistency of the input data.

3. The automatic ICD encoding method based on fusion of multiple tag information as claimed in claim 1, characterized in that: The loss function of the automatic ICD coding model is defined as: in, For the tags, For the The true value of the label, For the The document features corresponding to the labels are linearly projected to obtain the prediction results. A collection of tags.

4. An automatic ICD encoding system based on fusion of multiple tag information, using an automatic ICD encoding method based on fusion of multiple tag information as described in any one of claims 1 to 3, characterized in that: include: The feature extraction module is configured to: extract features from the acquired clinical text and ICD label descriptions respectively to obtain clinical text representation and label representation; The information enhancement module is configured to: interactively fuse the label representation with the clinical text representation to obtain the label-related context representation, extract the co-occurrence representation of the label representation, and fuse it with the description of the ICD label and the label-related context representation to obtain the enhanced label representation; The ICD coding module is configured to: interactively fuse the enhanced label representation with the clinical text representation, obtain the document representation corresponding to each label, and predict the ICD code based on the document representation corresponding to each label; The model training module is configured to: define the loss function, optimize the model parameters, and train the automatic ICD coding model.

5. An electronic device, characterized in that: The invention comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 3 is completed.

6. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 4.

7. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 4.

Images