Diagnosis name recommendation method and system based on Fourier transform and message passing

Abstract

The present invention discloses a diagnosis name recommendation method and system based on Fourier transform and message passing, which relates to the technical field of real-time recommendation of diagnosis names. It can automatically capture diagnosis fragments when the doctor fills in the diagnosis process, extract the diagnosis information filled in by the doctor with the help of an encoder, and use Fourier transform to perform feature interaction on the extracted diagnosis information. Afterwards, the disease name alignment is designed according to the output characteristics of the encoder, and the standardized disease name and coding mapping are used to recommend the standardized disease name in real time, helping doctors to achieve the unification and standardization of disease names at the medical record filling end, and providing strong technical support for the formulation of diagnosis and treatment plans. The present invention also designs a message passing mechanism based on relative distance weights, so that the model takes into account the local and global feature extraction capabilities, and can promptly pass messages to the real-time captured diagnosis fragments, ensuring the real-time nature of the diagnosis name recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of real-time diagnosis name recommendation, and in particular to a diagnosis name recommendation method and system based on Fourier transform and message passing. 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] The development of medical information technology has generated a vast amount of unstructured medical record text data. Diagnostic information within medical records plays a crucial role in helping doctors formulate treatment plans. Furthermore, the advancement of ICD coding reform has exposed the serious impact of doctors' individualized disease descriptions on the quality of both their diagnoses and hospital medical record front pages. Furthermore, personalized diagnoses can be ambiguous, preventing patients from clearly understanding their condition. Therefore, using artificial intelligence to recommend standardized diagnoses based on personalized diagnoses is a pressing issue.

[0004] However, the inventors discovered that due to the individualized nature of clinician diagnostic writing and the lack of standardized standards, there are instances of incompatibility with the standard diagnostic coding library. Furthermore, existing coding technologies for encoding medical records suffer from shortcomings such as an inability to fully utilize key information from extracted features, an inability to track physicians' latest diagnostic records in real time, and cumbersome information processing. These shortcomings extend recognition and recommendation time, resulting in low recognition accuracy, poor real-time performance, and an inability to accurately recommend corresponding diagnostic names. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a diagnostic name recommendation method and system based on Fourier transform and message passing. This method automatically captures diagnostic snippets as doctors fill out their medical records, extracts the diagnostic information provided by the doctor using an encoder, and uses Fourier transforms to perform feature-based interaction on the extracted diagnostic information. Disease name alignment is then designed based on the encoder's output features. By mapping standardized disease names to their codes, standardized disease names are recommended in real time. This helps doctors unify and standardize disease names when filling out medical records, providing strong technical support for the development of treatment plans.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A first aspect of the present invention provides a method for recommending a diagnosis name based on Fourier transform and message passing, comprising the following steps:

[0008] Obtain diagnostic text in real time, perform rule matching on diagnostic segments in the diagnostic text, and record the latest characters;

[0009] The matched diagnostic fragments are input into the diagnostic feature extraction model to obtain the diagnostic features including the latest character feature vector;

[0010] Use standard diagnostic labels to perform real-time recognition of diagnostic categories for the latest character diagnostic features and obtain diagnostic name recommendations;

[0011] The specific steps of inputting the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic features include:

[0012] Vectorize the matched diagnostic fragments to obtain a vector representation;

[0013] Use bidirectional long short-term memory networks to extract semantic information represented by vectors;

[0014] The semantic information and interaction features between characters are extracted using an encoder based on Fourier transform;

[0015] The message passing mechanism is used to pass the latest character to obtain the diagnostic feature containing the feature vector of the latest character.

[0016] A second aspect of the present invention provides a diagnosis name recommendation system based on Fourier transform and message passing, comprising:

[0017] A data acquisition module is configured to acquire the diagnosis text in real time, perform rule matching on the diagnosis segments in the diagnosis text, and record the latest characters;

[0018] a feature extraction module configured to input the matched diagnostic fragment into a diagnostic feature extraction model to obtain a diagnostic feature including an updated character feature vector;

[0019] The diagnosis name recommendation module is configured to use standard diagnostic labels to perform real-time recognition of diagnostic categories on the diagnostic features of the latest characters to obtain diagnostic name recommendation results;

[0020] The specific steps of inputting the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic features include:

[0021] Vectorize the matched diagnostic fragments to obtain a vector representation;

[0022] Use bidirectional long short-term memory networks to extract semantic information represented by vectors;

[0023] The semantic information and interaction features between characters are extracted using an encoder based on Fourier transform;

[0024] The message passing mechanism is used to pass the latest character to obtain the diagnostic feature containing the feature vector of the latest character.

[0025] A third aspect of the present invention provides a medium having a program stored thereon, which, when executed by a processor, implements the steps of the diagnostic name recommendation method based on Fourier transform and message passing as described in the first aspect of the present invention.

[0026] The fourth aspect of the present invention provides a device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the diagnostic name recommendation method based on Fourier transform and message passing as described in the first aspect of the present invention are implemented.

[0027] One or more of the above technical solutions have the following beneficial effects:

[0028] This embodiment proposes a diagnosis name recommendation method and system based on Fourier transform and message passing, which can automatically capture diagnosis fragments when the doctor fills in the diagnosis process, and completes the extraction of diagnostic information with the help of Fourier encoder and message passing mechanism. Since the present invention uses parameter-free Fourier transform, the model parameters are greatly reduced, the running speed is accelerated, and the time for recognition and recommendation is reduced. Secondly, the present invention designs disease name alignment based on the output characteristics of the encoder, and recommends standardized disease names in real time through standardized disease names and coding mappings, helping doctors to achieve the unification and standardization of disease names at the medical record filling end, and providing strong technical support for the formulation of diagnosis and treatment plans. The present invention also designs a message passing mechanism based on relative distance weights, so that the model takes into account both local and global feature extraction capabilities, and can promptly pass messages to the diagnosis fragments captured in real time, ensuring the real-time nature of diagnosis name recommendations.

[0029] The present invention effectively extracts diagnostic features and realizes real-time recommendation based on Fourier transform and message passing, which greatly improves the standardization and quality of medical records when doctors fill in medical records. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings, 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.

[0031] Figure 1 This is a flowchart of a diagnostic name recommendation method based on Fourier transform and message passing in Embodiment 1 of the present invention;

[0032] Figure 2 This is an architectural diagram of the diagnostic feature extraction model in Example 1 of the present invention. DETAILED DESCRIPTION

[0033] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0034] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "include" and / or "comprising" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations;

[0035] Example 1:

[0036] The first embodiment of the present invention provides a diagnostic name recommendation method based on Fourier transform and message passing, such as Figure 1 As shown, based on the case text that the doctor is filling out, real-time matching of diagnostic fragments in the diagnosis text is performed. Then, the text fragments captured by the rule matching engine are input into the Fourier encoder to extract diagnostic features. For the disease fragment that the doctor is filling out, the doctor's personalized disease diagnosis is aligned to the national standard version of ICD10 disease name code through disease name alignment, and the top standard disease names with the greatest similarity are recommended.

[0037] The specific steps include:

[0038] Step 1: Obtain the diagnosis text in real time, match the diagnosis fragments in the diagnosis text with rules, and record the latest characters.

[0039] Step 2: Input the matched diagnostic fragment into the diagnostic feature extraction model to obtain the diagnostic features containing the latest character feature vector.

[0040] Step 3: Use standard diagnostic labels to perform real-time recognition of diagnostic categories on the diagnostic features of the latest characters to obtain diagnostic name recommendation results.

[0041] In step 1, while the doctor is filling out the medical record, the backend rule matching engine automatically captures text snippets containing clear diagnostic information, such as "admission diagnosis" and "preliminary diagnosis," from the admission record and initial medical history record. The specific capture range is shown in Table 1:

[0042] Table 1 Captured segments and capture ranges

[0043]

[0044] As doctors fill out their diagnoses in medical records, this embodiment uses regular expressions to match the diagnosis text in real time and records the latest characters. After acquiring the data, this embodiment sends the diagnosis text to the diagnostic feature extraction model via an asynchronous request, extracting diagnostic features through Fourier transforms and message passing.

[0045] In step 2, if Figure 2 As shown, the matching diagnostic fragment and the latest character are input into the diagnostic feature extraction model. The specific steps of obtaining the diagnostic features include:

[0046] (1) Vectorize the matched diagnostic fragments to obtain a vector representation.

[0047] After receiving the diagnostic fragment containing the latest characters matched by the rule engine, the background disease extraction service will vectorize the captured diagnostic fragment.

[0048] Specifically, a word embedding model is used to obtain the word vector representation of each character;

[0049] The word vector representation set of each character is represented as the vector representation of the diagnostic fragment text.

[0050] For example, given a diagnostic text segment "Discharge diagnosis: laryngitis, thyroid nodules, lung nodules", this example first uses the word embedding model Glove to obtain the word vector representation of each word. Where x represents a character vector, represents the real number space, d c Indicates the vector dimension. Then, we can get the vector representation of the diagnostic fragment text X={x1,…,x n}, where x n A vector representing the nth Chinese character.

[0051] (2) Use bidirectional long short-term memory networks to extract semantic information represented by vectors.

[0052] In order to extract the semantic features of characters, this embodiment uses a bidirectional long short-term memory network BiLSTM to extract the semantic information of the diagnostic fragment text. The extraction process is shown in formula (1):

[0053] h i =BiLSTM(x i ;θ),1≤i≤n (1),

[0054] in, represents the semantic feature representation of the i-th character, d h is the dimension of semantic features, and θ represents the parameters of BiLSTM.

[0055] (3) Use a Fourier transform-based encoder to extract the semantic interaction features between semantic information characters.

[0056] In order to extract the semantic interaction features between characters, this embodiment designs a new encoder based on Fourier transform, which extracts the interaction features between semantic features through parameter-free Fourier transform. Compared with the traditional neural network model based on the attention mechanism, the method of this embodiment can greatly reduce the number of model parameters and improve the feature extraction speed.

[0057] Specifically, the vector representation of semantic information is Fourier transformed to obtain the transformed feature vector;

[0058] The transformed feature vector is further subjected to a forward propagation normalization layer to extract character semantic information and obtain the final interaction feature.

[0059] Assume that, given the sequence characteristics d f is the characteristic dimension. In this embodiment, Fourier transform is performed using formula (2):

[0060]

[0061] Where k represents the index of the Fourier coefficient, F(k) represents the transformed Fourier coefficient, which consists of a real part a and an imaginary part b, in the form of F(k)=a+bj, f l represents the lth eigenvalue of feature f, and j represents an imaginary number. In this embodiment, the transformed amplitude is used as the eigenvalue.

[0062] Afterwards, this embodiment calculates the modulus M(k) of each Fourier coefficient F(k) using formula (3):

[0063]

[0064] where a k The real part of the kth Fourier coefficient, b k Represents the imaginary part of the kth Fourier coefficient. Then we get the transformed eigenvector

[0065] According to the above Fourier transform formula, this embodiment first transforms the output features of BiLSTM in the row direction Perform Fourier transform to obtain the interaction features within the character features, and then perform Fourier transform according to the column direction to obtain the interaction features between character features. The implementation process is shown in formula (4):

[0066] H Fourier =Fourier 列 (Fourier 行 (H)) (4),

[0067] in is the characteristic matrix after Fourier transform, 列 and Fourier 行 They are column-wise Fourier transform operations and row-wise Fourier transform operations respectively.

[0068] Afterwards, in order to further extract features, this embodiment further extracts character semantic information through a forward propagation normalization layer. The extraction process is shown in formula (5):

[0069] H o =LayerNorm(ELU(H Fourier W o +η)) (5),

[0070] in, is the output feature of the Fourier encoder, where d o is the feature dimension, LayerNorm represents layer normalization, and ELU is an activation function. is the weight parameter, is the bias vector.

[0071] (4) The message passing mechanism is used to pass the latest character and obtain the diagnostic features containing the feature vector of the latest character.

[0072] To ensure the real-time nature of recommended diagnosis names, this embodiment abandons the traditional method of first predicting the disease sequence and then aligning it. Instead, a distance-based message passing mechanism is designed to aggregate text information to the latest characters being filled in by the doctor. Real-time recommendations of standard disease names are achieved by performing multi-label classification on the latest character features.

[0073] Specifically, the relative position weight of the latest character and other characters is calculated based on the position of the latest character;

[0074] The spatial domain graph neural network is used to obtain diagnostic features containing the latest character feature vectors.

[0075] Define the position of the latest character filled in by the doctor as p new , this embodiment calculates the relative position weight of the latest character and other characters by formula (6):

[0076]

[0077] Among them, score q Indicates the character at q and the latest character p newThe relative position weight score is calculated to fill the corresponding weight score into the corresponding position of the adjacency matrix to obtain the adjacency matrix A. abs() represents the absolute value operation, p q Indicates the position of the q character.

[0078] In order to realize the message transmission in the diagnosis text, this embodiment introduces the spatial domain graph neural network, connects the latest character filled in by the doctor with other characters, and calculates the relative position weight score. q Fill in the corresponding positions in the adjacency matrix to obtain the adjacency matrix based on relative position weights and the degree matrix This embodiment then implements message delivery within the diagnostic text using formula (7):

[0079]

[0080] in, is the output feature of the graph neural network, d g Feature dimension, W g is the weight parameter.

[0081] In step 3, this embodiment uses the ICD10 standard diagnosis issued by the National Medical Insurance Administration as the label, which contains 37,293 different diagnoses. This embodiment extracts the feature vector corresponding to the latest character filled in by the doctor from the feature matrix U Then, formula (8) is used to achieve real-time identification of diagnostic categories:

[0082]

[0083] in is the predicted standard diagnostic label, sigmoid is an activation function is the weight parameter.

[0084] Afterwards, this embodiment calculates the loss Loss of the diagnosis recognition model using formula (9):

[0085]

[0086] Among them, |Ω| represents the number of samples in the training set, Class t represents the number of categories of the t-th sample, Represents the loss value of the cth class in the tth sample.

[0087] In this embodiment, Calculated by formula (10):

[0088]

[0089] Among them, αc Hyperparameters, is the true label of the cth class in the tth sample, is the predicted label for the cth class in the tth sample. After obtaining the loss value, this embodiment optimizes the model parameters through gradient backpropagation. Finally, this embodiment returns the top most similar normalized diagnosis names for the doctor to select.

[0090] This embodiment proposes a real-time recommendation method for standardized diagnosis names based on Fourier encoders. The present invention can automatically capture diagnosis fragments when doctors fill in the diagnosis process, and completes the extraction of diagnostic information with the help of Fourier encoders and message passing mechanisms. Since this embodiment uses parameter-free Fourier transform, the model parameters are greatly reduced and the running speed is accelerated. Secondly, the message passing mechanism based on relative distance weights enables the model to take into account both local and global feature extraction capabilities. In actual use, the recommendation time of this embodiment is shortened to milliseconds, and an example is shown in Table 2. In summary, the present invention effectively extracts diagnostic features and realizes real-time recommendations based on Fourier transform and message passing, which greatly improves the standardization and medical record quality of doctors filling in medical records.

[0091] Table 2 Example of recommendation results

[0092]

[0093] Example 2:

[0094] A second embodiment of the present invention provides a diagnosis name recommendation system based on Fourier transform and message passing, including:

[0095] A diagnosis name recommendation system based on Fourier transform and message passing, including:

[0096] A data acquisition module is configured to acquire the diagnosis text in real time, perform rule matching on the diagnosis segments in the diagnosis text, and record the latest characters;

[0097] a feature extraction module configured to input the matched diagnostic fragment into a diagnostic feature extraction model to obtain a diagnostic feature including an updated character feature vector;

[0098] The diagnosis name recommendation module is configured to use standard diagnostic labels to perform real-time recognition of diagnostic categories on the diagnostic features of the latest characters to obtain diagnostic name recommendation results;

[0099] The specific steps of inputting the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic features include:

[0100] Vectorize the matched diagnostic fragments to obtain a vector representation;

[0101] Use bidirectional long short-term memory networks to extract semantic information represented by vectors;

[0102] The semantic information and interaction features between characters are extracted using an encoder based on Fourier transform;

[0103] The message passing mechanism is used to pass the latest character to obtain the diagnostic feature containing the feature vector of the latest character.

[0104] Example 3:

[0105] A third embodiment of the present invention provides a medium having a program stored thereon. When the program is executed by a processor, the steps of the diagnostic name recommendation method based on Fourier transform and message passing as described in the first embodiment of the present invention are implemented.

[0106] Example 4:

[0107] Embodiment 4 of the present invention provides a device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the diagnostic name recommendation method based on Fourier transform and message passing as described in Embodiment 1 of the present invention are implemented.

[0108] The steps involved in the above embodiments 2, 3, and 4 correspond to those in embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0109] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0110] Although the above describes the specific embodiments 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 any creative work are still within the scope of protection of the present invention.

Claims

1. A diagnostic name recommendation method based on Fourier transform and message passing, characterized in that: The following steps are involved: Obtain diagnostic text in real time, perform rule matching on diagnostic segments in the diagnostic text, and record the latest characters; Input the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic feature including the latest character feature vector; Using standard diagnostic labels, the diagnostic features containing the latest character feature vectors are used to perform real-time identification of diagnostic categories and obtain diagnostic name recommendations. The specific steps of inputting the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic feature including the latest character feature vector include: Vectorize the matched diagnostic fragments to obtain a vector representation; Use bidirectional long short-term memory networks to extract semantic information represented by vectors; The semantic information and interaction features between characters are extracted using a Fourier transform-based encoder. The specific steps include: Perform Fourier transform on the vector representation of semantic information to obtain the transformed feature vector. Specifically, first perform Fourier transform on the semantic information output by the bidirectional long short-term memory network in the row direction to obtain the interaction features within the character features, and then perform Fourier transform on the column direction to obtain the interaction features between the character features. The transformed feature vector is further extracted through the forward propagation normalization layer to obtain the final interactive features; The message passing mechanism is used to pass the latest character to obtain the diagnostic feature containing the feature vector of the latest character.

2. The diagnostic name recommendation method based on Fourier transform and message passing according to claim 1, characterized in that: The specific steps for vectorizing the matched diagnostic fragments to obtain a vector representation include: Use word embedding model to obtain word vector representation for each character; The word vector representation set of each character is represented as the vector representation of the diagnostic fragment text.

3. The diagnostic name recommendation method based on Fourier transform and message passing according to claim 1, characterized in that: The specific steps of using the message passing mechanism to pass the latest character and obtain the diagnostic features including the latest character feature vector include: Calculate the relative position weight of the latest character to other characters based on the position of the latest character; The spatial domain graph neural network is used to obtain diagnostic features containing the latest character feature vectors.

4. The diagnostic name recommendation method based on Fourier transform and message passing according to claim 3, characterized in that: The specific steps of using spatial domain graph neural network to obtain diagnostic features containing the latest character feature vectors include: The latest character is connected to other characters by edges, and the relative position weight scores are filled into the corresponding positions in the adjacency matrix to obtain the adjacency matrix and degree matrix based on the relative position weights, and then the diagnostic feature matrix containing the feature vector of the latest character is obtained.

5. The diagnostic name recommendation method based on Fourier transform and message passing according to claim 1, characterized in that: Use standard diagnostic labels to identify the diagnostic category of the latest character in real time, and return to the previous The most similar normalized diagnosis names are used for selection to obtain the diagnosis name recommendation results.

6. A diagnostic name recommendation system based on Fourier transform and message passing, characterized in that: include: A data acquisition module is configured to acquire the diagnosis text in real time, perform rule matching on the diagnosis segments in the diagnosis text, and record the latest characters; a feature extraction module configured to input the matched diagnostic fragment and the latest character into a diagnostic feature extraction model to obtain a diagnostic feature including a feature vector of the latest character; A diagnosis name recommendation module is configured to use standard diagnosis labels to perform real-time recognition of diagnosis categories on the diagnosis features containing the latest character feature vectors to obtain diagnosis name recommendation results; The specific steps of inputting the matched diagnostic fragment and the latest character into the diagnostic feature extraction model to obtain the diagnostic feature including the latest character feature vector include: Vectorize the matched diagnostic fragments to obtain a vector representation; Use bidirectional long short-term memory networks to extract semantic information represented by vectors; The semantic information and interaction features between characters are extracted using a Fourier transform-based encoder. The specific steps include: Perform Fourier transform on the vector representation of semantic information to obtain the transformed feature vector. Specifically, first perform Fourier transform on the semantic information output by the bidirectional long short-term memory network in the row direction to obtain the interaction features within the character features, and then perform Fourier transform on the column direction to obtain the interaction features between the character features. The transformed feature vector is further extracted through the forward propagation normalization layer to obtain the final interactive features; The message passing mechanism is used to pass the latest character to obtain the diagnostic feature containing the feature vector of the latest character.

7. A computer-readable storage medium, characterized in that A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the diagnosis name recommendation method based on Fourier transform and message passing according to any one of claims 1 to 5.

8. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the diagnostic name recommendation method based on Fourier transform and message passing according to any one of claims 1-5.

Images