Operation ICD code generation method

Abstract

The invention provides a surgical ICD code generation method, which comprises the following steps: collecting coded and archived surgical records, splitting training set data and test set data, and constructing an AI surgical ICD code model; training the AI operation ICD coding model by using the training set data, and testing the AI operation ICD coding model by using the test set data, so that the AI operation ICD coding model reaches a preset training condition; and based on the trained AI operation ICD coding model, identifying the operation record of the unarchived medical record to generate an operation ICD code corresponding to the operation record of the unarchived medical record. According to the technical scheme provided by the invention, the ICD coding speed and accuracy of the operation are improved, the ICD coding management cost of the operation is greatly reduced, and the user experience is improved.

Description

Technical Field

[0001] The present application relates to the field of medical coding technology, and in particular to a method for generating ICD codes for surgery. Background Art

[0002] The International Classification of Diseases (ICD) is an internationally unified disease classification method developed by the WHO. It classifies diseases according to their etiology, pathology, clinical manifestations, anatomical location and other characteristics, making them an orderly combination and a system represented by a coding method.

[0003] During the research process of conceiving and forming this application, the applicant discovered at least the following problems: during the operation in the hospital, the doctor needs to describe in the diagnosis and treatment record what disease the patient has, what kind of operation was performed on the patient, and what kind of treatment was performed. This is a description of the patient's condition and treatment process in natural language. The diagnosis and treatment records will eventually be transferred to the medical record room, and the medical record room staff will code the disease according to the records in the diagnosis and treatment files. Generally speaking, doctors are the ones who understand patients best, and most of the coding personnel do not understand medical care and patients, so there are many inaccuracies in their coding. This often causes a disconnect between the diagnosis and treatment situation and the disease coding. Summary of the invention

[0004] In order to alleviate the above problems, the present application provides a method for generating ICD codes for surgery, comprising: Collect coded and archived surgical records, split them into training set data and test set data, and build an AI surgical ICD coding model; Using the training set data to train the AI ​​surgery ICD coding model, and using the test set data to test the AI ​​surgery ICD coding model, so that the AI ​​surgery ICD coding model meets the preset training conditions; Based on the trained AI surgical ICD coding model, the surgical records of unarchived medical records are identified to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records.

[0005] Optionally, the process of collecting the coded and archived surgical records, splitting the training set data and the test set data, and building the AI ​​surgical ICD coding model also includes: Based on the coded and archived surgical records, data preprocessing is performed on the coded and archived surgical records of the medical records to obtain a sequence of surgical-related medical vocabulary to be marked; The surgery-related medical vocabulary sequence to be labeled is divided into training set data and test set data, and an AI surgery ICD coding model is constructed. The AI ​​surgery ICD coding model includes an input layer, a hidden layer and an output layer.

[0006] Optionally, the process of performing data preprocessing on the coded and archived surgical records to obtain a sequence of surgical-related medical vocabulary to be marked includes: Performing data extraction, data cleaning, and data specification preprocessing on the coded and archived surgical records, and removing common descriptions, stop words, and common vocabulary during the data cleaning process; The pre-processed surgical records are visually analyzed to determine the optimal length of the medical vocabulary sequence related to the surgery.

[0007] Optionally, the process of visually analyzing the pre-processed surgical records includes: The coded and archived surgical records are subjected to word segmentation and text analysis to accurately identify the surgical records.

[0008] Optionally, the process of performing word segmentation and text analysis on the coded and archived surgical records includes: Performing word segmentation on the preprocessed surgical records, dividing continuous character strings in the surgical records into words or phrases; Medical part-of-speech tagging is performed on each word or phrase to identify surgery-related entities in the text; Based on the identified surgery-related entities, the dependencies between words in each sentence are analyzed and the semantic role of each word in the sentence is determined; According to the semantic roles and dependencies, entity relationships between each operation-related entity are extracted from the operation record.

[0009] Optionally, the process of dividing the surgery-related medical vocabulary sequence to be labeled into training set data and test set data, and constructing an AI surgery ICD coding model, wherein the AI ​​surgery ICD coding model includes an input layer, a hidden layer, and an output layer, further includes: Extracting surgical feature values ​​such as surgical site feature value, surgical procedure feature value, approach feature value, filler feature value and personnel feature value based on the surgical related medical vocabulary sequence in the training set data; Determine the nodes of the input layer and the hidden layer, wherein the input layer includes a first input node, a second input node, a third input node, a fourth input node and a fifth input node, the first input node is used to input the surgical site characteristic value, the second input node is used to input the procedure characteristic value, the third input node is used to input the approach characteristic value, the fourth input node is used to input the filler characteristic value, the fifth input node is used to input the personnel characteristic value, and the nodes of the hidden layer are composed of Gaussian kernel functions.

[0010] Optionally, the AI ​​surgery ICD coding model includes an embedding layer; the step of training the AI ​​surgery ICD coding model using the training set data, and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets preset training conditions includes: Converting the surgery-related medical vocabulary sequence of the training set into a numerical representation vector in the embedding layer to capture the semantic information of the surgery-related medical vocabulary sequence; Inputting the representation vector and the hidden state of the previous time step into the hidden layer, and updating the hidden state of the hidden layer by a preset nonlinear function; Based on the hidden state at each time step, the temporal information of the surgery-related medical vocabulary sequence is captured; A gating mechanism is introduced to control the information flow inside the hidden layer to effectively learn the long-term dependencies between medical words; Using the gradient clipping technology, when the memory gradient of the hidden layer exceeds the preset threshold, the gradient exceeding the preset time length is clipped in chronological order to ensure the stability of the training process.

[0011] Optionally, the process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets preset training conditions also includes: The gradient of the parameters of the surgery-related medical vocabulary sequence of the training set is calculated using a back-propagation algorithm, and the parameters of the surgery-related medical vocabulary sequence of the training set are updated using a gradient descent algorithm to minimize the prediction error of the output feature representation, wherein the feature representation includes key features and the temporal relationship between the key features.

[0012] Optionally, the process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets preset training conditions also includes: The AI ​​surgery ICD coding model is verified using the surgery-related medical vocabulary sequence of the test set data, and the network structure and parameters of the AI ​​surgery ICD coding model are cyclically adjusted until the preset training conditions are met.

[0013] Optionally, the process of using the surgery-related medical vocabulary sequence of the test set data to verify the AI ​​surgery ICD coding model and cyclically adjusting the network structure and parameters of the AI ​​surgery ICD coding model includes: The difference between the coding recognition result and the actual medical record coding is calculated by the cross entropy loss function, the gradient of the difference to the model parameters is calculated by the back propagation algorithm, and the weights and bias parameters of the convolutional layer and the pooling layer are updated using the SGD optimizer for the gradient.

[0014] The method for generating ICD codes for surgery provided in the present application collects coded and archived surgical records, splits them into training set data and test set data, and constructs an AI surgical ICD coding model; uses the training set data to train the AI ​​surgical ICD coding model, and uses the test set data to test the AI ​​surgical ICD coding model, so that the AI ​​surgical ICD coding model meets preset training conditions; based on the trained AI surgical ICD coding model, identifies surgical records of unarchived medical records to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records, thereby improving the speed and accuracy of ICD coding for surgery, significantly reducing the ICD coding management cost for surgery, and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.

[0016] Figure 1 This is a flow chart of a method for generating ICD codes for surgery according to an embodiment of the present application.

[0017] The realization of the purpose, functional features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. The above-mentioned drawings have shown clear embodiments of this application, which will be described in more detail later. These drawings and textual descriptions are not intended to limit the scope of the concept of this application in any way, but to illustrate the concept of this application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0018] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0019] It should be noted that, in this article, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element. In addition, components, features, and elements with the same name in different embodiments of the present application may have the same meaning or different meanings, and their specific meanings need to be determined by their explanation in the specific embodiment or further combined with the context of the specific embodiment.

[0020] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] First embodiment This application provides a method for generating ICD codes for surgery. Figure 1 This is a flow chart of a method for generating ICD codes for surgery according to an embodiment of the present application.

[0022] like Figure 1 As shown, in one embodiment, the method for generating ICD code for surgery includes: S10: Collect the coded and archived surgical records, split them into training set data and test set data, and build an AI surgical ICD coding model.

[0023] For example, the International Classification of Diseases (ICD) is a system that classifies diseases according to rules based on certain characteristics of the diseases and represents them in a coding method. Electronic medical record data is collected by manual input, scanning paper medical records, importing other system data, etc., where the surgical record data includes the patient's personal information, surgical site, surgical procedure, approach, filler and other surgical related information, thereby obtaining a surgical record data sequence. The surgical record data sequence contains a variety of character types, including numbers, Chinese characters, letters, and special symbols such as punctuation marks. There is a known correspondence between the surgical records of the training set and the test set and the surgical ICD codes.

[0024] S20: Use the training set data to train the AI ​​surgery ICD coding model, and use the test set data to test the AI ​​surgery ICD coding model, so that the AI ​​surgery ICD coding model meets the preset training conditions.

[0025] Exemplarily, the training process of a large natural language model may include a pre-training and adjustment phase. By learning a large amount of unlabeled text data and labeled data for specific tasks, the model can master the basic structure and semantic laws of the language. During the training process, technologies such as Transformer architecture, parameter sharing, and regularization are used to achieve efficient and reliable training. The training process also involves model verification, evaluation, and tuning to ensure that the final model can provide high-quality natural language generation and understanding capabilities. Based on surgical records that already contain corresponding surgical ICD coding training sets and test sets, the AI ​​surgical ICD coding model can be trained to achieve the training purpose of accurate and efficient ICD coding for surgical records.

[0026] S30: Based on the trained AI surgical ICD coding model, the surgical records of the unarchived medical records are identified to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records.

[0027] Exemplarily, after the AI ​​surgical ICD coding model is trained, the surgical records of unarchived medical records are identified and the corresponding ICD codes are automatically generated.

[0028] The method for generating ICD codes for surgery provided in the present application collects coded and archived surgical records, splits them into training set data and test set data, and constructs an AI surgical ICD coding model; uses the training set data to train the AI ​​surgical ICD coding model, and uses the test set data to test the AI ​​surgical ICD coding model, so that the AI ​​surgical ICD coding model meets preset training conditions; based on the trained AI surgical ICD coding model, identifies surgical records of unarchived medical records to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records, which can improve the speed and accuracy of ICD coding for surgery, greatly reduce the ICD coding management cost for surgery, and improve user experience.

[0029] Optionally, the process of collecting the coded and archived surgical records, splitting the training set data and the test set data, and building the AI ​​surgical ICD coding model also includes: Based on the coded and archived surgical records, data preprocessing is performed on the coded and archived surgical records of the medical records to obtain a sequence of surgical-related medical vocabulary to be marked; The surgery-related medical vocabulary sequence to be labeled is divided into training set data and test set data, and an AI surgery ICD coding model is constructed. The AI ​​surgery ICD coding model includes an input layer, a hidden layer and an output layer.

[0030] For example, to ensure the quality of training, surgical records can be required to conform to basic specifications in form, such as using standardized document formats, clear layout structures, and standardized medical language expressions. In addition, due to the differences in writing habits and specialty characteristics of different doctors, the original medical records obtained may have large differences in content organization and detailed descriptions. Therefore, with the help of data preprocessing technology, medical record documents can be preliminarily structured and analyzed, and key information can be extracted and corrected, and converted into a standardized form of surgical-related medical vocabulary sequences to be marked for further analysis.

[0031] The input layer of a neural network is the first layer of the neural network. It is the only layer that interacts with the outside world. In a neural network, each neuron in the input layer corresponds to a feature in the input data. The number of neurons in the input layer depends on the number of features in the input data. For example, in a classification task, each pixel is a neuron in the input layer. A neural network can only have one input layer, which is the first layer of the neural network. Its function is to convert the input data into a format that can be processed by the neural network. The number of neurons in the input layer depends on the number of features in the input data. The function of the input layer is to convert the input data into a format that can be processed by the neural network, such as converting surgical record data into vector form. The output of the input layer is the input passed to the next layer of neurons. The next layer here can be a hidden layer or an output layer. Hidden layers are all layers in a neural network except the input layer and the output layer. Their function is to convert input data into output data. The output layer is the last layer in a neural network. Its output is the prediction or classification result of the neural network for the input data.

[0032] The hidden layer of a neural network is one or more layers located between the input layer and the output layer. Its function is to transform the input data into a higher-level feature representation. The number of neurons and the way they are connected in each hidden layer vary, depending on the specific neural network architecture. In a convolutional neural network (CNN), the convolutional layer and the pooling layer constitute the hidden layer. In a recurrent neural network (RNN), the recurrent layer constitutes the hidden layer. The more neurons in the hidden layer, the stronger the representation ability of the neural network.

[0033] The output layer of the neural network is the last layer in the neural network. Its output is the prediction or classification result of the neural network for the input data. The design of the output layer needs to be adjusted according to the specific task to adapt to different application scenarios.

[0034] In classification encoding tasks, the output layer can use the Softmax function to calculate the probability distribution of each category. For example, for a k-class classification problem, the number of neurons in the output layer should be k, each neuron corresponds to a category, and its output value represents the probability of the category. The Softmax function can normalize the output value of the neuron to a probability distribution, so that the probability value of each category is between 0 and 1, and the sum is 1.

[0035] Optionally, the process of performing data preprocessing on the coded and archived surgical records to obtain a sequence of surgical-related medical vocabulary to be marked includes: Performing data extraction, data cleaning, and data specification preprocessing on the coded and archived surgical records; The pre-processed surgical records are visually analyzed to determine the optimal length of the medical vocabulary sequence related to the surgery.

[0036] Data extraction is the process of extracting data from a data source, mainly in the form of full extraction and incremental extraction. Data cleaning – The process of re-examining and verifying data in order to remove duplicate information, correct existing errors, and provide data consistency. Data cleaning refers to the final procedure of discovering and correcting identifiable errors in data files, including checking data consistency, handling invalid values ​​and missing values. In the process of data cleaning, it can be used to remove common descriptions, stop words and common words. Data conventions are a series of conventions formulated to enable the correct completion of data transmission between multiplexed data ports.

[0037] Visual analysis is an analysis method that is mainly used for massive data association analysis. It can assist in the association analysis of data and make complete analysis charts. With the help of a powerful visual data analysis platform, it can assist in the association analysis of surgical record data and make complete analysis results. The analysis results can include relevant information of all events, and can also fully display the process of surgical record data analysis and the direction of the data chain. In order to better and more conveniently construct the entire data set index, big data visualization analysis can be performed on data sets from different sources to determine and set the optimal text sequence length as the standard for sentence filling length in the subsequent model.

[0038] Optionally, the process of visually analyzing the pre-processed surgical records includes: The coded and archived surgical records are subjected to word segmentation and text analysis to accurately identify the surgical records.

[0039] Exemplarily, the surgical records are identified by segmenting and tagging them, and then text content analysis is performed, thereby performing relationship analysis through entity recognition to accurately identify the content of the surgical records.

[0040] Optionally, the process of performing word segmentation and text analysis on the coded and archived surgical records includes: Performing word segmentation on the preprocessed surgical records, dividing continuous character strings in the surgical records into words or phrases; Medical part-of-speech tagging is performed on each word or phrase to identify surgery-related entities in the text; Based on the identified surgery-related entities, the dependencies between words in each sentence are analyzed and the semantic role of each word in the sentence is determined; According to the semantic roles and dependencies, entity relationships between each operation-related entity are extracted from the operation record.

[0041] Exemplarily, the continuous character string in the surgical record can be segmented into words or phrases. Then the part of speech is determined for each segmented word, such as noun, verb, adjective, etc. For example, the segmentation of "the patient received thoracotomy after admission" is ["patient", "at", "after admission", "receive", "thoracotomy", "treatment"]. Then the segmented words are annotated as ["patient / noun", "at / preposition", "after admission / adverb", "receive / verb", "thoracotomy / noun", "treatment / verb"].

[0042] After word segmentation and tagging, we can perform syntactic analysis through entity recognition to analyze the dependency relationship between words in the sentence, such as subject-predicate relationship, verb-object relationship, etc. We can determine the semantic role of each word in the sentence, such as subject, object, attributive, etc. For example, the relationship between "accept" and "open-chest surgery" is a verb-object relationship, and we can determine that "open-chest surgery" is the object of "accept".

[0043] For example, the main contents of the surgical record can be extracted to form a summary, and the surgical record can be divided into different categories, such as surgical name, site, procedure, approach, filler, etc.; the relationship between entities can be extracted from the surgical record. For example, the relationship between the surgical name and the surgical site can be extracted from the surgical record.

[0044] Optionally, the process of dividing the surgery-related medical vocabulary sequence to be labeled into training set data and test set data, and constructing an AI surgery ICD coding model, wherein the AI ​​surgery ICD coding model includes an input layer, a hidden layer, and an output layer, further includes: Extracting surgical feature values ​​such as surgical site feature value, surgical procedure feature value, approach feature value, filler feature value and personnel feature value based on the surgical related medical vocabulary sequence in the training set data; Determine the nodes of the input layer and the hidden layer, wherein the input layer includes a first input node, a second input node, a third input node, a fourth input node and a fifth input node, the first input node is used to input the surgical site characteristic value, the second input node is used to input the procedure characteristic value, the third input node is used to input the approach characteristic value, the fourth input node is used to input the filler characteristic value, the fifth input node is used to input the personnel characteristic value, and the nodes of the hidden layer are composed of Gaussian kernel functions.

[0045] Exemplarily, the surgical site feature value is the feature data for the body part involved in this operation. The surgical procedure feature value is the feature data for the surgical method used in this operation. In the medical field, a surgical procedure specifically refers to a series of specific operating steps and techniques used when performing an operation. These operating methods vary depending on the type of operation, the patient's condition, and the doctor's professional skills and experience. There are multiple surgical procedures depending on different surgical needs. For example, the same operation may use different surgical procedures due to different doctors or the characteristics of the medical center. These different surgical procedures may involve the choice of surgical approach, the use of surgical instruments, the order of surgical operations, etc.

[0046] For example, the approach feature value refers to the feature data of the path chosen by the doctor to enter the body cavity or body surface during the operation. Common surgical approaches include open surgery, laparoscopic surgery, and interventional surgery. The filler feature value refers to the feature data of medical materials that are filled into the human body subcutaneously through surgery or injection to achieve the purpose of replenishing volume and making the appearance more beautiful. The staff feature value refers to the feature data of the staff involved in the operation, such as doctors, caregivers, or nurses.

[0047] The input value of the input layer is the anchor text feature vector of the surgical record feature, and the anchor text feature vector is represented as a multi-dimensional matrix. The Gaussian kernel function generally refers to the radial basis function. The radial basis function is a real-valued function whose value depends only on the distance from the origin. In this embodiment, the five features of the surgical record are input separately through five nodes, and the Gaussian kernel function of the hidden layer is passed. A set of two-dimensional Gaussian kernels is generated according to the preset standard deviation, and the generated Gaussian kernel is convolved with the anchor text feature vector, and each set of vectors is weighted averaged. The data obtained by the Gaussian kernel convolution replaces the corresponding vector value of the feature vector in the original surgical record to smooth the pooling curve, so that the model can accurately output the AI ​​surgical ICD code corresponding to the identified surgical record.

[0048] Optionally, the AI ​​surgery ICD coding model includes an embedding layer; the embedding layer is connected between the input layer and the hidden layer.

[0049] Optionally, the step of using the training set data to train the AI ​​surgery ICD coding model, and using the test set data to test the AI ​​surgery ICD coding model so that the AI ​​surgery ICD coding model meets preset training conditions includes: Converting the surgery-related medical vocabulary sequence of the training set into a numerical representation vector in the embedding layer to capture the semantic information of the surgery-related medical vocabulary sequence; Inputting the representation vector and the hidden state of the previous time step into the hidden layer, and updating the hidden state of the hidden layer by a preset nonlinear function; Based on the hidden state at each time step, the temporal information of the surgery-related medical vocabulary sequence is captured; A gating mechanism is introduced to control the information flow inside the hidden layer to effectively learn the long-term dependencies between medical words; Using the gradient clipping technology, when the memory gradient of the hidden layer exceeds the preset threshold, the gradient exceeding the preset time length is clipped in chronological order to ensure the stability of the training process.

[0050] In the recurrent neural network (RNN) model, the process of hidden layer processing surgery-related medical vocabulary sequences mainly involves key technologies such as forward propagation, the role of memory units and gating mechanisms, and gradient clipping. The surgery-related medical vocabulary sequence data flows from the input layer to the output layer, and is processed by each hidden layer on the way. Each hidden layer extracts and transforms the input data and finally generates an output. For the surgery-related medical vocabulary sequence, RNN processes the vocabulary one by one in the order of the sequence, and uses the ability to process time series data that feedforward neural networks cannot have to pass the information of the previous vocabulary to the processing of the next vocabulary.

[0051] For example, the vocabulary can first be converted into a numerical representation vector, which is usually done through an embedding layer. The embedding layer maps the sequence of surgical medical vocabulary to a vector in a high-dimensional space that captures the semantic information of the vocabulary. This representation vector is then input into the hidden layer along with the hidden state of the previous time step. The hidden layer uses this information to update its state. This update process is done through a nonlinear function that takes into account the current input and the previous hidden state.

[0052] For example, the hidden layer of RNN usually contains memory cells, which can store and transmit information from previous moments, allowing the network to handle long-term dependency problems. When processing a sequence of surgical-related medical vocabulary, the memory cells will remember the information of previous vocabulary, such as the semantics of the vocabulary, contextual relationships, etc., and use it when processing subsequent vocabulary.

[0053] For example, at each time step, the hidden layer not only updates its state, but also potentially generates an output. This output can be used for various tasks, such as predicting the next word or classification tasks.

[0054] For example, gating mechanisms (such as the forget gate, input gate, and output gate in LSTM) further enhance the ability of RNN to handle long-term dependencies by controlling the flow of information to avoid the gradient vanishing or exploding problem. Due to the design of RNN, it updates its hidden state at every time step, which enables it to capture the timing information in the sequence. For example, when processing surgical steps, RNN is able to "remember" previous steps, which may be crucial to understanding the entire surgical process.

[0055] For example, during training, since the sequence of surgery-related medical vocabulary may be long, RNN may encounter the problem of gradient explosion. To solve this problem, gradient clipping technology is usually used.

[0056] Gradient clipping sets a threshold, and when the absolute value of the gradient exceeds this threshold, it is clipped to the size of the threshold. This can prevent gradient explosion and ensure the stability of training.

[0057] For example, two special RNN variants, LSTM or GRU, can be used to better control the flow of information by introducing a gating mechanism, so that long-term dependencies can be effectively learned, thereby solving the problem of gradient disappearance or gradient explosion.

[0058] In summary, when processing surgery-related medical vocabulary sequences, the recurrent neural network model can effectively capture the information in the sequence and make accurate predictions or classifications through key technologies such as forward propagation, the role of memory units and gating mechanisms, and gradient clipping.

[0059] Optionally, the process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets preset training conditions also includes: The gradient of the parameters of the surgery-related medical vocabulary sequence of the training set is calculated using a back-propagation algorithm, and the parameters of the surgery-related medical vocabulary sequence of the training set are updated using a gradient descent algorithm to minimize the prediction error of the output feature representation, wherein the feature representation includes key features and the temporal relationship between the key features.

[0060] For example, after processing the entire sequence, the RNN uses the back-propagation algorithm to calculate the gradient of its parameters. These parameters are then updated by an optimization algorithm (such as gradient descent) to minimize the prediction error. In this way, the hidden layer of the RNN is able to process surgical-related medical vocabulary sequences, learn the key features and temporal relationships in the sequence, and thus provide effective feature representations for various downstream tasks (such as IDC encoding).

[0061] Optionally, the processed feature vector can be transformed through stacked transformation layers so that the output layer of the AI ​​surgery ICD coding model performs coding recognition to obtain a coding recognition result.

[0062] Exemplarily, the word vector of the sentence corresponding to the surgical record is represented by the feature output of the hidden layer of the named entity recognition model to obtain a combined feature vector. The result of the combined feature vector obtained by stacking the conversion layers can be input into the output layer of the AI ​​surgical ICD coding model for recognition to obtain a labeled surgical record surgery-related medical vocabulary sequence.

[0063] Optionally, the process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets preset training conditions also includes: The AI ​​surgery ICD coding model is verified using the surgery-related medical vocabulary sequence of the test set data, and the network structure and parameters of the AI ​​surgery ICD coding model are cyclically adjusted until the preset training conditions are met.

[0064] The test set is used to verify the model accuracy and loss, and to find the iteration rounds at which the AI ​​surgery ICD coding model begins to overfit. For example, after the total number of iterations of the network begins to overfit, the training accuracy and training loss are relatively stable, and the verification accuracy no longer increases, and the verification loss no longer decreases. At this time, it can be considered that the preset training conditions have been met. Removing the number of iterations after this can not only reduce the computer's computing load, but also avoid model overfitting.

[0065] Optionally, the process of using the surgery-related medical vocabulary sequence of the test set data to verify the AI ​​surgery ICD coding model and cyclically adjusting the network structure and parameters of the AI ​​surgery ICD coding model includes: The difference between the coding recognition result and the actual medical record coding is calculated by the cross entropy loss function, the gradient of the difference to the model parameters is calculated by the back propagation algorithm, and the weights and bias parameters of the convolutional layer and the pooling layer are updated using the SGD optimizer for the gradient.

[0066] Exemplarily, the training set is input into the AI ​​surgery ICD coding model, and then the recommended improved model performs forward propagation on the training set to obtain the ICD coding result. The difference between the coding result and the actual value is then calculated using the cross entropy loss function, and the gradient of the loss to the model parameters is calculated using the back propagation algorithm. The calculated gradient is then clipped, and then the SGD optimizer is used to update the model weights and bias terms based on the calculated gradient. After each round of training, the model is verified using the test set, the network structure or hyperparameters are adjusted, and the training is repeated until the pre-set number of training rounds or the conditions for stopping training are reached. The trained AI surgery ICD coding improved model collects surgical records in real time, and calculates and filters the encoded surgical records.

[0067] Third embodiment The present application also provides a readable storage medium storing a computer program, which implements the steps of the above method when executed by a processor.

[0068] The method for generating ICD codes for surgery provided in the present application collects coded and archived surgical records, splits them into training set data and test set data, and constructs an AI surgical ICD coding model; uses the training set data to train the AI ​​surgical ICD coding model, and uses the test set data to test the AI ​​surgical ICD coding model, so that the AI ​​surgical ICD coding model meets preset training conditions; based on the trained AI surgical ICD coding model, identifies surgical records of unarchived medical records to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records, thereby improving the speed and accuracy of ICD coding for surgery, significantly reducing the ICD coding management cost for surgery, and improving user experience.

[0069] It should be noted that in the present application, step codes such as S10, S20, etc. are used for the purpose of expressing the corresponding content more clearly and concisely, and do not constitute a substantial limitation on the sequence. When implementing the step, those skilled in the art may execute S20 first and then S10, etc., but these should all be within the scope of protection of the present application.

[0070] In the embodiments of the system and storage medium provided in the present application, all technical features of any of the above-mentioned method embodiments may be included, and the expanded and explained contents of the specification are basically the same as those of the above-mentioned method embodiments, and will not be repeated here.

[0071] The embodiment of the present application further provides a computer program product, which includes a computer program code. When the computer program code runs on a computer, the computer executes the methods in the above various possible implementation modes.

[0072] An embodiment of the present application also provides a chip, including a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that a device equipped with the chip executes the methods in various possible implementation modes as described above.

[0073] It is understood that the above scenarios are only examples and do not constitute a limitation on the application scenarios of the technical solutions provided in the embodiments of the present application. The technical solutions of the present application can also be applied to other scenarios. For example, it is known to those skilled in the art that with the evolution of the system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0074] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0075] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs.

[0076] The units in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0077] In the present application, the same or similar terminology concepts, technical solutions and / or application scenario descriptions are generally described in detail only the first time they appear. When they appear again later, they are generally not repeated for the sake of brevity. When understanding the technical solutions and other contents of the present application, for the same or similar terminology concepts, technical solutions and / or application scenario descriptions that are not described in detail later, reference can be made to the previous related detailed descriptions.

[0078] In the present application, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0079] The various technical features of the technical solution of the present application can be arbitrarily combined. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application.

[0080] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for generating ICD codes for surgery, characterized in that: include: Collect coded and archived surgical records, split them into training set data and test set data, and build an AI surgical ICD coding model; Using the training set data to train the AI ​​surgery ICD coding model, and using the test set data to test the AI ​​surgery ICD coding model, so that the AI ​​surgery ICD coding model meets the preset training conditions; Based on the trained AI surgical ICD coding model, the surgical records of unarchived medical records are identified to generate surgical ICD codes corresponding to the surgical records of the unarchived medical records.

2. The method for generating ICD codes for surgery according to claim 1, characterized in that: The process of collecting the coded and archived surgical records, splitting the training set data and the test set data, and building the AI ​​surgical ICD coding model also includes: Based on the coded and archived surgical records, data preprocessing is performed on the coded and archived surgical records of the medical records to obtain a sequence of surgical-related medical vocabulary to be marked; The surgery-related medical vocabulary sequence to be labeled is divided into training set data and test set data, and an AI surgery ICD coding model is constructed. The AI ​​surgery ICD coding model includes an input layer, a hidden layer and an output layer.

3. The method for generating ICD codes for surgery according to claim 2, characterized in that: The process of performing data preprocessing on the coded and archived surgical records based on the coded and archived surgical records to obtain a sequence of surgical related medical vocabulary to be marked includes: Performing data extraction, data cleaning, and data specification preprocessing on the coded and archived surgical records, and removing common descriptions, stop words, and common vocabulary during the data cleaning process; The pre-processed surgical records are visually analyzed to determine the optimal length of the medical vocabulary sequence related to the surgery.

4. The method for generating ICD codes for surgery according to claim 3, characterized in that: The process of visually analyzing the pre-processed surgical records includes: The coded and archived surgical records are subjected to word segmentation and text analysis to accurately identify the surgical records.

5. The method for generating ICD codes for surgery according to claim 4, characterized in that: The process of performing word segmentation and text analysis on the coded and archived surgical records includes: Performing word segmentation on the preprocessed surgical records, dividing continuous character strings in the surgical records into words or phrases; Medical part-of-speech tagging is performed on each word or phrase to identify surgery-related entities in the text; Based on the identified surgery-related entities, the dependencies between words in each sentence are analyzed and the semantic role of each word in the sentence is determined; According to the semantic roles and dependencies, entity relationships between each operation-related entity are extracted from the operation record.

6. The method for generating ICD codes for surgery according to claim 5, characterized in that: The process of dividing the surgery-related medical vocabulary sequence to be labeled into training set data and test set data, and constructing an AI surgery ICD coding model, wherein the AI ​​surgery ICD coding model includes an input layer, a hidden layer, and an output layer, further includes: Extracting surgical site feature values, surgical procedure feature values, approach feature values, filler feature values, and personnel feature values ​​based on the surgery-related medical vocabulary sequences in the training set data; Determine the nodes of the input layer and the hidden layer, wherein the input layer includes a first input node, a second input node, a third input node, a fourth input node and a fifth input node, the first input node is used to input the surgical site characteristic value, the second input node is used to input the procedure characteristic value, the third input node is used to input the approach characteristic value, the fourth input node is used to input the filler characteristic value, the fifth input node is used to input the personnel characteristic value, and the nodes of the hidden layer are composed of Gaussian kernel functions.

7. The method for generating ICD codes for surgery according to claim 6, characterized in that: The AI ​​surgery ICD coding model includes an embedding layer; the step of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets the preset training conditions includes: Converting the surgery-related medical vocabulary sequence of the training set into a numerical representation vector in the embedding layer to capture the semantic information of the surgery-related medical vocabulary sequence; Inputting the representation vector and the hidden state of the previous time step into the hidden layer, and updating the hidden state of the hidden layer by a preset nonlinear function; Based on the hidden state at each time step, the temporal information of the surgery-related medical vocabulary sequence is captured; A gating mechanism is introduced to control the information flow inside the hidden layer to effectively learn the long-term dependencies between medical words; Using the gradient clipping technology, when the memory gradient of the hidden layer exceeds the preset threshold, the gradient exceeding the preset time length is clipped in chronological order to ensure the stability of the training process.

8. The method for generating ICD codes for surgery according to claim 7, characterized in that: The process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets the preset training conditions also includes: The gradient of the parameters of the surgery-related medical vocabulary sequence of the training set is calculated using a back-propagation algorithm, and the parameters of the surgery-related medical vocabulary sequence of the training set are updated using a gradient descent algorithm to minimize the prediction error of the output feature representation, wherein the feature representation includes key features and the temporal relationship between the key features.

9. A method for generating ICD codes for surgery according to any one of claims 2 to 8, characterized in that: The process of training the AI ​​surgery ICD coding model using the training set data and testing the AI ​​surgery ICD coding model using the test set data so that the AI ​​surgery ICD coding model meets the preset training conditions also includes: The AI ​​surgery ICD coding model is verified using the surgery-related medical vocabulary sequence of the test set data, and the network structure and parameters of the AI ​​surgery ICD coding model are cyclically adjusted until the preset training conditions are met.

10. The method for generating ICD codes for surgery according to claim 9, characterized in that: The process of using the surgery-related medical vocabulary sequence of the test set data to verify the AI ​​surgery ICD coding model and cyclically adjusting the network structure and parameters of the AI ​​surgery ICD coding model includes: The difference between the coding recognition result and the actual medical record coding is calculated by the cross entropy loss function, the gradient of the difference to the model parameters is calculated by the back propagation algorithm, and the weights and bias parameters of the convolutional layer and the pooling layer are updated using the SGD optimizer for the gradient.

Images