Automatic generation method for consultation opinions of patients in perioperative period

Abstract

The invention provides an automatic generation method for consultation opinions of patients in a perioperative period. The automatic generation method comprises the following steps: extracting key patient indexes from an electronic medical record by utilizing a natural language processing technology; cleaning and standardizing the key patient indexes to ensure the consistency of data formats, and obtaining preprocessed index data; mapping the preprocessed index data to a parameter format of a standardized risk assessment scale to obtain structured patient data; calling a plurality of clinically verified risk assessment scales to assess the structured patient data, and generating a comprehensive risk assessment score according to an assessment result; classifying the comprehensive risk of the patient according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result; and converting the comprehensive risk assessment result into a consultation opinion text by using a generative large language model. By combining natural language processing, machine learning and the generative large language model, the efficiency and accuracy of perioperative patient risk assessment and consultation opinion generation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method for automatically generating consultation opinions for perioperative patients. Background Art

[0002] With the advancement of medical technology and the aging of the population, the number of surgeries has increased year by year, especially the proportion of elderly and high-risk patients has increased significantly. This has led to a significant increase in the number of perioperative risk assessments and consultations, and has put forward higher requirements for the quality of risk assessments and consultations. However, the existing technology has the following problems:

[0003] Increased surgical volume and increasing complexity of patients' conditions: The increase in surgical volume and the increasing complexity of patients' conditions have led to a rising demand for preoperative anesthesia consultations, while the relatively short supply of anesthesiologists has led to a strain on medical resources.

[0004] Anesthesiologists have a heavy workload: Anesthesiologists need to handle preoperative evaluations for a large number of patients, which limits effective consultation time and results in insufficient consultation quality.

[0005] Inconsistent knowledge of risk assessment scales: There are many types of risk assessment scales available, but doctors have different levels of knowledge of them, resulting in insufficient consistency and accuracy in the assessment results.

[0006] Heterogeneity of consultation opinions: The format and content of consultation opinions lack standardization, which affects the efficiency of handover and the accuracy of information transmission.

[0007] Therefore, there is an urgent need for a method that can automatically generate perioperative patient consultation opinions to improve the efficiency and accuracy of risk assessment, reduce the workload of doctors, and ensure the standardization and consistency of consultation opinions. Summary of the invention

[0008] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for automatically generating perioperative patient consultation opinions, aiming to improve the efficiency and accuracy of perioperative patient risk assessment and consultation opinion generation through intelligent technology.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] A method for automatically generating perioperative patient consultation opinions, comprising:

[0011] Extract key patient metrics from electronic medical records using natural language processing techniques;

[0012] Cleaning and standardizing the key patient indicators to ensure data format consistency and obtain preprocessed indicator data;

[0013] Mapping the pretreatment indicator data to a parameter format of a standardized risk assessment scale to obtain structured patient data;

[0014] Invoking a plurality of clinically validated risk assessment scales to evaluate the structured patient data, and generating a comprehensive risk assessment score based on the evaluation results;

[0015] Based on a machine learning algorithm, the comprehensive risk of the patient is risk stratified according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result;

[0016] The comprehensive risk assessment results are converted into standardized consultation opinion text using a generative large language model.

[0017] Preferably, the key patient indicators include: age, gender, height, weight, BMI, past medical history and auxiliary examination results.

[0018] Preferably, natural language processing techniques are used to extract key patient indicators from electronic medical records, including:

[0019] segmenting free text in said electronic medical record into words;

[0020] Marking the part of speech for each of the words;

[0021] Using a named entity recognition model in the medical field to identify medical entities in the free text and restore the words to their root forms;

[0022] Performing text classification on the root forms in the free text to obtain classification data;

[0023] The key patient indicators in the classified data are extracted using a trained named entity recognition (NER) model.

[0024] Preferably, text classification is performed on the root forms in the free text to obtain classification data, including:

[0025] Inputting the root form in the free text into the text context feature extraction layer to extract context feature information;

[0026] Inputting the root form in the free text into the global feature extraction layer to extract global feature information;

[0027] Fusing the contextual feature information and the global feature information to obtain a fused text feature;

[0028] Constructing a loss function using the fused text features;

[0029] Continuously optimizing the loss function to obtain a text classification model;

[0030] The text classification model is used to complete text classification to obtain the classification data.

[0031] Preferably, the root form in the free text is input into the text context feature extraction layer to extract context feature information, including:

[0032] Extracting initial feature information of root forms in the free text using a pre-trained language model;

[0033] Inputting the initial feature information into a forward gated recurrent unit and a reverse gated recurrent unit;

[0034] The outputs of the forward gated recurrent unit and the reverse gated recurrent unit are concatenated to obtain contextual feature information; the expression of the contextual feature information is: Among them, H t ={H1,H2,…H l} represents the initial feature information input at time t, represents the output of the positive gated recurrent unit at time t, Represents the output of the reverse gated recurrent unit at time t, GRU represents the gated recurrent unit, and G represents the contextual relationship feature information.

[0035] Preferably, the expression of the global feature information is:

[0036] c i =f(ω●H+b)

[0037]

[0038] In the formula, f is the activation function, ω is the convolution kernel, h is the convolution kernel size, b is the bias value, c is i is the feature vector extracted by the i-th convolutional layer, It represents the value of the features extracted by three different convolution kernels after the maximum pooling layer, and C represents the extracted global feature information.

[0039] Preferably, the risk assessment scale includes: ASA classification table, Child-Pugh scoring table, NYHA heart function classification table, RCRI modified heart risk assessment table, ARISCAT postoperative pulmonary complications risk scoring table, Essen stroke risk scoring table.

[0040] Preferably, multiple clinically validated risk assessment scales are used to assess the structured patient data, and a comprehensive risk assessment score is generated based on the assessment results, including:

[0041] The scoring algorithm of each scale is called according to the structured patient data to calculate the scoring result of each scale;

[0042] Based on a weighted average algorithm, the comprehensive risk assessment score is generated according to the scoring results.

[0043] Preferably, based on a machine learning algorithm, the comprehensive risk of the patient is classified according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result, including:

[0044] Obtain preset comprehensive risk assessment sub-datasets;

[0045] Build an initial neural network deep learning model;

[0046] Training the initial neural network deep learning model according to the comprehensive risk assessment sub-data set to obtain a trained classifier;

[0047] Connecting the trained LSTM neural network after the classifier to obtain a comprehensive risk classification model;

[0048] The comprehensive risk assessment score is input into the comprehensive risk classification model to obtain the comprehensive risk assessment result.

[0049] Preferably, the generative large language model is a GPT language model.

[0050] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0051] The present invention provides an automated generation method for perioperative patient consultation opinions, including: extracting key patient indicators from electronic medical records using natural language processing technology; cleaning and standardizing the key patient indicators to ensure data format consistency to obtain pre-processed indicator data; mapping the pre-processed indicator data to the parameter format of a standardized risk assessment scale to obtain structured patient data; calling a variety of clinically validated risk assessment scales to evaluate the structured patient data, and generating a comprehensive risk assessment score based on the evaluation results; based on a machine learning algorithm, stratifying the comprehensive risk of the patient according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result; and using a generative large language model to convert the comprehensive risk assessment result into a standardized consultation opinion text. The present invention significantly improves the efficiency, accuracy and standardization of perioperative patient risk assessment and consultation opinion generation by combining advanced technologies such as natural language processing, machine learning and generative large language models, reduces the burden on medical workers, optimizes the medical service process, and provides patients with higher quality personalized medical services. The implementation of this system will bring significant efficiency improvements and technological advances to the medical industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0053] Figure 1 A flow chart of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The purpose of the present invention is to provide a method for automatically generating perioperative patient consultation opinions, which significantly improves the efficiency, accuracy and standardization of perioperative patient risk assessment and consultation opinion generation by combining advanced intelligent technologies such as natural language processing, machine learning and generative large language models.

[0056] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Figure 1 A flow chart of a method provided by an embodiment of the present invention, such as Figure 1 As shown, the present invention provides a method for automatically generating perioperative patient consultation opinions, comprising:

[0058] Step 100: Extract key patient indicators from electronic medical records using natural language processing techniques;

[0059] Step 200: cleaning and standardizing key patient indicators to ensure data format consistency and obtain pre-processed indicator data;

[0060] Step 300: Mapping the pre-processing indicator data to the parameter format of a standardized risk assessment scale to obtain structured patient data;

[0061] Step 400: Calling multiple clinically validated risk assessment scales to evaluate structured patient data, and generating a comprehensive risk assessment score based on the evaluation results;

[0062] Step 500: Based on a machine learning algorithm, risk stratify the comprehensive risk of the patient according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result;

[0063] Step 600: Use a generative large language model to convert the comprehensive risk assessment results into a standardized consultation opinion text.

[0064] Preferably, the key patient indicators include: age, gender, height, weight, BMI, past medical history and auxiliary examination results.

[0065] Preferably, natural language processing techniques are used to extract key patient indicators from electronic medical records, including:

[0066] segmenting free text in said electronic medical record into words;

[0067] Marking the part of speech for each of the words;

[0068] Using a named entity recognition model in the medical field to identify medical entities in the free text and restore the words to their root forms;

[0069] Performing text classification on the root forms in the free text to obtain classification data;

[0070] The key patient indicators in the classified data are extracted using a trained named entity recognition (NER) model.

[0071] Specifically, step 100 of this embodiment includes:

[0072] Step 1: Word segmentation and text segmentation

[0073] First, segment the free text in the electronic medical record into words or phrases. For Chinese text, you can use a word segmentation tool (such as jieba or HanLP) to segment sentences into words; for English text, you can use spaCy or NLTK for word segmentation. The purpose of word segmentation is to break down continuous text into basic language units (such as words or phrases) for subsequent processing. It should be noted that terms in the medical field (such as "hypertension" and "diabetes") may be fixed phrases, so it is necessary to combine medical dictionaries (such as ICD-10 or SNOMED CT) to identify and segment professional terms to avoid incorrect segmentation.

[0074] Step 2: Part-of-speech tagging

[0075] On the basis of word segmentation, each word is labeled with the part of speech (such as noun, verb, adjective, etc.). Part of speech tagging can help identify important information in the text. For example, disease names are usually nouns, and test results may be numerical values ​​or adjectives. Tools such as spaCy and Stanza can be used for part of speech tagging, and combined with part of speech tagging models in the medical field (such as BioBERT or ClinicalBERT) to improve the accuracy of tagging. For example, in "the patient's blood pressure is 120 / 80mmHg", "blood pressure" is labeled as a noun and "120 / 80" is labeled as a numerical value. Through part of speech tagging, contextual information can be provided for subsequent medical entity recognition.

[0076] Step 3: Named Entity Recognition and Lemma Restoration

[0077] Use the named entity recognition (NER) model in the medical field to identify medical entities (such as diseases, drugs, test results, etc.) in free text. NER models can be implemented based on deep learning (such as BioBERT, Med7) or rule matching. The model will identify the boundaries and categories of medical entities based on contextual information. For example, "hypertension" is identified as a disease entity and "aspirin" is identified as a drug entity. At the same time, the words are restored to their root form (lemma restoration), for example, "checked" is restored to "check" to unify the expression form and reduce redundancy. Lemmatization can be implemented using spaCy or NLTK's lemma restoration tool.

[0078] Step 4: Text Classification

[0079] Text classification is performed on the root forms in free text in order to divide the text into different categories (such as medical history, test results, diagnostic information, etc.). Text classification can be implemented using machine learning models (such as support vector machines SVM, random forests) or deep learning models (such as BERT, TextCNN). When training the model, it is necessary to use annotated electronic medical record data as a training set. For example, "the patient has a history of hypertension" is classified as "past medical history", and "blood pressure is 120 / 80mmHg" is classified as "examination results". The purpose of classification is to categorize the information in the text and provide contextual support for the subsequent extraction of key indicators.

[0080] Step 5: Extract key patient indicators

[0081] Use the trained named entity recognition (NER) model to extract key patient indicators from the classified data. The NER model will extract the patient's age, gender, height, weight, BMI, past medical history, and auxiliary examination results based on the classified text content. For example, from "patient age 65 years old, weight 70kg, height 170cm", age = 65, weight = 70kg, height = 170cm, and BMI = 24.2 are extracted; from "past medical history includes hypertension and diabetes", the disease entities "hypertension" and "diabetes" are extracted. The extracted indicators will be stored as structured data (such as JSON or database tables) to provide standardized input for subsequent risk assessment and consultation opinion generation.

[0082] Preferably, text classification is performed on the root forms in the free text to obtain classification data, including:

[0083] Inputting the root form in the free text into the text context feature extraction layer to extract context feature information;

[0084] Inputting the root form in the free text into the global feature extraction layer to extract global feature information;

[0085] Fusing the contextual feature information and the global feature information to obtain a fused text feature;

[0086] Constructing a loss function using the fused text features;

[0087] Continuously optimizing the loss function to obtain a text classification model;

[0088] The text classification model is used to complete text classification to obtain the classification data.

[0089] In practical applications, this embodiment can use the XLNet model to complete the preliminary feature extraction step. The XLNet model is a pre-trained language model based on the combination of autoregression and autoencoding. Its core part is the Transformer-XL architecture, which solves the defect of the context input length constraint of the BERT model (Bidirectional Encoder Representations from Transformers), so that the pre-trained model can learn more distant context information.

[0090] Preferably, the root form in the free text is input into the text context feature extraction layer to extract context feature information, including:

[0091] Extracting initial feature information of root forms in the free text using a pre-trained language model;

[0092] Inputting the initial feature information into a forward gated recurrent unit and a reverse gated recurrent unit;

[0093] The outputs of the forward gated recurrent unit and the reverse gated recurrent unit are concatenated to obtain contextual feature information; the expression of the contextual feature information is: Among them, H t ={H1,H2,…H l} represents the initial feature information input at time t, represents the output of the positive gated recurrent unit at time t, Represents the output of the reverse gated recurrent unit at time t, GRU represents the gated recurrent unit, and G represents the contextual relationship feature information.

[0094] Specifically, this embodiment can more fully learn the context relationship of the text and obtain the context information by splicing the outputs of the forward gated recurrent unit and the reverse gated recurrent unit.

[0095] Preferably, the expression of the global feature information is:

[0096] c i =f(ω·H+b)

[0097]

[0098] In the formula, f is the activation function, ω is the convolution kernel, h is the convolution kernel size, b is the bias value, c is i is the feature vector extracted by the i-th convolutional layer, It represents the value of the features extracted by three different convolution kernels after the maximum pooling layer, and C represents the extracted global feature information.

[0099] Specifically, this embodiment fuses the contextual feature information and the global feature information to obtain a fused text feature, which specifically includes:

[0100] The contextual feature information and the global feature information are concatenated to obtain a first fused text feature; wherein the first fused text feature is:

[0101] The first fused text feature is subjected to maximum pooling processing to obtain the second fused text feature; the fused feature O is subjected to maximum pooling processing to extract the maximum feature in the first fused text feature O.

[0102] The first fused text feature is subjected to mean pooling processing to obtain a third fused text feature. The fused feature O is subjected to mean pooling processing to extract the mean feature in the fused feature O.

[0103] Furthermore, the loss function is constructed using the fused text features, including:

[0104] The first fused text feature is input into the fully connected layer and the Softmax layer to output the first probability value of the text category; wherein the expression of the first probability value output by the fully connected layer and the Softmax layer is:

[0105] Y1=tanh(W O ·O+b O )

[0106]

[0107] Among them, W O represents the weight matrix of the fully connected layer when the input is the first fused text feature, O represents the first fused text feature, b O Represents the bias vector of the fully connected layer when the input is the first fused text feature, When the input is the first fused text feature, the output layer weight matrix, θ represents the network parameters, P(y i |Y1,θ) means Y1 belongs to the yth i The probability value of each category.

[0108] The second fused text feature is input into the fully connected layer and the Softmax layer to output the second probability value of the text category; wherein the expression of the second probability value output by the fully connected layer and the Softmax layer is:

[0109]

[0110] in, Represents the weight matrix of the fully connected layer when the input is the second fused text feature, M p represents the second fused text feature, Represents the bias vector of the fully connected layer when the input is the second fused text feature, When the input is the second fusion text feature, the output layer weight matrix, θ represents the network parameters, P(y i |Y2,θ) means Y2 belongs to the yth i The probability value of each category;

[0111] The third fused text feature is input into the fully connected layer and the Softmax layer to output the third probability value of the text category; wherein the expression of the third probability value output by the fully connected layer and the Softmax layer is:

[0112]

[0113] in, Represents the weight matrix of the fully connected layer when the input is the second fused text feature, A p represents the third fused text feature, Represents the bias vector of the fully connected layer when the input is the third fused text feature, When the input is the third fusion text feature, the output layer weight matrix, θ represents the network parameters, P(y i |Y3,θ) means Y3 belongs to the yth i The probability value of each category;

[0114] Taking the maximum value among the first probability value, the second probability value and the third probability value as the prediction result;

[0115] The prediction results are used to construct a loss function; specifically, the expression of the loss function is:

[0116]

[0117] in, represents the true category of the text to be classified, m represents the total number of text categories, λ represents the regularization parameter, and y k Indicates the prediction result.

[0118] The present invention obtains fused text features by fusing the contextual feature information and global feature information of the text using a gated recurrent unit, a convolutional layer and a pooling layer, and constructs and optimizes a loss function based on this to complete text classification, thereby improving the classification performance and effect of the text.

[0119] Preferably, the risk assessment scale includes: ASA classification table, Child-Pugh scoring table, NYHA heart function classification table, RCRI modified heart risk assessment table, ARISCAT postoperative pulmonary complications risk scoring table, Essen stroke risk scoring table.

[0120] Specifically, step 300 of this embodiment includes:

[0121] First, the parameter requirements and scoring rules of each risk assessment scale need to be clarified. For example, the ASA classification requires information such as the patient's medical history and activity tolerance; the Child-Pugh score requires liver function-related indicators (such as bilirubin, albumin, prothrombin time, etc.); the RCRI modified cardiac risk assessment requires cardiovascular-related risk factors (such as previous myocardial infarction, diabetes, etc.); the ARISCAT score requires the patient's age, preoperative blood oxygen saturation, preoperative infection, type of surgery, etc. According to the parameter requirements of these scales, a standardized parameter mapping table is established to correspond the preprocessing indicator data to the scale parameters one by one. For example, the patient's "age" is directly mapped to the "age" parameter of the ARISCAT score, and the "hypertension" in the "past medical history" is mapped to the "risk factor" parameter of the RCRI score.

[0122] After the parameter format is clarified, the preprocessing indicator data is converted according to the mapping table. For example, "patient age 87 years old" extracted from the electronic medical record is directly mapped to the "age" field of the ARISCAT score; "past medical history: hypertension for 10 years" is mapped to the "risk factor" field of the RCRI score; "bilirubin 1.0 mg / dL" is mapped to the "bilirubin" field of the Child-Pugh score. For parameters that need to be calculated (such as BMI), they are calculated according to the formula (weight / height2) and then mapped to the parameter field of the relevant scale. In this way, the preprocessing indicator data is converted into a standardized format that meets the requirements of each risk assessment scale to ensure the integrity and consistency of the data.

[0123] After completing the parameter mapping, the parameter data of all scales are integrated into a structured patient data format (such as JSON or a database table). For example, the generated structured data may contain the following fields: {"age":87,"BMI":22,"ASA grade":2,"Child-Pugh score":"Grade A","RCRI risk factors":1,"ARISCAT score":47}. This structured data format facilitates the subsequent risk assessment algorithm call and processing while ensuring the standardization and scalability of the data. In this way, the system can quickly and accurately map the patient's pretreatment indicator data to the parameter format of each risk assessment scale, providing a reliable data basis for comprehensive risk assessment and consultation opinion generation.

[0124] Preferably, multiple clinically validated risk assessment scales are used to assess the structured patient data, and a comprehensive risk assessment score is generated based on the assessment results, including:

[0125] The scoring algorithm of each scale is called according to the structured patient data to calculate the scoring result of each scale;

[0126] Based on a weighted average algorithm, the comprehensive risk assessment score is generated according to the scoring results.

[0127] The first paragraph: Call the scoring algorithm of each scale to calculate the scoring results

[0128] Specifically, according to the structured patient data, the scoring algorithm of each risk assessment scale is called one by one to calculate the scoring result of each scale. The scoring algorithm of each scale is based on its specific parameter requirements and scoring rules. For example, the ASA grade directly outputs the grade result according to the patient's medical history and activity tolerance; the Child-Pugh score calculates the total score and determines the grade according to parameters such as bilirubin, albumin and prothrombin time; the RCRI score obtains the score by counting the number of cardiovascular-related risk factors (such as previous myocardial infarction, diabetes, etc.); the ARISCAT score calculates the total score based on the patient's age, preoperative blood oxygen saturation, preoperative infection and other parameters. By calling these algorithms, the system can generate independent scoring results for each scale, such as "ASA grade = 2", "Child-Pugh score = A", "RCRI score = 1 point", "ARISCAT score = 47 points".

[0129] After obtaining the scoring results of each scale, a weighted average algorithm is used to generate a comprehensive risk assessment score. First, a weight is assigned to each scale according to its importance, for example, the weight of the ASA grade is 0.3, the weight of the Child-Pugh score is 0.2, the weight of the RCRI score is 0.3, and the weight of the ARISCAT score is 0.2. Then, the scoring results of each scale are multiplied by the corresponding weights and accumulated to obtain a comprehensive risk assessment score. For example, if the ASA grade = 2 (corresponding to a score of 0.4), the Child-Pugh score = grade A (corresponding to a score of 0.2), the RCRI score = 1 (corresponding to a score of 0.3), and the ARISCAT score = 47 (corresponding to a score of 0.5), the comprehensive risk assessment score is (0.4×0.3)+(0.2×0.2)+(0.3×0.3)+(0.5×0.2)=0.35. The final comprehensive risk assessment score can be used for further classification (such as low risk, medium risk, high risk) or to generate consultation opinions.

[0130] Preferably, based on a machine learning algorithm, the comprehensive risk of the patient is classified according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result, including:

[0131] Obtain preset comprehensive risk assessment sub-datasets;

[0132] Build an initial neural network deep learning model;

[0133] Training the initial neural network deep learning model according to the comprehensive risk assessment sub-data set to obtain a trained classifier;

[0134] Connecting the trained LSTM neural network after the classifier to obtain a comprehensive risk classification model;

[0135] The comprehensive risk assessment score is input into the comprehensive risk classification model to obtain the comprehensive risk assessment result.

[0136] Specifically, this embodiment first collects and organizes the comprehensive risk assessment data set for training. The data set should contain the patient's comprehensive risk assessment score and its corresponding risk classification label (such as low risk, medium risk, high risk). The data source can be historical electronic medical record data, surgical records or clinical research data to ensure that the data covers different patient groups and risk levels. In order to improve the generalization ability of the model, the data set needs to be cleaned and standardized, such as removing outliers, filling missing values, and normalizing the comprehensive risk assessment score (such as mapping the score to a range of 0 to 1). In addition, the data set is divided into a training set, a validation set, and a test set (such as a ratio of 8:1:1) for subsequent model training and performance evaluation.

[0137] Based on the comprehensive risk assessment data set, a DNN model is constructed to classify comprehensive risks. The input layer of the model accepts the comprehensive risk assessment score, and the hidden layer consists of several fully connected layers, each of which contains a certain number of neurons (such as 64 or 128), and uses activation functions (such as ReLU) to introduce nonlinear characteristics. The output layer uses the Softmax activation function to convert the model output into a probability distribution of risk classification (such as the probability of low risk, medium risk, and high risk). The cross-entropy loss function is selected as the loss function of the model, and the Adam optimizer is selected as the optimizer. The model is iteratively trained through the training set, and the performance of the model (such as accuracy, F1 score) is monitored on the validation set, and the hyperparameters (such as learning rate, number of hidden layers) are adjusted to optimize the model effect.

[0138] After the initial neural network model is trained, it is used as a classifier and connected to a trained LSTM (Long Short-Term Memory) neural network to capture the time series characteristics or potential complex patterns of the comprehensive risk assessment score. The input of the LSTM network can be the patient's comprehensive risk assessment score sequence (such as the trend of score changes over multiple assessments), and the output is the probability distribution of risk classification. The hidden layer of the LSTM network can contain a certain number of units (such as 64 or 128), and the Tanh and Sigmoid activation functions are used to process time series data. Through joint training or staged training, the classifier and LSTM network are integrated into a comprehensive risk classification model to further enhance the model's ability to recognize complex risk patterns.

[0139] The patient's comprehensive risk assessment score is input into the trained comprehensive risk classification model. The model will perform forward propagation based on the input data and output the probability distribution of each risk category (such as low risk = 0.7, medium risk = 0.2, high risk = 0.1). The patient's comprehensive risk assessment result (such as low risk) is determined based on the maximum value of the output probability. In addition, the model can also provide confidence in the classification to help doctors better understand the reliability of the assessment results. The final comprehensive risk assessment results can be used to generate standardized consultation opinion texts to provide support for perioperative clinical decision-making.

[0140] Preferably, the generative large language model is a GPT language model.

[0141] Optionally, this embodiment first organizes the comprehensive risk assessment results (such as low risk, medium risk, high risk) and related patient information (such as age, gender, past medical history, auxiliary examination results, etc.) into structured input data as the input of the generative large language model. For example, the input data can be in JSON format, including fields such as {"age":65,"gender":"male","comprehensive risk assessment results":"medium risk","past medical history":["hypertension","diabetes"],"auxiliary examination results":{"blood pressure":"140 / 90mmHg","blood sugar":"8.5mmol / L"}}. At the same time, a standardized consultation opinion template is designed to clarify the structure and content of the generated text, such as basic patient information, comprehensive risk assessment results, analysis of major risk factors, and recommended preoperative preparations or interventions. This templated design can provide clear guidance for the generative model to ensure that the generated text meets clinical specifications.

[0142] In order to ensure that the generated consultation opinion text conforms to the professionalism and standardization of the medical field, the GPT language model can be fine-tuned. Fine-tuning requires the preparation of a high-quality medical text dataset, including a large number of consultation opinion instances, covering different risk assessment results and patient conditions. Through supervised learning, these medical texts are used to train the GPT model so that it can generate text that conforms to the style and logic of medical language. During the fine-tuning process, specific training goals can be set. For example, the generated text must contain key content such as the patient's comprehensive risk level, major risk factors, and specific preoperative recommendations. The fine-tuned model will be more suitable for medical scenarios and can generate more accurate and standardized consultation opinions.

[0143] After fine-tuning, the comprehensive risk assessment results and patient information are input into the GPT model to generate the consultation opinion text. For example, the input prompt can be "Generate a consultation opinion based on the following patient information: age 65, male, comprehensive risk assessment result is medium risk, past medical history includes hypertension and diabetes, auxiliary examination results show blood pressure 140 / 90mmHg, blood sugar 8.5mmol / L." The model will output a standardized consultation opinion, such as: "Patient, male, 65 years old, with a history of hypertension and diabetes. The comprehensive risk assessment result is medium risk, and the main risk factors include poor blood pressure control (140 / 90mmHg) and elevated blood sugar (8.5mmol / L). It is recommended to strengthen blood pressure and blood sugar management before surgery, and adjust antihypertensive drugs and insulin doses if necessary." After generation, the text can be post-processed through rule verification or manual review to ensure the accuracy and standardization of the content, and finally the consultation opinion text will be used for clinical decision support.

[0144] As an optional implementation, the consultation opinion of this embodiment is shown as follows:

[0145] Patient information: Zhang San 0005599888*** Ward 01 Bed

[0146] Chief complaint: abdominal pain for 3 days

[0147] Diagnosis: Hepatic space-occupying lesions

[0148] For handover: The patient is an elderly female, 150cm tall, 49.5kg in weight, BMI=22, ASA grade 2. Child-Pugh liver function grade 0 points A. NYHA heart function grade 2. Activity tolerance 4METs, 1 RCRI modified cardiac risk assessment risk factor, corresponding to a rate of cardiac death / non-fatal myocardial infarction / non-fatal cardiac arrest of approximately 1.0%. ARISCAT postoperative pulmonary complication risk prediction score of 47 points, corresponding to a high risk of postoperative pulmonary complications, Essen stroke risk score of 3 points, and a moderate risk of perioperative stroke, with a risk of 7-9%. The incidence of adverse cardiovascular events within 30 days after non-cardiac surgery by type of surgery is moderate, with an incidence of 1-5%.

[0149] The medical history has been reviewed and the patient has been visited.

[0150] The patient is an elderly female, 87 years old. She had hypertension for 10 years and took nifedipine regularly to control her blood pressure. She was scheduled to undergo laparoscopic liver resection. For more details, please refer to the medical record.

[0151] Physical examination: mouth opening > 3 horizontal fingers, normal head and neck mobility, Mahalanobis grade 1, no removable dentures or loose teeth.

[0152] Auxiliary tests and examinations: No obvious abnormalities were found in the electrocardiogram and chest X-ray, and the remaining tests and examinations have been reviewed.

[0153] Recommendation: 1. The patient is an elderly female, 150cm tall, 49.5kg in weight, BMI=22, ASA grade 2.

[0154] Risk assessment: NYHA heart function class 2. Activity tolerance 4METs, 1 RCRI modified cardiac risk assessment risk factor, corresponding to a rate of cardiac death / non-fatal myocardial infarction / non-fatal cardiac arrest of approximately 1.0%. ARISCAT postoperative pulmonary complication risk prediction score of 47 points, corresponding to a high risk of postoperative pulmonary complications, Essen stroke risk score of 3 points, a medium risk of perioperative stroke, a risk of 7-9%. Anesthesia risk is medium-high risk.

[0155] Communicate relevant risks with patients and their families and obtain informed consent: The risk of postoperative pulmonary complications such as pulmonary infection, atelectasis, respiratory failure, hypoxemia, etc. is extremely high. Transfer to the ICU for further treatment after surgery cannot be ruled out. The risk of perioperative cerebrovascular events, embolic bleeding, and infection increases accordingly, such as cerebral hemorrhage, cerebral ischemia, cerebral infarction, delirium, cognitive dysfunction, etc.

[0156] Pay attention to perioperative circulation stability, avoid drastic fluctuations in blood pressure and heart rate, perform advanced circulatory monitoring during surgery, and ensure perfusion of important organs. Strengthen respiratory management, protective ventilation strategies, avoid excessive ventilation and carbon dioxide accumulation. Pay attention to maintaining water and electrolyte balance and keeping warm. Provide adequate analgesia and antiemetic after surgery.

[0157] I will follow up with you in my department, thank you for the invitation!

[0158] Furthermore, the patient information table for extracting collected data in this embodiment is as follows:

[0159]

[0160]

[0161]

[0162] The beneficial effects of the present invention are as follows:

[0163] (1) The present invention utilizes natural language processing technology to automatically extract key patient indicators from electronic medical records and clean and standardize the data, thereby significantly reducing the time for manual data collection and organization and improving the efficiency of risk assessment. The automated risk assessment process enables anesthesiologists to obtain comprehensive risk assessment results for patients more quickly, thereby focusing on personalized assessment and decision-making.

[0164] (2) The present invention integrates a variety of clinically validated risk assessment scales (such as ASA classification, Child-Pugh score, RCRI score, etc.), and generates a comprehensive risk assessment score through the "voting + weighting" method to ensure the comprehensiveness and accuracy of the assessment results; the comprehensive risk assessment scores are classified based on machine learning algorithms, which further improves the accuracy and consistency of risk assessment and reduces the influence of human subjective factors.

[0165] (3) The present invention reduces the repetitive workload of anesthesiologists in preoperative consultations through automated and standardized processes, enabling them to more efficiently manage the growing number of patients; the automatically generated consultation opinion text reduces the time doctors spend writing reports and optimizes the workflow.

[0166] (4) The consultation opinion text generated by the present invention adopts a standardized and modular format, which is convenient for information sharing and handover among medical teams, improves communication efficiency, and reduces errors in information transmission; the present invention strictly abides by the laws and regulations on medical data security and privacy protection, and adopts multi-level data encryption technology and access control mechanism to ensure the security and privacy of patient data and protect the interests of patients and medical institutions.

[0167] (5) The present invention utilizes a generative large language model to convert the comprehensive risk assessment results into standardized consultation opinion text, ensuring the professionalism, consistency, and readability of the report content, and reducing the understanding bias caused by differences in personal expression among doctors.

[0168] (6) The present invention provides more accurate and personalized medical services for perioperative patients through comprehensive risk assessment and standardized consultation opinions, thereby improving the quality and safety of patient care. The automatically generated consultation report can only be officially released after being reviewed by an authorized anesthesiologist, ensuring the accuracy and professionalism of the report while reducing the repetitive work of doctors.

[0169] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0170] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for automatically generating perioperative patient consultation opinions, characterized in that: include: Extract key patient metrics from electronic medical records using natural language processing techniques; Cleaning and standardizing the key patient indicators to ensure data format consistency and obtain preprocessed indicator data; Mapping the pretreatment indicator data to a parameter format of a standardized risk assessment scale to obtain structured patient data; Invoking a plurality of clinically validated risk assessment scales to evaluate the structured patient data, and generating a comprehensive risk assessment score based on the evaluation results; Based on a machine learning algorithm, the comprehensive risk of the patient is risk stratified according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result; The comprehensive risk assessment results are converted into standardized consultation opinion text using a generative large language model.

2. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: The key patient indicators include: age, gender, height, weight, BMI, medical history and auxiliary examination results.

3. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: Leverage natural language processing to extract key patient metrics from electronic medical records, including: segmenting free text in said electronic medical record into words; Marking the part of speech for each of the words; Using a named entity recognition model in the medical field to identify medical entities in the free text and restore the words to their root forms; Performing text classification on the root forms in the free text to obtain classification data; The key patient indicators in the classified data are extracted using a trained named entity recognition model.

4. The method for automatically generating perioperative patient consultation opinions according to claim 3, characterized in that: Text classification is performed on the root forms in the free text to obtain classification data, including: Inputting the root form in the free text into the text context feature extraction layer to extract context feature information; Inputting the root form in the free text into the global feature extraction layer to extract global feature information; Fusing the contextual feature information and the global feature information to obtain a fused text feature; Constructing a loss function using the fused text features; Continuously optimizing the loss function to obtain a text classification model; The text classification model is used to complete text classification to obtain the classification data.

5. The method for automatically generating perioperative patient consultation opinions according to claim 4, characterized in that: The root form in the free text is input into the text context feature extraction layer to extract context feature information, including: Extracting initial feature information of root forms in the free text using a pre-trained language model; Inputting the initial feature information into a forward gated recurrent unit and a reverse gated recurrent unit; The outputs of the forward gated recurrent unit and the reverse gated recurrent unit are concatenated to obtain contextual feature information; the expression of the contextual feature information is: Among them, H t ={H1,H2,…H l } represents the initial feature information input at time t, represents the output of the positive gated recurrent unit at time t, Represents the output of the reverse gated recurrent unit at time t, GRU represents the gated recurrent unit, and G represents the contextual relationship feature information.

6. The method for automatically generating perioperative patient consultation opinions according to claim 5, characterized in that: The expression of the global feature information is: c i =f(ω·H+b) In the formula, f is the activation function, ω is the convolution kernel, h is the convolution kernel size, b is the bias value, c is i is the feature vector extracted by the i-th convolutional layer, It represents the value of the features extracted by three different convolution kernels after the maximum pooling layer, and C represents the extracted global feature information.

7. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: The risk assessment scales include: ASA classification table, Child-Pugh scoring table, NYHA heart function classification table, RCRI modified heart risk assessment table, ARISCAT postoperative pulmonary complications risk scoring table, and Essen stroke risk scoring table.

8. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: A variety of clinically validated risk assessment scales are used to evaluate the structured patient data, and a comprehensive risk assessment score is generated based on the evaluation results, including: The scoring algorithm of each scale is called according to the structured patient data to calculate the scoring result of each scale; Based on a weighted average algorithm, the comprehensive risk assessment score is generated according to the scoring results.

9. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: Based on the machine learning algorithm, the comprehensive risk of the patient is risk stratified according to the comprehensive risk assessment score to obtain a comprehensive risk assessment result, including: Obtain preset comprehensive risk assessment sub-datasets; Build an initial neural network deep learning model; Training the initial neural network deep learning model according to the comprehensive risk assessment sub-data set to obtain a trained classifier; Connecting the trained LSTM neural network after the classifier to obtain a comprehensive risk classification model; The comprehensive risk assessment score is input into the comprehensive risk classification model to obtain the comprehensive risk assessment result.

10. The method for automatically generating perioperative patient consultation opinions according to claim 1, characterized in that: The generative large language model is a GPT language model.

Images