A medical record intelligent agent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases by using medical record information

Abstract

A medical record intelligent agent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases by using medical record information, the present application relates to a medical record intelligent agent system and belongs to the field of medical information processing. The present application aims to solve the problems that the existing technology system does not sufficiently mine medical record information, does not make good use of various types of data, and the overall reasoning process is invisible, which makes the system poor in interpretability. The system comprises: a data processing module, a numerical type hepatobiliary and pancreatic disease classification module, a natural language type hepatobiliary and pancreatic disease classification module, an image examination data type hepatobiliary and pancreatic disease classification module, an integration module, and an evidence judgment module; the numerical type hepatobiliary and pancreatic disease classification module is used for obtaining numerical type hepatobiliary and pancreatic disease classification results; the natural language type hepatobiliary and pancreatic disease classification module is used for obtaining natural language type hepatobiliary and pancreatic disease classification results; and the image examination data type hepatobiliary and pancreatic disease classification module is used for obtaining image examination type hepatobiliary and pancreatic disease classification results.

Description

Technical Field

[0001] This invention relates to a medical record intelligent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information, and belongs to the field of medical information processing. Background Technology

[0002] With the rapid development of artificial intelligence technology, more and more medical fields are beginning to explore the application of technologies such as machine learning and deep learning in disease prediction. Currently, three methods—disease diagnosis based on symptom description, disease diagnosis based on tabular numerical data, and inference rules based on disease evolution—have become important research directions in the field of intelligent healthcare, each with unique advantages and application scenarios.

[0003] Disease diagnosis based on symptom description:

[0004] This method predicts the type of disease a patient may have by analyzing symptom information provided by the patient. Its core technology is natural language processing, which uses machine learning algorithms to analyze large amounts of medical text data, learn a model of the relationship between symptoms and diseases, and thus predict the patient's possible disease type. For example, by establishing a relatively complete symptom-disease knowledge base, and then using a Naive Bayes-based symptom matching model and deep learning techniques, significant progress has been made in symptom-based diagnosis. Applying genetic algorithms to decision trees and Bayesian classification methods further mines the rich semantics of medical record information, resulting in a significant improvement in the accuracy of disease diagnosis. The advantage of this method is that it can quickly make a preliminary diagnosis through simple question-and-answer methods and can reduce the interference of subjective factors from doctors in the diagnostic process. However, this method requires a large amount of medical text data for training; insufficient data can easily lead to low prediction accuracy.

[0005] Disease diagnosis based on tabular numerical data:

[0006] This method utilizes machine learning algorithms to analyze and model tabular data in medical records to predict the types of diseases a patient may have. Given that tabular data contains a large amount of numerical information, which differs from natural language information such as symptom descriptions, a highly scalable, unified format processing tool is needed. Wei et al. used the CDSS architecture to study a general case analysis method that supports rapid analysis of cases of different types of diseases for case storage (Wei Hongpeng. Research and Implementation of Openness in Clinical Diagnosis and Treatment Decision Support System [D]. Xi'an University of Electronic Science and Technology, 2017). By annotating the columns of the tabular data, the relationships between columns are fully explored, improving the prediction accuracy in this field. The advantage of this method is that it can fully utilize the structured information in medical record data to make more comprehensive and accurate predictions of the patient's condition, while avoiding interference from the doctor's subjective factors. The disadvantage of this method is that it requires the establishment of a complex data model and a large amount of medical record data for training; insufficient data can easily lead to low prediction accuracy.

[0007] Inference rules based on disease evolution:

[0008] This method is based on in-depth research into the evolution of diseases in the medical field, constructing inference rules to better predict disease development trends and treatment options. Cosnes et al. (Cosnes, Jacques, et al. "Long-term evolution of disease behavior of Crohn's disease." Inflammatory bowel diseases 8.4 (2002): 244-250.) retrospectively studied data from 2,002 Crohn's disease patients to identify the occurrence and predictive factors of stenosis and / or penetrating complications. Finally, these inference methods were automated using finite state automata and other methods. The core of this method is expert knowledge and a comprehensive understanding of the disease development process to establish a sound disease model and inference rules. The advantage of this method is that it can better understand the nature and evolution of diseases, thereby improving the treatment effect and survival rate of patients. However, the disadvantage of this method is that it requires a large amount of clinical data for training and extensive medical expert knowledge to guide the construction of rules.

[0009] While each method has its own advantages and disadvantages, combining them can maximize their strengths and minimize the impact of their weaknesses on disease prediction. For example, in disease prediction, a pre-trained disease diagnosis model based on symptom description can be used for rapid initial diagnosis, followed by further analysis and prediction using a pre-trained model based on tabular data and inference rules based on disease evolution. In this scenario, a preliminary diagnosis can be made quickly using symptom information provided by the patient, and then further diagnosis and inference can be made based on more information such as medical records and clinical experience, thus improving prediction accuracy.

[0010] Current research methods suffer from three main shortcomings. First, they often utilize only one or two types of data from medical records. While many methods can extract sufficient information from specific data types, such as chief complaint information, they still discard structured data like examination indicators, limiting their application scenarios. Second, medical records contain a large amount of patient personal information, which is highly sensitive to privacy. However, many current methods do not explicitly specify how to anonymize this data. Third, while most current methods employ deep learning technology, the drawback is that the entire reasoning process is invisible. In sensitive fields like medicine, doctors and patients often cannot trust unfounded reasoning, which also contradicts evidence-based medicine, making the practical application and promotion of various intelligent diagnostic and treatment technologies extremely difficult. Summary of the Invention

[0011] The purpose of this invention is to address the problems of insufficient mining of medical record information, failure to make good use of various types of data, and poor interpretability of the system due to the invisibility of the overall reasoning process in existing technology systems. Therefore, this invention proposes a medical record intelligent agent system that uses medical record information for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases.

[0012] A medical record intelligent system for intelligent assisted diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information includes:

[0013] The module includes: data processing module, numerical type hepatobiliary and pancreatic disease classification module, natural language type hepatobiliary and pancreatic disease classification module, imaging examination data type hepatobiliary and pancreatic disease classification module, integration module, and evidence judgment module.

[0014] The data processing module is used to collect hepatobiliary and pancreatic medical record data, unify the format of the collected hepatobiliary and pancreatic medical record data, perform privacy desensitization on the unified format hepatobiliary and pancreatic related medical record data, and classify the privacy desensitized hepatobiliary and pancreatic medical record data.

[0015] The numerical type hepatobiliary and pancreatic disease classification module is used to process the numerical type data in the medical records after the data processing module has processed it, and obtain the numerical type hepatobiliary and pancreatic disease classification results.

[0016] The natural language type hepatobiliary and pancreatic disease classification module is used to process the natural language type data in the medical records after the data processing module has processed it, and obtain the natural language type hepatobiliary and pancreatic disease classification results.

[0017] The image examination data type classification module for hepatobiliary and pancreatic diseases is used to process the basic image examination type data in the medical records after the data processing module has processed it, and to obtain the image examination type classification results for hepatobiliary and pancreatic diseases.

[0018] The integration module is used to integrate numerical type hepatobiliary and pancreatic disease classification results, natural language type hepatobiliary and pancreatic disease classification results, and imaging examination type hepatobiliary and pancreatic disease classification results to obtain integrated classification results.

[0019] The evidence judgment module is used to determine whether the integrated classification results are consistent with the doctor's diagnosis. If they are consistent, it outputs supporting evidence; if they are inconsistent, it outputs opposing evidence.

[0020] The beneficial effects of this invention are as follows:

[0021] The method proposed in this invention not only solves the problems of automated and intelligent diagnosis of common hepatobiliary and pancreatic diseases, but also focuses on exploring ways to enhance the interpretability and reliability of the entire intelligent diagnostic process. This will make the entire intelligent diagnosis process transparent and the entire system decision-making process credible, truly helping doctors make the final decision.

[0022] The Medical Record Intelligent Agent aims to develop an artificial intelligence-based medical assistance tool to help doctors and medical teams make disease diagnosis and treatment decisions. This invention analyzes patient medical record data, including symptoms, medical history, and physical signs, and utilizes advanced machine learning algorithms and big data analytics to predict the patient's possible disease type and provide doctors with relevant diagnostic suggestions and treatment plans. The field of Medical Record Intelligent Agents is in a phase of rapid development, with current research focusing on improving the accuracy and understanding of natural language processing technology in medical texts to extract key information more precisely. Machine learning algorithms play a crucial role in assisting doctors in diagnosis, developing personalized treatment plans, and predicting patient conditions.

[0023] It can truly assist doctors and medical teams in making disease diagnosis and treatment decisions. By analyzing patients' medical records, including symptoms, medical history, signs, and other information, and using advanced machine learning algorithms and big data analysis technology, it can predict the possible types of diseases patients may have and provide doctors with relevant diagnostic suggestions, as well as corresponding diagnostic reasons and evidence.

[0024] The medical record intelligent agent method of this invention has high accuracy and precision, can quickly process large amounts of medical record data, and provide high-quality diagnostic and treatment suggestions. Furthermore, the medical record intelligent agent method of this invention can continuously update and optimize its algorithm model at a low cost, expanding to meet new functionalities or business needs. Attached Figure Description

[0025] Figure 1 This is a system flowchart of the present invention;

[0026] Figure 2 This is a schematic diagram of the input and output of this method;

[0027] Figure 3 This is a diagram illustrating a specific application example of this method. Detailed Implementation

[0028] Specific Implementation Method 1: This implementation method provides a medical record intelligent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information, comprising:

[0029] The module includes: data processing module, numerical type hepatobiliary and pancreatic disease classification module, natural language type hepatobiliary and pancreatic disease classification module, imaging examination data type hepatobiliary and pancreatic disease classification module, integration module, and evidence judgment module.

[0030] The data processing module is used to collect liver, gallbladder and pancreas related medical record data, unify the format of the collected liver, gallbladder and pancreas related medical record data, perform privacy desensitization on the unified format liver, gallbladder and pancreas related medical record data, and classify the privacy desensitized liver, gallbladder and pancreas related medical record data.

[0031] The numerical type hepatobiliary and pancreatic disease classification module is used to process the numerical type data in the medical records after the data processing module has processed it, and obtain the numerical type hepatobiliary and pancreatic disease classification results.

[0032] The natural language type hepatobiliary and pancreatic disease classification module is used to process the natural language type data in the medical records after the data processing module has processed it, and obtain the natural language type hepatobiliary and pancreatic disease classification results.

[0033] The image examination data type classification module for hepatobiliary and pancreatic diseases is used to process the basic image examination type data in the medical records after the data processing module has processed it, and to obtain the image examination type classification results for hepatobiliary and pancreatic diseases.

[0034] The integration module is used to integrate numerical type hepatobiliary and pancreatic disease classification results, natural language type hepatobiliary and pancreatic disease classification results, and imaging examination type hepatobiliary and pancreatic disease classification results to obtain integrated classification results.

[0035] The evidence judgment module is used to determine whether the integrated classification results are consistent with the doctor's diagnosis. If they are consistent, it outputs supporting evidence; if they are inconsistent, it outputs opposing evidence.

[0036] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that: the data processing module is used to collect liver, gallbladder and pancreas related medical record data, unify the format of the collected liver, gallbladder and pancreas related medical record data, perform privacy desensitization on the format-unified liver, gallbladder and pancreas related medical record data, and classify the privacy-desensitized liver, gallbladder and pancreas related medical record data.

[0037] The specific process is as follows:

[0038] 1) Collect relevant medical records for the liver, gallbladder, and pancreas; the specific process is as follows:

[0039] The hepatobiliary and pancreatic related medical record dataset contains K types of diseases, and the proportion of medical record data for each disease in the dataset is no less than [a certain percentage].

[0040] The hepatobiliary and pancreatic related medical record dataset contains numerical data (examination indicators), natural language data (chief complaint information), and imaging examination data.

[0041] This method utilizes the burgeoning deep learning and machine learning technologies. A key aspect of using these technologies lies in data collection and processing. Based on practical experimental experience, each disease should have at least 100 corresponding medical records, and these records must be directly related to the patient's diagnosis of the corresponding hepatobiliary and pancreatic disease, rather than other historical information. Furthermore, this method combines the patient's chief complaint, examination indicators, and imaging results from the medical records. Therefore, the medical record data must include all three. Additionally, because the model training results are strongly influenced by data distribution, two points need to be emphasized.

[0042] 2) Standardize the format of the collected hepatobiliary and pancreatic related medical records; the specific process is as follows:

[0043] The collected liver, gallbladder, and pancreas-related medical records were standardized and aligned to JSON format.

[0044] This is to prevent the final trained model from having significantly biased prediction results. Secondly, the distribution of natural language data in medical records varies slightly across different hospitals or regions, and the units or evaluation standards for laboratory indicators also differ. Therefore, in practical use, if the application is specific to a particular hospital, it is best to use that hospital's actual clinical data. If the application is to multiple hospitals, it is essential to combine patient data from multiple different hospitals in the dataset to enhance the system's generalization ability. The evaluation or measurement standards for examination indicators must also be standardized.

[0045] 3) Perform privacy anonymization on the standardized formatted hepatobiliary and pancreatic related medical records; the specific process is as follows:

[0046] Natural language processing or text matching technology is used to scan and identify liver, gallbladder and pancreas related medical records after they have been standardized, and to find text containing personal identification information, which is then removed or replaced.

[0047] In the medical field, protecting patient privacy and personal information is of paramount importance (Kayaalp, Mehmet. "Patient privacy in the era of big data." Balkan Medical Journal 35.1 (2018): 8-17.). Therefore, when using medical record data for medical research or developing intelligent medical aids, this invention must take a series of measures to protect patient privacy and personal information. The following is the process by which this invention processes privacy information in medical record data:

[0048] First, this invention must identify and remove personally identifiable information contained in medical records, such as name, date of birth, phone number, address, etc.

[0049] Examples include gender, race, religious belief, and marital status. Similarly, this invention uses similar technology to scan and identify medical record data, and removes or replaces text containing sensitive information.

[0050] For example, in this method, the data usage is as follows: 1280 medical records are used, and data information in the medical records that is irrelevant to this method is anonymized and protected for privacy. Names, regions, and Chinese resident identity information are all hidden as meaningless text.

[0051] 4) Classify the liver, gallbladder, and pancreas-related medical records after privacy anonymization; the specific process is as follows:

[0052] The types of liver, gallbladder, and pancreas diseases are set as follows: gallbladder polyps, liver cysts, cholecystitis, gallstones, liver cancer, pancreatic cancer, bile duct stones, bile duct cysts, pancreatitis, and bile duct cancer.

[0053] Classify the liver, gallbladder, and pancreas-related medical records after privacy anonymization.

[0054] Based on common types of hepatobiliary and pancreatic diseases, they are categorized into 10 types: gallbladder polyps, liver cysts, cholecystitis, gallstones, liver cancer, pancreatic cancer, bile duct stones, bile duct cysts, pancreatitis, and bile duct cancer. Case data for each type of disease are categorized and organized according to the chief complaint, numerical data, and imaging examination information in the medical record.

[0055] The other steps and parameters are the same as in Specific Implementation Method 1.

[0056] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that: the numerical type hepatobiliary and pancreatic disease classification module is used to process the numerical type data in the medical records after the data processing module has processed, and obtain the numerical type hepatobiliary and pancreatic disease classification results; the specific process is as follows:

[0057] 1) Process the numerical data (lab reports) in the medical records after the data processing module has finished processing; the specific process is as follows:

[0058] If the data in the medical record is of numeric type, it will not be processed;

[0059] If the data in the medical record is non-numerical, then convert the non-numerical data into numerical information, for example, convert positive / negative to 1 / -1;

[0060] Classifying and organizing numerical data is a crucial task in medical record analysis, typically involving various examination indicators. In special examinations, the model's output is not a continuous numerical value, but rather presented as categories such as negative / positive. Therefore, to handle these results in the model, rules need to be established under prior conditions to convert them into numerical information, such as 1 / -1.

[0061] 2) The numerical classification model for hepatobiliary and pancreatic diseases is the XGBoost model;

[0062] Gradient Boosting Decision Trees (GBDT) is an effective ensemble learning technique. In ensemble learning, multiple weak learners (usually decision trees) are combined to build a powerful model. The core idea of ​​gradient boosting is to iteratively train a series of weak learners, each trained on the residuals of the previous learner. The final prediction is a weighted sum of all the weak learners. For the selection of base learners, this method uses the widely adopted XGBoost (eXtreme Gradient Boosting) model as a reference and defaults to using decision trees as base learners. A decision tree is a recursive binary tree structure where each node represents a decision condition for a feature, and each leaf node represents a predicted value. XGBoost uses a CART (Classification and Regression Trees) type decision tree.

[0063] 3) Train the numerical classification model for hepatobiliary and pancreatic diseases to obtain a well-trained numerical classification model for hepatobiliary and pancreatic diseases; the specific process is as follows:

[0064] The input to the XGBoost model is the numerical data in the processed medical records, and the output of the XGBoost model is the numerical classification results of hepatobiliary and pancreatic diseases.

[0065] The objective function is L t The loss function is the cross-entropy loss function.

[0066] Obtain a trained numerical classification model for hepatobiliary and pancreatic diseases;

[0067] 4) Obtain the relevant medical record data of the liver, gallbladder and pancreas to be tested, process it through the data processing module, process the numerical data (lab reports) in the medical record after the data processing module, and obtain the numerical data to be tested.

[0068] The numerical classification model for hepatobiliary and pancreatic diseases is used to classify the numerical data to be tested, and the numerical classification results for hepatobiliary and pancreatic diseases are obtained.

[0069] Other steps and parameters are the same as in specific implementation method one or two.

[0070] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the objective function L... t for:

[0071]

[0072] In the formula, L t Let y represent the objective function of the t-th decision tree, and n represent the number of data points in a batch of update data; i Indicates sample x i The tag, This indicates that the first t-1 decision trees for sample x i The prediction result is generally obtained by adding the prediction results of the first t-1 decision trees, f t (x i ) represents the current t-th decision tree pair for sample x. i The prediction results; Ω(f t ) represents the complexity of the t-th decision tree;

[0073] loss represents the loss for The cross-entropy loss function can be specifically written as:

[0074]

[0075] In the formula p(x i ) represents the true sample distribution, q(x) i ) represents the sample distribution predicted by the model, and n represents the number of data points in a batch of updated data.

[0076] The complexity Ω(f) of the t-th decision tree t The acquisition process represented by ) is as follows:

[0077] Ω(f t The value represents the complexity of the entire decision tree, and is generally a linear combination of L1 and L2 regularization terms. The L1 regularization term is calculated as the sum of the absolute values ​​of the features. The L2 regularization term is calculated as the sum of the squares of the features.

[0078] For the t-th decision tree f t L1 regularization term Ω1(f t The calculation is as follows:

[0079]

[0080] Where Q is the total number of leaf nodes in the t-th decision tree; w j γ is the weight of the j-th leaf node; γ is a hyperparameter that controls the strength of regularization.

[0081] For the t-th decision tree f t L2 regularization term Ω2(f t The calculation is as follows:

[0082]

[0083] Where Q is the total number of leaf nodes in the t-th decision tree; w j λ is the weight of the j-th leaf node; λ is a hyperparameter that controls the strength of regularization.

[0084] The complexity of the t-th decision tree is Ω(f t ) is a linear combination of L1 and L2 regularization terms:

[0085]

[0086] In practical applications, the strength of regularization can be controlled by adjusting the hyperparameters γ and λ. Larger γ and λ will result in stronger regularization, thereby limiting the growth of the tree and reducing the risk of overfitting.

[0087] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0088] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One to Four in that the information gain entropy H(k) of different features k (output features of different layers in the model) obtained in the numerical type hepatobiliary and pancreatic disease classification model is calculated using the following formula:

[0089]

[0090] In the formula, Z represents the number of all decision trees;

[0091] C(t) represents the total number of non-leaf nodes in the t-th decision tree;

[0092] B(t) represents the total number of nodes in the t-th decision tree (including the total number of non-leaf nodes and leaf nodes); O represents the O-th node in the t-th decision tree;

[0093] m represents whether the current node belongs to feature k, with 1 indicating that it belongs and 0 indicating that it does not.

[0094] Loss(t,i,L) and Loss(t,i,R) represent the losses on the left and right nodes of the i-th non-leaf node in the t-th decision tree, respectively.

[0095] D(t,i,L) and D(t,i,R) are the second derivatives of Loss(t,i,L) and Loss(t,i,R), respectively.

[0096] b and γ are the hyperparameters of the regularization term;

[0097] The splitting of the node to which different features k belong in the numerical hepatobiliary and pancreatic disease classification model is I(k) in the entire decision tree; the calculation formula is shown in the following formula:

[0098]

[0099] In the formula, B(t) represents the total number of nodes in the t-th decision tree (including the total number of non-leaf nodes and the total number of leaf nodes in the t-th decision tree);

[0100] Based on the information gain entropy H(k) and the splitting situation I(k), the importance (k) of feature k for the numerical classification model of hepatobiliary and pancreatic diseases is obtained; the calculation formula is shown in the following formula:

[0101] Importance(k)=αH(k)+(1-α)I(k)Importance(k)

[0102] In the formula, α represents the adjustment parameter;

[0103] H(k) is useful for finding the top few features most important for a decision tree, but it often yields low computational results for enumerated variables, leading to them being ignored. However, in medical diagnosis, patients' test results often show enumerated data. For example, a certain indicator may be positive or weakly positive.

[0104] Therefore, this invention introduces the I(k) index, which is more favorable for enumerated features, and the influence of the two is balanced by adjusting the parameter α.

[0105] Of course, different diseases will result in different feature rankings during training. Depending on the patient's input, some features may not be input, while others may be amplified. Therefore, although the model is static, the important features found in the end will also differ due to different inputs.

[0106] Decision tree reasoning is an invisible black box process. To make the reasoning process of the medical record agent transparent and improve the interpretability of the conclusions, this invention proposes a novel method for calculating the contribution gain of decision tree nodes. By comprehensively considering the changes in the loss function and node splitting during the training phase, the most important feature nodes for model decision-making are calculated. The features extracted from the patient data for these nodes are then used as the basis for the model's decision-making.

[0107] The other steps and parameters are the same as those in specific implementation methods one through four.

[0108] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that: the natural language type hepatobiliary and pancreatic disease classification module is used to process the natural language type data in the medical records after the data processing module has processed it, and obtain the natural language type hepatobiliary and pancreatic disease classification results; the specific process is as follows:

[0109] 1) The natural language-based classification model for hepatobiliary and pancreatic diseases is the Bert-chinese model;

[0110] 2) Train the natural language-based classification model for hepatobiliary and pancreatic diseases to obtain a well-trained natural language-based classification model for hepatobiliary and pancreatic diseases; the specific process is as follows:

[0111] The input to the natural language type hepatobiliary and pancreatic disease classification model is the natural language type data in the medical records after processing by the data processing module (the natural language type data (chief complaint information) in the medical records is processed based on the BERT model to obtain the processed natural language type data); the output of the natural language type hepatobiliary and pancreatic disease classification model is the hepatobiliary and pancreatic disease type.

[0112] The loss function is the cross-entropy loss function;

[0113] Obtain a well-trained natural language classification model for hepatobiliary and pancreatic diseases;

[0114] 3) Obtain the relevant medical record data of the liver, gallbladder and pancreas to be tested, process it through the data processing module, process the natural language type data (chief complaint information) in the medical record after the data processing module, and obtain the natural language type data to be tested.

[0115] Based on the trained natural language type classification model for hepatobiliary and pancreatic diseases, the test natural language type data is classified to obtain the natural language type classification results for hepatobiliary and pancreatic diseases.

[0116] The medical record information data is processed into training data and validation set data for classification tasks. In this method, the bidirectional language model based on a pre-trained Chinese language model is the most well-known open-source model in the industry, BERT-chinese. BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained model based on the Transformer architecture, proposed by Google in 2018. One of BERT's main innovations is the use of bidirectional contextual information to pre-train the model, which has led to significant performance improvements on multiple natural language processing tasks, including text classification. During the pre-training phase, BERT learns language representations through unsupervised learning on large-scale text corpora. The model structure is as follows: Figure 2 As shown. Bert-chinese is based on Bert, but with the Chinese vocabulary replaced and the training of a full-text Chinese mask added.

[0117] BERT uses a Transformer architecture containing multiple self-attention layers. These layers allow the model to consider all positions in the input sequence simultaneously, not just individual words. This helps the model capture global contextual information. For text classification tasks, the input is one or more text sequences. Each text sequence is marked with a special "[CLS]" tag to indicate that it is a classification task. Each word in the sentence is embedded as a vector. During the fine-tuning phase, BERT's output layer is typically modified to suit the specific classification task. The output layer usually consists of a fully connected layer and a softmax activation function to predict the class of the text sequence. The loss function is generally the cross-entropy loss function.

[0118]

[0119] In the formula p(x i ) represents the true sample distribution, q(x) i ) represents the sample distribution predicted by the model, and n represents the number of data points in a batch of updated data.

[0120] The hyperparameters that need to be set in advance during model training are shown in Table 1 below. These parameters are typically related to neural network training in machine learning, especially when using deep learning models for classification tasks. The following is an explanation of these parameters:

[0121] ①hidden_dropout_prob

[0122] Significance: To control the dropout probability of hidden layers in a neural network.

[0123] Explanation: Dropout is a regularization technique used to prevent overfitting in neural networks. `hidden_dropout_prob` specifies the probability of randomly dropping hidden layer neurons in each training step. During training, these neurons will be randomly ignored with the specified probability, which helps improve the model's generalization ability.

[0124] ②num_labels

[0125] Significance: Defines the number of categories in a classification task.

[0126] Explanation: In classification problems, `num_labels` represents the number of distinct classes that the model needs to distinguish. In this example, the model is designed to handle a classification problem involving two classes.

[0127] ③learning_rate

[0128] Meaning: Learning rate, used to control the magnitude of model weight updates in each training iteration.

[0129] Explanation: The learning rate is an important hyperparameter that determines the step size at which the model updates the weights in each training iteration. A smaller learning rate may lead to slower convergence but greater stability; a larger learning rate may lead to faster convergence but may be unstable. `learning_rate = 2e-5` means the learning rate is 2 times 10 to the power of -5.

[0130] ④weight_decay

[0131] Significance: Weights for L2 regularization.

[0132] Explanation: Weight Decay is a method to prevent overfitting by adding a regularization term to the loss function. The weight_decay parameter controls the weight of the regularization term, penalizing the model's weights and encouraging the model to use smaller weight values.

[0133] ⑤ epochs = 5:

[0134] Meaning: The number of training iterations.

[0135] Explanation: epochs specify the number of times the entire training dataset is used by the model for learning. In each epoch, the model performs one forward and one backward propagation through the entire training dataset. Increasing the number of epochs helps the model learn patterns in the data better, but may increase training time.

[0136] ⑥ batch_size

[0137] Meaning: The number of samples used in each training iteration.

[0138] Explanation: The training dataset is divided into batches, each containing a certain number of samples. `batch_size` specifies the number of samples in each batch. Larger batches can improve training speed but may require more memory.

[0139] Table 1. List of hyperparameters for the Bert model

[0140]

[0141]

[0142] The other steps and parameters are the same as those in specific implementation methods one through five.

[0143] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One through Six in that the formula for calculating the dot product attention in the Bert-chinese model is:

[0144]

[0145] Where Dot(x) represents dot product attention;

[0146] Q represents the output of the input text X after passing through a linear layer of shape [Embed, Embed]; Embed represents the embedding length.

[0147] K represents the output of the input text X after passing through a linear layer of shape [Embed,Embed].

[0148] Q and K represent the outputs of input X after passing through two different linear layers of shape [Embed,Embed], respectively;

[0149] Q×K TThe result is a matrix of [Sequence_length, Sequence_length]. This matrix can be viewed as a weight matrix of the influence of each token in the input on the output. The i-th row represents the influence factor of the i-th token in the sentence on the result; Sequence_length is the length of the input text X; d represents the embedding dimension; the superscript T indicates transpose.

[0150] Set a threshold T′, and extract the text of all tokens greater than the threshold T′ from the input, as...

[0151] The basis for Bert-chinese model decision-making (the decision result of Bert-chinese model is determined by the extracted sentences);

[0152] token represents the word token.

[0153] Clearly, the selection of the threshold T is crucial for extracting diagnostic criteria.

[0154] During model inference, the input is transformed into a Tensor vector of shape [Batch_size, Sequence_length, Embed]. Batch_size represents the number of samples in a batch, Sequence_length is the length of the input text, and Embed is an embedding length defined during model training.

[0155] The reasoning process in the BERT model is also an invisible black box process. To make the reasoning process of the medical record agent transparent and improve the interpretability of the conclusions, this invention proposes a novel interpretability method based on dot product attention and quadratic verification analysis.

[0156] Therefore, this invention utilizes a two-stage verification process. The extracted sentences are used as new input, allowing the model to re-evaluate. If the result matches the previous input, then these sentences can be considered sufficient evidence for the model's diagnosis. If the model output differs from the previous one, it indicates missing information. This invention appropriately increases the threshold T and iteratively obtains the final result.

[0157] The other steps and parameters are the same as those in specific implementation methods one through six.

[0158] Specific Implementation Method Eight: This implementation method differs from Specific Implementation Methods One through Seven in that: the imaging examination data type hepatobiliary and pancreatic disease classification module is used to process the imaging examination type data in the medical records after the data processing module has processed it, and obtain the imaging examination type hepatobiliary and pancreatic disease classification results; the specific process is as follows:

[0159] The imaging examination data in the medical records are processed into regular expressions to obtain the classification results of hepatobiliary and pancreatic diseases in the form of regular expressions.

[0160] Extracting specific disease diagnoses from medical imaging results typically involves extracting and analyzing textual information. Regular expression matching and finite state automata (FSA) are possible methods, but they need to be customized based on the specific text structure and patterns. Analyzing the text structure of medical imaging results helps understand the general format and patterns of different disease diagnoses. This may involve regular expression matching patterns, such as keywords, phrases, or formats specific to a diagnosis. Regular expressions are used to match key information in the disease diagnosis results. This may include information such as disease name, severity, and location. Regular expressions can be used to capture patterns in this information. If the structure of the medical imaging results is complex, finite state automata can be used to more flexibly capture specific text patterns. A finite state automata is a formal tool that can describe various patterns in strings. Below is a demonstration of regular expression design.

[0161] 1) Extract disease name:

[0162] Regular expression: (liver cancer|liver cyst|gallbladder polyp|cholecystitis|tumor)

[0163] 2) Locating the lesion location:

[0164] Regular expression: (left-hand|right-hand|both-handed)

[0165] 3) Determination of lesion extent:

[0166] Regular expression: (mild|moderate|severe)

[0167] 4) Extract keywords from image descriptions:

[0168] Regular expression: (image display|discovery|observation|visibility)

[0169] 5) Extracting digital information (such as size, density, etc.):

[0170] Regular expression: (\d+(\.\d+)?)\s*(cm|mm|cm) 2 |mm 2 )

[0171] The `(\d+(\.\d+)?)` part is used to match the number part, which can be an integer or a decimal. `\d+`: matches one or more digits, i.e., the integer part. For example, 123 or 45 can be matched. `(\.\d+)?`: matches the optional decimal part. Specifically, `\.`: matches the decimal point. `\d+`: matches one or more digits, i.e., the decimal part. `()?`: indicates that the decimal point and decimal part are optional; that is, the decimal part can be present or absent. If there is no decimal point, the integer part can also be matched alone. Therefore, this subexpression `(\d+(\.\d+)?)` can match the following: integers: such as 123; numbers with decimals: such as 45.67; decimals alone: ​​such as 0.5; The `\s*` part is used to match whitespace characters between digits and units. `\s`: matches any whitespace character, including spaces, tabs, newlines, etc. `*`: represents zero or more whitespace characters. So it can match even if there are no spaces between the digits and units; cm|mm|cm 2 |mm 2 The '|' symbol indicates 'or'.

[0172] The other steps and parameters are the same as those in specific implementation methods one through seven.

[0173] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that the integration module is used to integrate numerical type hepatobiliary and pancreatic disease classification results, natural language type hepatobiliary and pancreatic disease classification results, and imaging examination type hepatobiliary and pancreatic disease classification results to obtain integrated classification results;

[0174] The specific process is as follows:

[0175] 1) The integrated module classification model is a multilayer perceptron (MLP);

[0176] The working principle of a multilayer perceptron is as follows:

[0177] MLP(X) = WX + B

[0178] Where X is the input of the multilayer perceptron, and W and B are the parameters to be trained.

[0179] 2) Train the integrated module classification model to obtain a trained integrated module classification model; the specific process is as follows:

[0180] The input to the integrated module classification model is the predicted label and the predicted probability. The predicted label is the classification result of hepatobiliary and pancreatic diseases in numerical type, natural language type, and imaging examination type. The predicted probability refers to the probability that the current sample belongs to a certain type of hepatobiliary and pancreatic disease.

[0181] The output of the integrated module classification model is the integrated type of hepatobiliary and pancreatic diseases.

[0182] The loss function is the cross-entropy loss function;

[0183] Obtain a trained integrated module classification model;

[0184] 3) Based on the trained integrated module classification model, the obtained numerical type hepatobiliary and pancreatic disease classification results, natural language type hepatobiliary and pancreatic disease classification results, and imaging examination type hepatobiliary and pancreatic disease classification results are processed to obtain integrated classification results.

[0185] The integration of results from the above three parts needs to be carried out according to professional medical guidelines. Table 2 below lists the commonly used diagnostic methods for hepatobiliary and pancreatic diseases used in this experiment. Specifically, for example, gallbladder polyps can generally be diagnosed using only imaging results. Therefore, the diagnosis of this disease should focus on the results of step four, with the results of step three serving as supplementary information, while the results of step two should not be given too much importance.

[0186] As noted in Table 2, some diseases require multi-dimensional diagnostic criteria. For example, the numerical data (real-world laboratory test indicators) processed in step two and the imaging results analysis in step four may both be important for diagnosing a particular disease. If the conclusions of these two steps are consistent, the overall result can be obtained directly. If the conclusions are inconsistent, then it is necessary to filter and integrate the two conclusions.

[0187] To address this issue, this invention trains a perceptron MLP. When the conclusions from the above steps conflict, the perceptron is used to filter and fuse the conclusions. The working principle of the perceptron is as follows.

[0188] MLP(X) = WX + B

[0189] To some extent, this is equivalent to learning more granular distinctions between different dimensions of diagnostic results for specific diseases in the training samples.

[0190] The input X consists of two parts: the predicted label and the predicted probability. The predicted label represents the disease prediction results obtained in steps two, three, and four. The predicted probability refers to the probability that the current sample belongs to this disease. The specific method for obtaining this probability is as follows:

[0191] In step two, the XGBoost model calculates the probability of each class label using the SoftMax formula in multi-class classification tasks and outputs the label with the highest probability as the result. This method does not require explicit secondary calculation; for example, in Python libraries, it can be directly obtained using the `XGBoost.predict_proba()` method.

[0192] Step three of the BERT-Chinese model also outputs probabilities according to the SoftMax formula. First, obtain the vector corresponding to the position of the [CLS] marker in the last layer of the model from the BERT input, and then use the formula to calculate it: torch.softmax(logits, dim = -1). The highest probability among these is taken as the prediction probability for this step.

[0193] Table 2 Predictive probabilities for different hepatobiliary and pancreatic diseases

[0194]

[0195] Step four does not involve deep learning models. In this method, based on medical clinical diagnostic criteria (as shown in Table 3) and practical experience, the predicted probabilities of different diseases are given manually as shown in Table 2 above.

[0196] To some extent, this is equivalent to learning more granular distinctions between different dimensions of diagnostic results for specific diseases in the training samples.

[0197] Table 3 Diagnostic criteria for common hepatobiliary and pancreatic diseases

[0198]

[0199]

[0200] The other steps and parameters are the same as those in specific implementation methods one through eight.

[0201] Specific Implementation Method Ten: This implementation method differs from Specific Implementation Methods One through Nine in that: the evidence judgment module is used to determine whether the integrated classification results are consistent with the doctor's diagnostic opinion. If they are consistent, supporting evidence is output; if they are inconsistent, opposing evidence is output. The specific process is as follows:

[0202] If the integrated classification result obtained by the integration module is consistent with the doctor's, then the integrated classification result obtained by the integration module is supporting evidence;

[0203] If the integrated classification result obtained by the integration module is inconsistent with the doctor's, then the integrated classification result obtained by the integration module is considered opposing evidence.

[0204] If the final diagnosis of the medical record agent aligns with the doctor's, then the medical record information upon which the agent's decision is based becomes supporting evidence. This evidence consists of three parts: First, key examination indicators obtained in step two, which are often crucial evidence for disease diagnosis in medicine. This data can be obtained from the nodes in the decision tree of step two. Each node in the decision tree represents a corresponding examination result. The more critical the node, the higher the information gain and the smaller its depth in the decision tree, indicating that the examination indicator is more important. A series of important examination indicators can be output as inference evidence. Second, relevant descriptions in the chief complaint obtained in step three, such as disease symptoms and patient physical information. Third, relevant information obtained from imaging examination reports in step four. Different diseases may rely on different imaging examinations, such as enhanced CT, MRI, or color Doppler ultrasound. The results of imaging examinations are often important criteria for distinguishing similar diseases in medicine. Relevant report information located in the imaging results can be used as the basis for the medical record agent's diagnosis. An example of the input and output of this method is shown below. Figure 2 As shown.

[0205] Furthermore, the medical record AI should also provide evidence of missing data based on the diagnostic guidelines for different diseases outlined in Step 5. For example, if a patient is diagnosed with pancreatitis based solely on a CT scan, the medical record AI will, according to the guidelines in Step 5, provide the doctor with information on missing serum amylase and lipase test results as diagnostic evidence.

[0206] The other steps and parameters are the same as those in specific implementation methods one through nine.

[0207] The evaluation metrics used in this invention are precision, recall, and F1 score. Precision, recall, and F1 score are commonly used metrics for evaluating model performance in classification problems. Precision refers to the proportion of samples that are actually positive out of all samples predicted as positive by the model. Precision measures the accuracy of the model in predicting positive classes, i.e., how many of the model's predicted positive examples are true positive examples. Recall refers to the proportion of samples that are correctly predicted as positive out of all actual positive class samples. Recall measures the model's coverage of positive class samples, i.e., how many positive examples are successfully predicted. The F1 score is the harmonic mean of precision and recall, comprehensively considering the model's accuracy and coverage. The F1 score ranges from 0 to 1, with values ​​closer to 1 indicating better overall model performance. The F1 score is suitable for situations where precision and recall need to be balanced in different problems.

[0208] Commonly used machine learning methods for disease diagnosis include Naive Bayes and classifier methods based on the RoBERTa (Robustly optimized BERT approach) model.

[0209] The Naive Bayes method is a probabilistic statistical classification algorithm based on Bayes' theorem, commonly used in applications such as text classification. Its basic idea is to classify features by calculating the probability of input features given a class. Based on the training data, the prior probability (c) of each class and the conditional probability P(x|c) of each feature in each class are calculated. For a given input feature vector x, the posterior probability of each class is calculated using Bayes' theorem.

[0210]

[0211] P(x) is a normalization factor that sums over all classes such that the sum of the posterior probabilities is 1. The class with the highest posterior probability is selected as the prediction result.

[0212] RoBERTa (Robustly Optimized BERT Approach) is a Natural Language Processing (NLP) model proposed by Facebook AI Research in 2019. It is an improvement and optimization of BERT (Bidirectional Encoder Representations from Transformers). RoBERTa's design goal is to improve the performance of pre-trained models and address some problems inherent in BERT. RoBERTa's improvements have enabled it to achieve excellent performance on multiple NLP tasks, making it one of the most advanced pre-trained language models at the time. Because RoBERTa is based on...

[0213] It is an improvement on BERT, still using the Transformer architecture and generating context-sensitive word vectors, which makes it perform well in various classification tasks.

[0214] Table 4 compares the experimental results with those of common disease prediction and classification models.

[0215]

[0216] As shown in Table 3 above, the method of this invention outperforms some commonly used machine learning and deep learning methods in all classification metrics, reflecting the accuracy and reliability of the method in intelligent disease-assisted diagnosis. Furthermore, Table 4 below shows the classification results and metrics of the method of this invention for 10 common diseases in the hepatobiliary and pancreatic fields.

[0217] Table 5. Classification results and indicators of 10 common hepatobiliary and pancreatic diseases using this method.

[0218]

[0219] Besides its superior performance in disease prediction and classification, a key characteristic that distinguishes this invention from other intelligent assisted diagnosis methods is its ability to provide a chain of evidence for the entire disease inference process. Existing disease prediction models based on machine learning or deep learning are essentially black-box reasoning processes, lacking good interpretability. This invention cannot explore the internal reasoning process and basis for diagnosing diseases within the model. When discrepancies arise between model predictions and human expert diagnoses, other methods often fail to provide explanations or evidence, making it difficult for patients and doctors to trust the model's predictions. This also leads to significant difficulties in the practical application of other methods, hindering their ability to truly assist in the entire medical consultation process.

[0220] The method of this invention not only fully utilizes multi-source information such as chief complaints, examination indicators, and imaging results in medical records to improve the accuracy of disease prediction, but also provides sufficient evidence and justification for diagnosis. In this way, by examining and scrutinizing the diagnostic evidence output by the model, doctors can choose whether to trust the model's predictions. Furthermore, it provides doctors with new ideas and discoveries during the diagnostic process, truly enabling artificial intelligence to serve medical scenarios. (Demonstration follows.) Figure 3 As shown.

[0221] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A medical record intelligent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information, characterized in that: The system includes: The module includes: data processing module, numerical type hepatobiliary and pancreatic disease classification module, natural language type hepatobiliary and pancreatic disease classification module, imaging examination data type hepatobiliary and pancreatic disease classification module, integration module, and evidence judgment module. The data processing module is used to collect hepatobiliary and pancreatic medical record data, unify the format of the collected hepatobiliary and pancreatic medical record data, perform privacy desensitization on the unified format hepatobiliary and pancreatic related medical record data, and classify the privacy desensitized hepatobiliary and pancreatic medical record data. The numerical type hepatobiliary and pancreatic disease classification module is used to process the numerical type data in the medical records after the data processing module has processed it, and obtain the numerical type hepatobiliary and pancreatic disease classification results; the specific process is as follows: 1) Process the numerical data in the medical records after the data processing module has processed them; the specific process is as follows: If the data in the medical record is of numeric type, it will not be processed; If the data in the medical record is non-numeric, then convert the non-numeric data into numeric information; 2) The numerical classification model for hepatobiliary and pancreatic diseases is the XGBoost model; 3) Train the numerical classification model for hepatobiliary and pancreatic diseases to obtain a well-trained numerical classification model for hepatobiliary and pancreatic diseases; the specific process is as follows: The input to the XGBoost model is the numerical data in the processed medical records, and the output of the XGBoost model is the numerical classification results of hepatobiliary and pancreatic diseases. The objective function is L t The loss function is the cross-entropy loss function. Obtain a trained numerical classification model for hepatobiliary and pancreatic diseases; 4) Obtain the liver, gallbladder and pancreas medical record data to be tested, process it through the data processing module, and process the numerical data in the medical record after the data processing module to obtain the numerical data to be tested. Based on the trained numerical classification model for hepatobiliary and pancreatic diseases, the numerical data to be tested is classified to obtain the numerical classification results for hepatobiliary and pancreatic diseases. The natural language-based hepatobiliary and pancreatic disease classification module is used to process the natural language-based data in the medical records after the data processing module has processed it, and to obtain the natural language-based hepatobiliary and pancreatic disease classification results; the specific process is as follows: 1) The natural language-based classification model for hepatobiliary and pancreatic diseases is the Bert-chinese model; 2) Train the natural language-based classification model for hepatobiliary and pancreatic diseases to obtain a well-trained natural language-based classification model for hepatobiliary and pancreatic diseases; the specific process is as follows: The input to the natural language type hepatobiliary and pancreatic disease classification model is the natural language type data in the medical records after the data processing module, and the output of the natural language type hepatobiliary and pancreatic disease classification model is the hepatobiliary and pancreatic disease type. The loss function is the cross-entropy loss function; Obtain a well-trained natural language classification model for hepatobiliary and pancreatic diseases; 3) Obtain the liver, gallbladder and pancreas medical record data to be tested, process it through the data processing module, and process the natural language data in the medical record after the data processing module to obtain the natural language data to be tested. Based on the trained natural language type hepatobiliary and pancreatic disease classification model, the test natural language type data is classified to obtain the natural language type hepatobiliary and pancreatic disease classification results. The formula for calculating dot product attention in the Bert-chinese model is as follows: Where Dot(x) represents dot product attention; Q represents the output of the input text X after passing through a linear layer of shape [Embed, Embed]; Embed represents the embedding length. K represents the output of the input text X after passing through a linear layer of shape [Embed,Embed]. Q×K T The result is a matrix of [Sequence_length, Sequence_length], where Sequence_length is the length of the input text X; d represents the embedding dimension; and the superscript T indicates transpose. Set a threshold T′, and extract the text of all tokens whose dot product attention is greater than the threshold T′ from the input, which will be used as the basis for the Bert-chinese model's decision. token represents the word token; The imaging examination data type classification module for hepatobiliary and pancreatic diseases is used to process the basic imaging examination type data in the medical records after the data processing module has processed it, and to obtain the imaging examination type classification results for hepatobiliary and pancreatic diseases; the specific process is as follows: The imaging examination data in the medical records are processed into regular expressions to obtain the classification results of hepatobiliary and pancreatic diseases in the form of regular expressions. The predictive probability of liver cancer is 0.92; The predictive probability of pancreatic cancer is 0.76; The predicted probability of bile duct stones is 0.96; The predicted probability of gallbladder cancer is 0.

64. The predicted probability of gallbladder polyps is 0.96; The predicted probability of bile duct cysts is 0.

82. The predicted probability of pancreatitis is 0.

45. The predicted probability of liver cysts is 0.56; The predicted probability of cholecystitis is 0.63; The predictive probability of cholangiocarcinoma is 0.79; The integration module is used to integrate numerical, natural language, and imaging examination-based classification results of hepatobiliary and pancreatic diseases to obtain an integrated classification result; the specific process is as follows: 1) The integrated module classification model is a multilayer perceptron (MLP); The working principle of a multilayer perceptron is as follows: MLP(X) = WX + B Where X is the input of the multilayer perceptron, and W and B are the parameters to be trained; 2) Train the integrated module classification model to obtain a trained integrated module classification model; the specific process is as follows: The input to the integrated module classification model is the predicted label and the predicted probability. The predicted label is the classification result of hepatobiliary and pancreatic diseases in numerical type, natural language type, and imaging examination type. The predicted probability refers to the probability that the current sample belongs to a certain type of hepatobiliary and pancreatic disease. The output of the integrated module classification model is the integrated type of hepatobiliary and pancreatic diseases. The loss function is the cross-entropy loss function; Obtain a trained integrated module classification model; 3) Based on the trained integrated module classification model, the obtained numerical type hepatobiliary and pancreatic disease classification results, natural language type hepatobiliary and pancreatic disease classification results, and imaging examination type hepatobiliary and pancreatic disease classification results are processed to obtain integrated classification results; The evidence judgment module is used to determine whether the integrated classification results are consistent with the doctor's diagnosis. If they are consistent, it outputs supporting evidence; if they are inconsistent, it outputs opposing evidence.

2. The intelligent medical record system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information as described in claim 1, characterized in that: The data processing module is used to collect hepatobiliary and pancreatic medical record data, unify the format of the collected hepatobiliary and pancreatic medical record data, perform privacy desensitization on the format-unified hepatobiliary and pancreatic medical record data, and classify the privacy-desensitized hepatobiliary and pancreatic medical record data. The specific process is as follows: 1) Collect medical records of liver, gallbladder, and pancreas patients; the specific process is as follows: The hepatobiliary and pancreatic medical record dataset contains K types of diseases, and the proportion of medical record data corresponding to each disease in the dataset is no less than [a certain percentage]. The hepatobiliary and pancreatic medical record dataset contains numerical data, natural language data, and imaging examination data. 2) Standardize the format of the collected hepatobiliary and pancreatic medical records; the specific process is as follows: The collected hepatobiliary and pancreatic medical records were standardized and aligned to JSON format. 3) Perform privacy anonymization on the standardized liver, gallbladder, and pancreas medical record data; 4) Classify the liver, gallbladder, and pancreas medical records after privacy anonymization; the specific process is as follows: The types of liver, gallbladder, and pancreas diseases are set as follows: gallbladder polyps, liver cysts, cholecystitis, gallstones, liver cancer, pancreatic cancer, bile duct stones, bile duct cysts, pancreatitis, and bile duct cancer. Classify the liver, gallbladder and pancreas medical records after privacy anonymization.

3. The intelligent medical record system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information as described in claim 2, characterized in that: The objective function L t for: In the formula, L t Let y represent the objective function of the t-th decision tree, and n represent the number of data points in a batch of update data; i Indicates sample x i The tag, This indicates that the first t-1 decision trees for sample x i The prediction result, f t (x i ) represents the current t-th decision tree pair for sample x. i The prediction results; Ω(f t ) represents the complexity of the t-th decision tree; loss represents the loss for The cross-entropy loss function; The complexity Ω(f) of the t-th decision tree t The process of obtaining ) is as follows: For the t-th decision tree f t L1 regularization term Ω1(f t The calculation is as follows: Where Q is the total number of leaf nodes in the t-th decision tree; w j γ is the weight of the j-th leaf node; γ is a hyperparameter that controls the strength of regularization. For the t-th decision tree f t L2 regularization term Ω2(f t The calculation is as follows: Where Q is the total number of leaf nodes in the t-th decision tree; w j λ is the weight of the j-th leaf node; λ is a hyperparameter that controls the strength of regularization. The complexity of the t-th decision tree is Ω(f t ) is a linear combination of L1 and L2 regularization terms:

4. The intelligent medical record system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information as described in claim 3, characterized in that: The formula for calculating the information gain entropy H(k) of different features k obtained in the numerical hepatobiliary and pancreatic disease classification model is as follows: In the formula, Z represents the number of all decision trees; C(t) represents the total number of non-leaf nodes in the t-th decision tree; B(t) represents the total number of nodes in the t-th decision tree; O represents the 0th node in the t-th decision tree; m represents whether the current node belongs to feature k, with 1 indicating that it belongs and 0 indicating that it does not. Loss(t,i,L) and Loss(t,i,R) represent the losses on the left and right nodes of the i-th non-leaf node in the t-th decision tree, respectively. D(t,i,L) and D(t,i,R) are the second derivatives of Loss(t,i,L) and Loss(t,i,R), respectively. b and γ are the hyperparameters of the regularization term; The splitting of the node to which different features k belong in the numerical hepatobiliary and pancreatic disease classification model is I(k) in the entire decision tree; the calculation formula is shown in the following formula: In the formula, B(t) represents the total number of nodes in the t-th decision tree; Based on the information gain entropy H(k) and the splitting situation I(k), the importance (k) of feature k for the numerical classification model of hepatobiliary and pancreatic diseases is obtained; the calculation formula is shown in the following formula: Importance(k)=αH(k)+(1-α)I(k)Importance(k) In the formula, α represents the adjustment parameter.

5. A medical record intelligent agent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information according to claim 4, characterized in that: The imaging examination data type classification module for hepatobiliary and pancreatic diseases is used to process the imaging examination type data in the medical records after the data processing module has processed it, and to obtain the imaging examination type classification results for hepatobiliary and pancreatic diseases; the specific process is as follows: The imaging examination data in the medical records are processed into regular expressions to obtain the classification results of hepatobiliary and pancreatic diseases in the form of regular expressions. 1) Extract disease name: Regular expression: (liver cancer|liver cyst|gallbladder polyp|cholecystitis|tumor) 2) Locating the lesion location: Regular expression: (left-hand|right-hand|both-handed) 3) Determination of lesion extent: Regular expression: (mild|moderate|severe) 4) Extract keywords from image descriptions: Regular expression: (image display|discovery|observation|visibility) 5) Extracting digital information (such as size, density, etc.): Regular expression: (\d+(\.\d+)?)\s*(cm|mm|cm) 2 |mm 2 ).

6. A medical record intelligent agent system for intelligent auxiliary diagnosis and treatment of hepatobiliary and pancreatic diseases using medical record information, as described in claim 5, is characterized in that: The evidence judgment module is used to determine whether the integrated classification results are consistent with the doctor's diagnosis. If they are consistent, it outputs supporting evidence; if they are inconsistent, it outputs opposing evidence. The specific process is as follows: If the integrated classification result obtained by the integration module is consistent with the doctor's, then the integrated classification result obtained by the integration module is supporting evidence; If the integrated classification result obtained by the integration module is inconsistent with the doctor's, then the integrated classification result obtained by the integration module is considered opposing evidence.

Images