Predictive data processing method based on electronic medical record data and medical knowledge data

Abstract

The embodiment of the present application relates to a kind of prediction data processing methods based on electronic medical record data and medical knowledge data, comprising obtaining the first electronic medical record data of first patient, and based on the preset medical knowledge base model, the first electronic medical record data is preprocessed, and first feature vector is obtained;First feature vector is input into preset disease prediction model, and the predicted disease probability of first patient is output, and feature vector data and the feature vector importance corresponding to feature vector are output.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a predictive data processing method based on electronic medical record data and medical knowledge data. Background Technology

[0002] With the continuous development of medical technology, the medical field has accumulated a wealth of information in electronic medical records. The in-depth analysis and mining of these records to build intelligent clinical decision support systems is becoming increasingly widespread in the medical field. In particular, the use of diagnostic prediction models in outpatient electronic medical record data processing has become an important tool in doctors' daily work.

[0003] In existing technologies, auxiliary diagnostic models can only be used for classification tasks of a single disease dataset. Different diagnostic prediction models need to be deployed for different diseases. However, in actual clinical applications, doctors face a large and complex patient population with a wide range of diseases requiring diagnosis. There is a lack of a predictive model that can simultaneously perform data analysis and diagnostic prediction for multiple diseases. Furthermore, machine learning or deep learning models are usually accompanied by complex computational structures, especially deep learning models, which require more computational resources and have higher prediction costs.

[0004] Therefore, constructing a clinical auxiliary data processing method that can simultaneously predict multiple disease data and respond quickly is an essential data processing method for disease data prediction based on outpatient electronic medical record data. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a data processing method based on clinical electronic medical record data and medical knowledge data, which can simultaneously predict multiple disease data and respond quickly.

[0006] To achieve the above objectives, the present invention provides a predictive data processing method based on electronic medical record data and medical knowledge data, comprising: The first electronic medical record data of the first patient is obtained, and the first electronic medical record data is preprocessed based on a preset medical knowledge base model to obtain the first feature vector; The first feature vector is input into a preset disease prediction model, and the predicted disease probability of the first patient is output.

[0007] Preferably, the preprocessing of the first electronic medical record data based on a preset medical knowledge base model to obtain the first feature vector specifically includes: The first electronic medical record data is processed using a first preset natural language processing method to obtain first structured data; The first structured data is standardized based on the preset medical knowledge base model to obtain the first feature vector.

[0008] Preferably, the construction of the preset medical knowledge base model includes: Acquire the aforementioned medical knowledge data; The medical knowledge data is structured and extracted to obtain a first number of first medical knowledge data; wherein, the first medical knowledge data includes at least one of disease clinical manifestation data, laboratory test indicator data, imaging feature data, and pathological result data; Based on the first medical knowledge data, a symptom-disease relationship mapping and an examination result-diagnosis conclusion mapping are constructed and stored as a structured medical knowledge base model.

[0009] More preferably, the method further includes generating diagnostic rule data based on the structured medical knowledge base model.

[0010] More preferably, the construction of the preset disease prediction model includes: Obtain the electronic medical record data; The electronic medical record data is preprocessed based on a pre-defined medical knowledge base model to obtain a set of training word vectors for medical records. The preset machine learning method is trained using the medical record training word vector set to construct the preset disease prediction model.

[0011] More preferably, the electronic medical record data includes a second number of second electronic medical record data, and the preprocessing of the electronic medical record data based on a preset medical knowledge base model to obtain a medical record training word vector set specifically includes: The second preset natural language processing method is used to perform structured processing on the second number of the second electronic medical record data, and then standardized processing is performed based on the preset medical knowledge base model to obtain the second number of second structured data; wherein, each second structured data includes time information; For each of the second structured data, sorting and splicing are performed according to time information to obtain electronic medical record serialized data; The electronic medical record serialization data is converted into medical record training word vectors using a preset word vector generation method; The medical record training word vector set is formed based on the second number of medical record training word vectors.

[0012] More preferably, the step of using the medical record training word vector set to train the preset machine learning method and construct the preset disease prediction model specifically includes: The medical record training word vector set is divided into a training set and a test set according to a preset ratio; The training set and the test set are used to train and test a third number of preset deep learning models respectively, to obtain a third number of trained deep learning models; wherein, the trained deep learning model includes a rule engine; The prediction accuracy of each of the trained deep learning models is calculated separately to obtain the prediction accuracy of each of the trained deep learning models. Based on the first preset rule and the prediction accuracy of the trained deep learning model, the model accuracy evaluation value of the trained deep learning model is determined, and the first trained deep learning model corresponding to the maximum model accuracy evaluation value is determined. The diagnostic rule data is embedded into the rule engine of the first trained deep learning model; The first trained deep learning model is determined as the preset disease prediction model.

[0013] More preferably, the electronic medical record data includes a second number of second electronic medical record data, and the second electronic medical record data includes age and gender. Before preprocessing the electronic medical record data based on a preset medical knowledge base model, the method further includes: Based on the stated age and gender, the second number of second electronic medical record data are divided into a fourth number of pediatric electronic medical record data, a fifth number of male electronic medical record data, and a sixth number of female electronic medical record data.

[0014] More preferably, the preprocessing of the electronic medical record data based on a preset medical knowledge base model to obtain a medical record training word vector set specifically includes, Based on a pre-defined medical knowledge base model, the fourth number of children's electronic medical record data, the fifth number of male electronic medical record data, and the sixth number of female electronic medical record data are preprocessed to obtain corresponding training word vector sets for children's medical records, male medical records, and female medical records.

[0015] Preferably, after outputting the predicted disease probability of the first patient, the method further includes: The first feature vector is input into the SHAP interpretation model, which outputs feature vector data and feature vector importance.

[0016] The present invention provides a predictive data processing method based on electronic medical record data and medical knowledge data. It is based on both electronic medical record data and medical knowledge data, and can simultaneously predict multiple disease data. It can respond quickly and realize general-level auxiliary diagnosis, and has the characteristics of fast diagnosis speed and high accuracy. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a predictive data processing method based on electronic medical record data and medical knowledge data, provided in an embodiment of the present invention; Figure 2 A flowchart for constructing a medical knowledge base model provided in an embodiment of the present invention. Figure 3 A flowchart for constructing a disease prediction model provided in an embodiment of the present invention; Figure 4 A flowchart for electronic medical record data preprocessing provided in an embodiment of the present invention; Figure 5 Long Short-Term Memory Network Parameters for a Machine Learning Model Provided in an Embodiment of the Present Invention; Figure 6 A flowchart for training a preset machine learning method to construct a preset disease prediction model using a set of word vectors trained on medical records, provided in an embodiment of the present invention; Figure 7 A flowchart of another predictive data processing method based on electronic medical record data and medical knowledge data provided in an embodiment of the present invention; Figure 8 The prediction accuracy data corresponding to the trained deep learning model provided in the embodiments of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0020] This invention provides a predictive data processing method based on electronic medical record data and medical knowledge data. It applies electronic medical record data and medical knowledge data to a deep learning-based auxiliary diagnostic model, changing the existing auxiliary diagnostic models that only target the classification task of a single disease. This provides a clinical auxiliary data processing method that can simultaneously predict multiple disease data and has a rapid response capability.

[0021] Figure 1 A flowchart of a predictive data processing method based on electronic medical record data and medical knowledge data provided in this embodiment of the invention is shown below. Figure 1 The technical solution of the present invention will be described with reference to specific embodiments.

[0022] This invention provides a predictive data processing method based on electronic medical record data and medical knowledge data, used to assist doctors in analyzing patients' conditions and providing reference predictive data for doctors' diagnoses. The method specifically includes the following steps: Step S1000: Obtain the first electronic medical record data of the first patient, and preprocess the first electronic medical record data based on the preset medical knowledge base model to obtain the first feature vector.

[0023] Specifically, in this embodiment of the invention, the first patient refers to the patient currently receiving treatment, and the first electronic medical record data refers to the electronic medical record data of the patient currently receiving treatment, including chief complaint, present illness, past medical history, personal history, family history, marital and reproductive history, age, gender, date of visit, laboratory test results, and laboratory examination results. When implementing the method provided by this invention, the first electronic medical record data is obtained by inputting from an external device, for example, by inputting from a doctor.

[0024] In this embodiment of the invention, the preset medical knowledge base model is a medical knowledge base model pre-constructed based on medical knowledge data. Figure 2 A flowchart for constructing a medical knowledge base model provided in an embodiment of the present invention, such as... Figure 2 As shown, in a preferred embodiment of the present invention, the construction of the preset medical knowledge base model includes the following steps: Step S1100: Obtain medical knowledge data.

[0025] Specifically, in this embodiment of the invention, the medical knowledge data refers to medical knowledge data obtained / collected from publicly available channels, including clinical guidelines, expert consensus, medical textbooks, and medical journals. In this embodiment, approximately 7,000 clinical guidelines, expert consensus documents, and 900,000 medical documents were collected as authoritative standard data for disease diagnosis. This data is in text format and covers multiple dimensions, including clinical manifestations, laboratory test indicators, imaging features, and pathological results of diseases.

[0026] Step S1200: Perform structured extraction on the medical knowledge data to obtain a first number of first medical knowledge data.

[0027] The first medical knowledge data includes at least one of the following: clinical manifestation data of diseases, laboratory test indicator data, imaging feature data, and pathological result data.

[0028] Specifically, Natural Language Processing (NLP) technology is used to extract and store medical knowledge data stored in text information in a structured manner. Entities, attributes, and relational information such as disease type, symptoms, and examination items are extracted from the text information, which is to obtain a first number of structured first medical knowledge data. Here, the first number represents the number of extracted structured first medical knowledge data. Each first medical knowledge data includes at least one of the following: clinical manifestation data of disease, laboratory test index data, imaging feature data, and pathological result data.

[0029] The extracted structured primary medical knowledge data is organized according to dimensions such as disease type, symptoms, and examination items. Simultaneously, this data is stored using a unified standardized format, such as the International Classification of Diseases (ICD), to ensure data interoperability and comparability.

[0030] Step S1300: Construct a symptom-disease relationship mapping and an examination result-diagnosis conclusion mapping based on the first medical knowledge data, and store them as a structured medical knowledge base model.

[0031] Specifically, the system constructs a mapping system including symptom-disease relationships and examination results-diagnosis conclusions to achieve structured storage of medical knowledge data. Based on these mapping relationships, diagnostic rules are generated according to specific rules, ultimately forming a medical knowledge base model with specific diagnostic rules.

[0032] The method provided in this embodiment of the invention further includes formulating precise diagnostic rules based on authoritative data in a medical knowledge base. Specifically, it generates diagnostic rule data based on a structured medical knowledge base model. These rules can involve a direct correlation between a single symptom and a disease, or they can be a comprehensive analytical logic encompassing multiple symptoms and examination results, or they can incorporate expert experience and the latest research findings. For example, in a specific example of this invention, a glycated hemoglobin (HbA1c) ≥ 6.5% can be diagnosed as diabetes, meaning the test result-diagnosis conclusion is mapped as: Glycated hemoglobin (HbA1c) ≥ 6.5% - Diabetes.

[0033] In an optional embodiment of the present invention, preprocessing the first electronic medical record data based on a preset medical knowledge base model to obtain a first feature vector specifically includes: performing structured processing on the first electronic medical record data using a first preset natural language processing method to obtain first structured data; and performing standardized processing on the first structured data based on the preset medical knowledge base model to obtain the resulting first feature vector. The first preset natural language processing method is a pre-defined natural language processing method, which can be a commonly used natural language processing method selected from the prior art and pre-defined in the method of the present invention.

[0034] In this embodiment of the invention, the first electronic medical record data includes chief complaint, present illness, past medical history, personal history, family history, marital and reproductive history, age, gender, date of visit, laboratory test results, and laboratory examination results. The unstructured information is structured using NLP technology and standardized based on a knowledge base to obtain a Chinese sequence. For example, in a specific example of an embodiment of the present invention, the first electronic medical record data of the first patient includes a medical record examination content: "2018-05-24 16:22:1, chest X-ray examination, increased and blurred lung markings, patchy blurred shadows, slightly worse than before. Both diaphragms are smooth, and the right costophrenic angle is not sharp." After performing structured processing and standardization processing based on a preset medical knowledge base model, the resulting sequence text data, that is, the first feature vector, is "2018-05-24 16:22:1, chest X-ray examination, both lungs, increased and blurred markings, patchy blurred shadows, slightly worse than before, both diaphragms, smooth, right side, costophrenic angle, not sharp".

[0035] Step S2000: Input the first feature vector into the preset disease prediction model and output the predicted disease probability of the first patient.

[0036] Specifically, the first feature vector corresponding to the first patient is used as the input to a preset disease prediction model. After processing by the preset disease prediction model, the predicted disease probability of the first patient is output. In this embodiment of the invention, the predicted disease probability includes disease types and probability values ​​corresponding to the disease types. The predicted disease probability can include multiple diseases and multiple probability values ​​corresponding to each disease, or it can be a single disease and its corresponding probability value.

[0037] In this embodiment of the invention, the preset disease prediction model is a pre-built disease prediction model that is pre-set in the method provided in this embodiment of the invention.

[0038] Figure 3 A flowchart for constructing a disease prediction model provided in an embodiment of the present invention, such as Figure 3 As shown, the preferred embodiment of the present invention provides a disease prediction model that includes the following steps: Step S2100: Obtain electronic medical record data.

[0039] Specifically, in this embodiment of the invention, the electronic medical record data consists of approximately 5.6 million high-quality electronic medical records collected / acquired by authoritative hospitals. Each electronic medical record is considered a second set of electronic medical record data. That is, the electronic medical record data includes a second number of second electronic medical record data. Each second electronic medical record data of the collected / acquired patients includes the chief complaint, present illness, past medical history, personal history, family history, marital and reproductive history, age, gender, date of visit, laboratory test results, and laboratory examination results. Alternatively, the second number can be considered to be 5.6 million.

[0040] In a preferred embodiment of the present invention, after obtaining a second number of second electronic medical record data, the second number of second electronic medical record data are divided into three categories according to the age and gender of each second electronic medical record data: a fourth number of children's electronic medical record data, a fifth number of male electronic medical record data, and a sixth number of female electronic medical record data. The fourth, fifth, and sixth numbers represent the number of children's electronic medical record data, male electronic medical record data, and female children's electronic medical record data, respectively. Considering that some second electronic medical record data may be abnormal in the processing room, it is known that the sum of the fourth, fifth, and sixth numbers is less than or equal to the second number.

[0041] Step S2200: Preprocess the electronic medical record data based on the preset medical knowledge base model to obtain the medical record training word vector set.

[0042] Specifically, in this embodiment of the invention, the preset medical knowledge base model in this step is a medical knowledge base model pre-constructed based on medical knowledge data, that is, the preset medical knowledge base model in step S1000.

[0043] In an optional embodiment of the present invention, the process of preprocessing electronic medical record data based on a preset medical knowledge base model to obtain a medical record training word vector set is as follows: Figure 4 As shown, the specific steps include: Step S2210: The second preset natural language processing method is used to perform structured processing on the second number of second electronic medical record data, and the data is standardized based on the preset medical knowledge base model to obtain the second number of second structured data.

[0044] Each of the second structured data sets includes time information.

[0045] Specifically, in this embodiment of the invention, the second preset natural language processing method is a preset natural language processing method, which can be selected from relevant methods in the prior art. It can be the same as the first preset natural language processing method in this invention, or it can be a method different from the first preset natural language processing method.

[0046] In this embodiment of the invention, unstructured information in each second electronic medical record is processed using NLP technology for structuring, and standardized based on a knowledge base to obtain a Chinese sequence, which is the second structured data. Each second structured data includes time information. For example, in a specific example of this invention, the second electronic medical record data of a tuberculosis patient includes the medical record examination content: "2018-05-24 16:22:1, chest X-ray examination, increased and blurred lung markings, patchy blurred shadows, slightly worse than before. Both diaphragms are smooth, right costophrenic angle is not sharp." After structuring and standardization, the sequence text data is obtained, that is, the second structured data is "2018-05-24 16:22:1, chest X-ray examination, both lungs, increased and blurred markings, patchy blurred shadows, slightly worse than before, both diaphragms, smooth, right costophrenic angle, not sharp". In actual diagnosis, each second electronic medical record data can include multiple content items, such as outpatient visit records, outpatient medical records and outpatient examination reports. Therefore, this step can produce multiple second structured data, and each second structured data has time information.

[0047] Step S2220: For each second structured data, sort and splice it according to the time information to obtain the electronic medical record serialized data.

[0048] Specifically, in this embodiment of the invention, the second structured data is sorted and spliced ​​according to the time information in each second structured data to obtain electronic medical record serialized data.

[0049] In this embodiment of the invention, after the second number of second electronic medical record data are processed in this step, a second number of electronic medical record serialization data are obtained.

[0050] Step S2230: Use a preset word vector generation method to convert the electronic medical record serialized data into medical record training word vectors.

[0051] Specifically, in this embodiment of the invention, the preset word vector generation method is a pre-defined method that can be selected from common methods in the prior art. For example, in this embodiment, the word2vec vector model is selected as the preset word vector generation method and preset in the method provided by this invention. The preset word vector generation method can be used to convert electronic medical record serialization data into medical record training word vectors.

[0052] Specifically, in this embodiment of the invention, all second-number number of electronic medical record serialization data are transformed to obtain second-number number of medical record training word vectors.

[0053] Step S2240: Form a medical record training word vector set based on the second number of medical record training word vectors.

[0054] Specifically, in this embodiment of the invention, a second number of medical record training word vectors are grouped together to form a medical record training word vector set.

[0055] In a preferred embodiment of the present invention, the second number of second electronic medical record data are divided into a fourth number of children's electronic medical record data, a fifth number of male electronic medical record data, and a sixth number of female electronic medical record data. Then, steps S2210 to S2240 are used to process the fourth number of children's electronic medical record data, the fifth number of male electronic medical record data, and the sixth number of female electronic medical record data respectively to obtain the corresponding children's medical record training word vector set, the male medical record training word vector set, and the female medical record training word vector set.

[0056] Step S2300: Use the medical record training word vector set to train the preset machine learning method and build a preset disease prediction model.

[0057] Specifically, in this embodiment of the invention, the preset machine learning method is a general machine learning method selected from the prior art, and the machine learning method adopts a Long Short-Term Memory (LSTM) network, with model parameters as follows: Figure 5 As shown.

[0058] In a preferred embodiment of the present invention, the process of using a medical record training word vector set to train a preset machine learning method and construct a preset disease prediction model is as follows: Figure 6 As shown, the specific steps include: Step S2310: Divide the medical record training word vector set into a training set and a test set according to a preset ratio.

[0059] Specifically, in this embodiment of the invention, the preset ratio is the ratio between the training set and the test set. In one specific example of this embodiment, the preset ratio is 9:1, meaning that the medical record training word vector set is divided into the training set and the test set according to a 9:1 ratio. In this embodiment, the ratio between the training set and the test set can also be determined according to the selected preset machine learning method, for example, it can be divided according to preset ratios such as 8:2, 7:3, and 4:6.

[0060] In a preferred embodiment of the present invention, the training word vector sets for children's medical records, the training word vector sets for male medical records, and the training word vector sets for female medical records are used as the training set and the test set respectively, in a ratio of 9:1.

[0061] Step S2320: Train and test the third number of preset deep learning models using the training set and test set respectively, to obtain the third number of trained deep learning models.

[0062] Specifically, in this embodiment of the invention, in order to obtain the optimal disease prediction model, a third number of preset deep learning models are provided for training and testing, resulting in a third number of trained deep learning models. Each trained deep learning model includes a rule engine.

[0063] In a preferred embodiment of the present invention, the training word vector sets for pediatric medical records, male medical records, and female medical records are used as training and testing sets respectively, in a 9:1 ratio. A third number of preset deep learning models are then trained and tested on these sets, resulting in three trained deep learning models corresponding to the pediatric, male, and female medical record training word vector sets. Each trained deep learning model includes a rule engine.

[0064] Step S2330: Calculate the prediction accuracy of each trained deep learning model to obtain the prediction accuracy of each trained deep learning model.

[0065] Specifically, the prediction accuracy is calculated for each trained deep learning model, resulting in a prediction accuracy corresponding to each model. This prediction accuracy includes the accuracy for predicting the first disease (also referred to as the first disease prediction accuracy), the accuracy for predicting the second disease, and the accuracy for predicting the third disease. The first disease prediction accuracy represents the accuracy rate at which the predicted diagnosis of the first disease matches the actual diagnosis; the second disease prediction accuracy represents the accuracy rate at which the predicted diagnoses of the first two diseases include the actual disease diagnoses; and the third disease prediction accuracy represents the accuracy rate at which the predicted diagnoses of the first three diseases include the actual disease diagnoses.

[0066] In a specific example of this invention, the third quantity is 3. After training the third quantity of preset deep learning models, the accuracy of predicting the first disease, the accuracy of predicting the second disease, and the accuracy of predicting the third disease are calculated for each trained deep learning model to measure the accuracy and reliability of each trained deep learning model in predicting disease diagnosis.

[0067] In a preferred embodiment of the present invention, the prediction accuracy of three trained deep learning models (a third number × 3) obtained by training and testing a third number of preset deep learning models using pediatric medical record training word vector sets, male medical record training word vector sets, and female medical record training word vector sets is calculated to obtain the corresponding third number × 3 prediction accuracy. In a specific example of the present invention, the third number is 3. The prediction accuracy obtained after training and testing a third number of preset deep learning models using pediatric medical record training word vector sets, male medical record training word vector sets, and female medical record training word vector sets is as follows: Figure 8 As shown, the values ​​include the prediction accuracy for the first disease, the prediction accuracy for the second disease, and the prediction accuracy for the third disease.

[0068] Step S2340: Based on the first preset rule and the prediction accuracy of the trained deep learning model, determine the model accuracy evaluation value of the trained deep learning model, and determine the first trained deep learning model corresponding to the maximum model accuracy evaluation value.

[0069] Specifically, in this embodiment of the invention, the first preset rule is a pre-set determination rule, which can be a calculation rule or other rules that can determine the model accuracy evaluation value from the prediction accuracy. The prediction accuracy of each trained deep learning model is processed according to the first preset rule to obtain the model accuracy evaluation value corresponding to each trained deep learning model, i.e., the model accuracy evaluation value of the trained deep learning model.

[0070] In this embodiment of the invention, the prediction accuracy includes the prediction accuracy of the first disease, the prediction accuracy of the second disease, and the prediction accuracy of the third disease. Therefore, the prediction accuracy of each trained deep learning model is processed according to a first preset rule. Specifically, the prediction accuracy of the first disease, the prediction accuracy of the second disease, and / or the prediction accuracy of the third disease corresponding to each trained deep learning model are processed according to the first preset rule to obtain the model accuracy evaluation value corresponding to each trained deep learning model. For example, in an optional solution provided by this embodiment of the invention, the prediction accuracy of the first disease is directly determined as the model accuracy evaluation value corresponding to the trained deep learning model. In another optional solution provided by this embodiment of the invention, the prediction accuracy of the second disease is directly determined as the model accuracy evaluation value corresponding to the trained deep learning model. In yet another optional solution provided by this embodiment of the invention, the prediction accuracy of the third disease is directly determined as the model accuracy evaluation value corresponding to the trained deep learning model. In another optional embodiment of the present invention, the first preset rule is to calculate the weighted value of the prediction accuracy of the first disease and the prediction accuracy of the first two diseases. Then, for each trained deep learning model, the prediction accuracy of the first disease and the prediction accuracy of the first two diseases are weighted to obtain a model accuracy evaluation value. In another optional embodiment of the present invention, the first preset rule is to calculate the weighted value of the prediction accuracy of the first disease and the prediction accuracy of the first three diseases. Then, for each trained deep learning model, the prediction accuracy of the first disease and the prediction accuracy of the first three diseases are weighted to obtain a model accuracy evaluation value. In another optional embodiment of the present invention, the first preset rule is to calculate the weighted value of the prediction accuracy of the first two diseases and the prediction accuracy of the first three diseases. Then, for each trained deep learning model, the prediction accuracy of the first two diseases and the prediction accuracy of the first three diseases are weighted to obtain a model accuracy evaluation value. In another optional embodiment of the present invention, if the first preset rule is to calculate the weighted value of the prediction accuracy of the first disease, the prediction accuracy of the second disease, and the prediction accuracy of the third disease, then the weighted value of the prediction accuracy of the first disease, the prediction accuracy of the second disease, and the prediction accuracy of the third disease of the trained deep learning model is calculated to obtain the model accuracy evaluation value of the trained deep learning model.

[0071] After determining the model accuracy evaluation value for each post-trained deep learning model, these model accuracy evaluation values ​​are compared, and the largest value is determined as the maximum model accuracy evaluation value. The post-trained deep learning model corresponding to the maximum model accuracy evaluation value is then determined as the first post-trained deep learning model.

[0072] Step S2350: Embed the diagnostic rule data into the rule engine of the first trained deep learning model.

[0073] Specifically, the diagnostic rule data of the pre-set medical knowledge base model is embedded into the rule engine of the first trained deep learning model.

[0074] Step S2360: The first trained deep learning model is determined as the preset disease prediction model.

[0075] Specifically, the first trained deep learning model after embedding diagnostic rule data is determined as the disease prediction model and is pre-set as a preset disease prediction model in the method of this embodiment of the invention.

[0076] This invention also provides another predictive data processing method based on electronic medical record data and medical knowledge data. Figure 7 A flowchart of another predictive data processing method based on electronic medical record data and medical knowledge data provided in an embodiment of the present invention is shown below. Figure 7 As shown, in addition to steps S1000 and S2000, the following steps are also included: Step S3000: Input the first feature vector into the SHAP interpretation model and output the feature vector data and feature vector importance.

[0077] Specifically, the SHAP (SHapley Additive exPlanations) interpretability model is a game theory-based machine learning tool designed to unlock the internal logic of a "black box" model and quantify the contribution of each feature to the prediction result. In this embodiment of the invention, to obtain the factors influencing disease prediction, after obtaining the predicted disease probability, a first feature vector is input into the SHAP interpretability model to output the factors affecting the predicted disease probability, i.e., the feature vector data. Simultaneously, the feature vector importance corresponding to the feature vector is also output, i.e., the degree of influence of the feature vector on the diagnostic result. The output feature vector data and feature vector importance can serve as diagnostic references, assisting doctors in making more accurate judgments.

[0078] The present invention provides a predictive data processing method based on electronic medical record data and medical knowledge data. It is based on both electronic medical record data and medical knowledge data, and can simultaneously predict multiple disease data. It can respond quickly and realize general-level auxiliary diagnosis, and has the characteristics of fast diagnosis speed and high accuracy.

[0079] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0080] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0081] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A predictive data processing method based on electronic medical record data and medical knowledge data, characterized in that, The method includes: The first electronic medical record data of the first patient is obtained, and the first electronic medical record data is preprocessed based on a preset medical knowledge base model to obtain the first feature vector; The first feature vector is input into a preset disease prediction model, and the predicted disease probability of the first patient is output. The construction of the preset disease prediction model includes: Obtain the electronic medical record data; wherein the electronic medical record data includes a second number of second electronic medical record data, and the second electronic medical record data includes age and gender; Based on the age and gender, the second number of second electronic medical record data are divided into a fourth number of children's electronic medical record data, a fifth number of male electronic medical record data, and a sixth number of female electronic medical record data. The second preset natural language processing method is used to perform structured processing on the second number of the second electronic medical record data, and then standardized processing is performed based on the preset medical knowledge base model to obtain the second number of second structured data; wherein, each second structured data includes time information; For each of the second structured data, sorting and splicing are performed according to time information to obtain electronic medical record serialized data; The electronic medical record serialization data is converted into medical record training word vectors using a preset word vector generation method; A medical record training word vector set is formed based on the second number of medical record training word vectors; wherein, the medical record training word vector set includes a children's medical record training word vector set, a male medical record training word vector set, and a female medical record training word vector set corresponding to the fourth number of children's electronic medical record data, the fifth number of male electronic medical record data, and the sixth number of female electronic medical record data; The preset disease prediction model is constructed by training the preset machine learning method using the medical record training word vector set.

2. The predictive data processing method according to claim 1, characterized in that, The preprocessing of the first electronic medical record data based on a preset medical knowledge base model to obtain the first feature vector specifically includes: The first electronic medical record data is processed using a first preset natural language processing method to obtain first structured data; The first structured data is standardized based on the preset medical knowledge base model to obtain the first feature vector.

3. The predictive data processing method according to claim 1, characterized in that, The construction of the preset medical knowledge base model includes: Acquire the aforementioned medical knowledge data; The medical knowledge data is structured and extracted to obtain a first number of first medical knowledge data; wherein, the first medical knowledge data includes at least one of disease clinical manifestation data, laboratory test indicator data, imaging feature data, and pathological result data; Based on the first medical knowledge data, a symptom-disease relationship mapping and an examination result-diagnosis conclusion mapping are constructed and stored as a structured medical knowledge base model.

4. The predictive data processing method according to claim 3, characterized in that, The method also includes generating diagnostic rule data based on the structured medical knowledge base model.

5. The predictive data processing method according to claim 4, characterized in that, The step of using the medical record training word vector set to train the preset machine learning method and construct the preset disease prediction model specifically includes: The medical record training word vector set is divided into a training set and a test set according to a preset ratio; The training set and the test set are used to train and test a third number of preset deep learning models respectively, to obtain a third number of trained deep learning models; wherein, the trained deep learning model includes a rule engine; The prediction accuracy of each of the trained deep learning models is calculated separately to obtain the prediction accuracy of each of the trained deep learning models. Based on the first preset rule and the prediction accuracy of the trained deep learning model, the model accuracy evaluation value of the trained deep learning model is determined, and the first trained deep learning model corresponding to the maximum model accuracy evaluation value is determined. The diagnostic rule data is embedded into the rule engine of the first trained deep learning model; The first trained deep learning model is determined as the preset disease prediction model.

6. The predictive data processing method according to claim 1, characterized in that, After outputting the predicted disease probability of the first patient, the method further includes: The first feature vector is input into the SHAP interpretation model, which outputs feature vector data and the feature vector importance corresponding to the feature vector.

Images