Classification model training, virus reference information acquisition method and device

Abstract

The embodiment of the application provides a classification model training method and a virus reference information acquisition method and device. The classification model training method based on virus characteristics comprises the following steps: acquiring virus characteristics and personnel classification information of at least one tested person; and training a preset to-be-trained classification model by using the virus characteristics and the personnel classification information of the tested person, to obtain a classification model based on virus characteristics. The classification model based on virus characteristics for classifying personnel can provide auxiliary suggestions for medical personnel to classify the examined personnel, and improve the efficiency and accuracy of the medical personnel to classify the examined personnel.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method for training a classification model based on virus characteristics, a method for obtaining virus reference information, a device for training a classification model based on virus characteristics, and a device for obtaining virus reference information. Background Technology

[0002] After being infected with a virus, the human body is susceptible to viral diseases. Generally speaking, viral diseases can be characterized by specific clinical manifestations, and the presence of a specific virus in the patient's body can be detected through nucleic acid testing.

[0003] However, viral diseases can present with similar clinical manifestations to other illnesses, making diagnosis based solely on clinical symptoms challenging. Furthermore, the accuracy of nucleic acid testing is affected by sampling methods and testing capabilities, leading to the possibility of false positives and false negatives. This makes it difficult for medical personnel to distinguish individuals with viral diseases from healthy individuals or those with other illnesses. Summary of the Invention

[0004] In view of the above problems, embodiments of this application are proposed to provide a method for training a classification model based on virus features, a method for obtaining virus reference information, a device for training a classification model based on virus features, and a device for obtaining virus reference information, in order to overcome or at least partially solve the above problems.

[0005] To address the aforementioned problems, this application discloses a method for training a classification model based on virus characteristics, comprising:

[0006] Obtain the viral characteristics and personnel classification information of at least one person being tested;

[0007] Using the virus characteristics and personnel classification information of the tested personnel, a preset classification model is trained to obtain a classification model based on virus characteristics.

[0008] Optionally, the virus characteristics include structured characteristics and unstructured characteristics;

[0009] The step of training a pre-defined classification model based on virus characteristics using the virus characteristics and personnel classification information of the tested personnel to obtain a classification model based on virus characteristics includes:

[0010] From the unstructured features, extract medical-related features;

[0011] At least one target training feature is selected from the structured features and the medical-related features;

[0012] Using the target training features and personnel classification information of the tested personnel, a preset classification model to be trained is trained to obtain a classification model based on virus features.

[0013] Optionally, the step of obtaining the viral characteristics and personnel classification information of at least one tested person includes:

[0014] Obtain the medical records of at least one person being tested;

[0015] Using the medical record information, the virus characteristics and personnel classification information are extracted.

[0016] Optionally, the structured features include basic information about the person being tested, and / or, information about the items being checked.

[0017] Optionally, the unstructured features include textual information and image information; wherein the textual information includes the medical record information of the test subject, and / or, inspection text information, and the image information includes inspection image information.

[0018] Optionally, the step of extracting medical-related features from the unstructured features includes:

[0019] Extract at least one medical entity word from the text information;

[0020] Medical association features are determined by using a preset naming standard and the medical entity terms.

[0021] Optionally, the medical association features include at least one of lesion location information, lesion size information, and lesion category information;

[0022] The step of extracting medical-related features from the unstructured features includes:

[0023] The location of the lesion is determined from the examined image information to obtain lesion location information;

[0024] And / or,

[0025] From the examined image information, the image information of the lesion site is extracted; based on the image information of the lesion site, the size of the lesion site is determined, and the lesion size information is obtained;

[0026] And / or,

[0027] A preset image classification model is used to determine the lesion category information corresponding to the examination image information.

[0028] Optionally, the step of selecting at least one target training feature from the structured features and the medical association features includes:

[0029] Determine the importance of the structured features and the medical-related features;

[0030] Based on the importance, at least one target training feature is selected from the structured features and the medical-related features.

[0031] Optionally, the personnel classification information includes virus category information and / or severity information.

[0032] This application also discloses a method for obtaining virus reference information, including:

[0033] Obtain the characteristics of the virus to be tested from the individuals being tested;

[0034] The virus characteristics of the person to be tested are input into a preset virus characteristic-based classification model to obtain the virus reference information of the person to be tested; wherein, the virus characteristic-based classification model obtains the virus characteristics and personnel classification information of at least one person to be tested; the virus characteristics and personnel classification information of the person to be tested are used to train the preset classification model to be trained to obtain the virus characteristic-based classification model.

[0035] Optionally, the virus features to be tested include structured features to be tested and unstructured features to be tested;

[0036] The step of inputting the virus characteristics of the person to be tested into a preset virus characteristic-based classification model to obtain the virus reference information of the person to be tested includes:

[0037] Extract the medical-related features from the unstructured features to be tested;

[0038] The structured features to be tested and the medical association features to be tested are input into a preset classification model based on virus features to obtain the virus reference information of the person to be tested.

[0039] Optionally, the virus reference information includes classification reference information and / or reference basis information.

[0040] Optionally, the classification reference information includes category reference information and / or severity reference information.

[0041] Optionally, the step of inputting the structured features to be tested and the medical association features to be tested into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested includes:

[0042] The structured features to be tested and the medical association features to be tested are input into a preset classification model based on virus features to obtain the classification reference information;

[0043] The reference information is determined based on the classification process of the classification model; or, the reference information is determined based on the preset importance of the structured features to be tested and the medical association features to be tested.

[0044] Optionally, the unstructured features to be tested include text-based information and image-based information; wherein the text-based information includes medical record information of the person to be tested, and / or examination text information, and the image-based information includes examination image information.

[0045] Optionally, the step of extracting the medical-related features from the unstructured features to be tested includes:

[0046] Extract at least one medical entity word from the text information to be tested;

[0047] Using a preset naming standard and the medical entity terms, the medical association features to be tested are determined.

[0048] Optionally, the medical association features to be tested include at least one of the following: information on the location of the lesion to be tested, information on the size of the lesion to be tested, and information on the category of the lesion to be tested;

[0049] The step of extracting the medical-related features from the unstructured features to be tested includes:

[0050] The location of the lesion is determined from the image information to be examined, and the information of the lesion location is obtained.

[0051] And / or,

[0052] From the image information to be examined, the image information of the lesion to be examined is extracted; based on the image information of the lesion to be examined, the size of the lesion is determined, and the size information of the lesion to be examined is obtained.

[0053] And / or,

[0054] A preset image classification model is used to determine the lesion category information corresponding to the image information to be examined.

[0055] This application also discloses a classification model training device based on virus characteristics, comprising:

[0056] The test feature acquisition module is used to acquire the virus characteristics and personnel classification information of at least one test subject.

[0057] The training module is used to train a preset classification model based on virus characteristics using the virus characteristics and personnel classification information of the tested personnel, so as to obtain a classification model based on virus characteristics.

[0058] Optionally, the training module includes:

[0059] The first extraction submodule is used to extract medical-related features from the unstructured features;

[0060] The target feature selection submodule is used to select at least one target training feature from the structured features and the medical association features;

[0061] The training submodule is used to train a preset classification model to obtain a classification model based on virus features by using the target training features and personnel classification information of the tested personnel.

[0062] Optionally, the measured feature acquisition module includes:

[0063] The medical record acquisition submodule is used to acquire the medical record information of at least one subject.

[0064] The detected feature acquisition submodule is used to extract the virus features and personnel classification information using the medical record information.

[0065] Optionally, the structured features include basic information about the person being tested, and / or, information about the items being checked.

[0066] Optionally, the unstructured features include textual information and image information; wherein the textual information includes the medical record information of the test subject, and / or, inspection text information, and the image information includes inspection image information.

[0067] Optionally, the first extraction submodule includes:

[0068] The first entity word extraction unit is used to extract at least one medical entity word from the text information.

[0069] The first association feature determination unit is used to determine medical association features using a preset naming standard and the medical entity terms.

[0070] Optionally, the medical association features include at least one of lesion location information, lesion size information, and lesion category information;

[0071] The first extraction submodule steps include:

[0072] The first location determination unit is used to determine the location of the lesion in the examination image information and obtain lesion location information;

[0073] And / or,

[0074] The first size determination unit is used to extract lesion site image information from the examination image information; and determine the size of the lesion site based on the lesion site image information to obtain the lesion size information.

[0075] And / or,

[0076] The first category determination unit is used to determine the lesion category information corresponding to the examination image information using a preset image classification model.

[0077] Optionally, the target feature selection submodule includes:

[0078] An importance determination unit is used to determine the importance of the structured features and the medical association features;

[0079] The target feature selection unit is used to select at least one target training feature from the structured features and the medical association features according to the importance.

[0080] Optionally, the personnel classification information includes virus category information and / or severity information.

[0081] This application also discloses a device for obtaining virus reference information, including:

[0082] The test feature acquisition module is used to acquire the test virus features of the test personnel; the test virus features include test structured features and test unstructured features;

[0083] The reference information acquisition module is used to input the virus features of the person to be tested into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested; wherein, the virus feature-based classification model obtains the virus features and personnel classification information of at least one person to be tested; using the virus features and personnel classification information of the person to be tested, the preset classification model to be trained is trained to obtain the virus feature-based classification model.

[0084] Optionally, the virus features to be tested include structured features to be tested and unstructured features to be tested;

[0085] The reference information acquisition module includes:

[0086] The second extraction submodule is used to extract the medical-related features to be tested from the unstructured features to be tested;

[0087] The reference information acquisition submodule is used to input the structured features to be tested and the medical association features to be tested into a preset classification model based on virus features to obtain the virus reference information of the person to be tested.

[0088] Optionally, the virus reference information includes classification reference information and / or reference basis information.

[0089] Optionally, the classification reference information includes category reference information and / or severity reference information.

[0090] Optionally, the reference information acquisition submodule includes:

[0091] The classification reference information acquisition unit is used to input the structured features to be tested and the medical association features to be tested into a preset classification model based on virus features to obtain the classification reference information.

[0092] The reference information acquisition unit is used to determine the reference information according to the classification process of the classification model; or, to determine the reference information according to the preset importance of the structured feature to be tested and the medical association feature to be tested.

[0093] Optionally, the unstructured features to be tested include text-based information and image-based information; wherein the text-based information includes medical record information of the person to be tested, and / or examination text information, and the image-based information includes examination image information.

[0094] Optionally, the second extraction submodule includes:

[0095] The second entity word extraction unit is used to extract at least one medical entity word from the text information to be tested;

[0096] The second association feature determination unit is used to determine the medical association features to be tested using a preset naming standard and the medical entity words.

[0097] Optionally, the medical association features to be tested include at least one of the following: information on the location of the lesion to be tested, information on the size of the lesion to be tested, and information on the category of the lesion to be tested;

[0098] The second extraction submodule includes:

[0099] The second part determination unit is used to determine the location of the lesion in the image information to be examined, and to obtain the lesion location information to be examined.

[0100] And / or,

[0101] The second size determination unit is used to extract the image information of the lesion site to be tested from the image information to be tested; and to determine the size of the lesion site based on the image information of the lesion site to be tested, thereby obtaining the size information of the lesion site to be tested.

[0102] And / or,

[0103] The second category determination unit is used to determine the category information of the lesion to be tested corresponding to the image information to be tested using a preset image classification model.

[0104] This application also discloses an apparatus, including:

[0105] One or more processors; and

[0106] One or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, cause the apparatus to perform one or more methods as described in the embodiments of this application.

[0107] This application also discloses one or more machine-readable media storing instructions that, when executed by one or more processors, cause the processors to perform one or more methods as described in this application.

[0108] The embodiments of this application have the following advantages:

[0109] The virus-feature-based classification model training method of this application acquires the virus features and personnel classification information of at least one tested individual. Using the virus features and personnel classification information of the tested individual, a preset classification model is trained to obtain a virus-feature-based classification model. Thus, the virus features and personnel classification information of the tested individual can be used to train the classification model to obtain a classification model capable of classifying individuals based on virus features. This provides auxiliary opinions for medical personnel in classifying examined individuals, improving the efficiency and accuracy of medical personnel in classifying examined individuals. Attached Figure Description

[0110] Figure 1 This is a flowchart illustrating the steps of an embodiment of a classification model training method based on virus characteristics according to this application.

[0111] Figure 2 This is a flowchart illustrating the steps of another embodiment of the classification model training method based on virus characteristics in this application.

[0112] Figure 3 This is a flowchart illustrating the steps of an embodiment of a method for obtaining virus reference information according to this application;

[0113] Figure 4 This is a flowchart illustrating the steps of another embodiment of the method for obtaining virus reference information in this application;

[0114] Figure 5 This is a structural block diagram of an embodiment of a classification device based on virus characteristics according to this application;

[0115] Figure 6This is a structural block diagram of an embodiment of a virus reference information acquisition device according to this application. Detailed Implementation

[0116] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0117] This application refers to Figure 1 The diagram illustrates a flowchart of an embodiment of a classification model training method based on virus features, which may specifically include the following steps:

[0118] Step 101: Obtain the viral characteristics and personnel classification information of at least one person being tested;

[0119] In this embodiment of the application, in order to train the classification model, the virus characteristics of at least one person being tested, as well as the person classification information of the person being tested, can be obtained.

[0120] The viral characteristics can be features associated with viral diseases. Specifically, the viral characteristics may include basic physical information of the tested person, such as name, height, and weight. The viral characteristics may also include image, audio, and video information related to the virus associated with the tested person.

[0121] The test subject can be someone who has already been tested and whose classification results from medical personnel have been obtained. Therefore, the test subject can have personnel classification information. This personnel classification information can be the classification information to which the test subject belongs. This personnel classification information can be used to indicate whether the test subject is likely to have a viral disease. As an example of this application, the personnel classification information can be someone with a viral disease, someone with other diseases, or a healthy person.

[0122] Step 102: Using the virus characteristics and personnel classification information of the tested personnel, the preset classification model to be trained is trained to obtain a classification model based on virus characteristics.

[0123] In the embodiments of this application, the virus characteristics and personnel classification information of the tested personnel can be used to train a preset classification model to be trained, thereby obtaining a classification model for classifying personnel based on virus characteristics.

[0124] The classification model to be trained can be a nearest neighbor algorithm model, a regression model, a tree model (such as a decision tree model, a random forest model, an Adaboost (Adaptive Boosting) model, an Xgboost (Extreme Gradient Boosting) model, a Catboost (Gradient Boosting and Categorical Features) model, a GBDT (Gradient Boosting Decision Tree) model, etc., and this application does not impose any restrictions on it.

[0125] In a specific implementation, the subjects can be divided into a training set and a test set. The target training features of the subjects in the training set are used as the input to the classification model to be trained, and the personnel classification information is used as the output of the classification model to be trained. The classification model is then trained. Subsequently, the classification accuracy of the classification model can be tested using the target training features and personnel classification information of the subjects in the test set. If the accuracy is greater than a preset threshold, the classification model is considered complete and the classification model is obtained.

[0126] In a specific implementation, the virus characteristics of the tested personnel can be used as input to the classification model to be trained, and the personnel classification information can be used as output to iteratively train the classification model. Subsequently, the loss function of the classification model to be trained can be calculated, and based on the loss function, it can be determined whether the classification model to be trained has converged. If the classification model to be trained has converged, it can be considered that the training of the classification model has been completed, and the classification model is obtained.

[0127] The virus-feature-based classification model training method of this application obtains the virus features and personnel classification information of at least one tested individual. Using the virus features and personnel classification information of the tested individual, a preset classification model is trained to obtain a virus-feature-based classification model. Therefore, the virus features and personnel classification information of the tested individual can be used to train the classification model, resulting in a classification model that can classify individuals based on virus features. This provides auxiliary opinions for medical personnel in classifying examined individuals, improving the efficiency and accuracy of medical personnel in classifying examined individuals.

[0128] Reference Figure 2 The diagram illustrates a flowchart of an embodiment of a classification model training method based on virus features, which may specifically include the following steps:

[0129] Step 201: Obtain the medical record information of at least one subject.

[0130] In this embodiment of the application, in order to extract the viral characteristics and personnel classification information of the tested individuals, the medical record information of at least one tested individual can be obtained. The medical record information may include the basic information of the tested individual, medical records recorded by medical personnel for the tested individual, and examination reports and images obtained from the medical examinations conducted on the tested individual.

[0131] Step 202: Using the medical record information, extract the virus characteristics and personnel classification information;

[0132] In this embodiment, the medical record information can be used to extract viral characteristics associated with viral diseases. Simultaneously, medical personnel can classify and record the tested individuals in medical records, and can also extract the personnel classification information of the tested individuals.

[0133] The viral features can be those associated with viral diseases. These viral features can include structured and unstructured features. Structured features can be features belonging to at least one predefined category, such as name, height, and weight. Unstructured features can be features that do not belong to a predefined category. Unstructured features may not be easily represented by simple numerical values ​​or text and are difficult to clearly categorize into a predefined category; for example, complex text information, image information, audio information, and video information.

[0134] In one embodiment of this application, the structured features include basic information of the person being tested, and / or, inspection item information.

[0135] In embodiments of this application, the structured features may include the basic information of the person being tested, and / or, inspection item information.

[0136] The basic information of the test subject can be the individual's basic information, such as name, height, weight, and gender. This basic information can be filled out by the test subject or obtained through measurements performed by medical personnel.

[0137] The examination item information can be the examination items received by the person being tested. For example, this includes the name of the examination item, the name of the examination indicator included in the examination result report, and the corresponding numerical value of the examination indicator. For instance, if the person being tested underwent a complete blood count (CBC), the examination item name could be CBC, and the examination indicator names included in the examination result report could be red blood cell count, hemoglobin, white blood cell count, white blood cell differential count, etc. The corresponding numerical values ​​for the examination indicator names could be 9.57 for white blood cells and 92.0 for hemoglobin, etc.

[0138] Generally, the basic information of the person being tested and the information on the examination items can be recorded in a preset format. For example, a preset table can be used to record the information, so that the basic information of the person being tested and the information on the examination items can be directly extracted based on the category and corresponding text or value in each column of the table.

[0139] Optionally, since different hospitals may use different naming methods for the same examination item, examination item lists from multiple hospitals can be collected as needed. The examination item information can then be categorized based on the name similarity between different examination item information in the lists. A unified naming standard can then be determined based on the categorization results. For examination item information whose names cannot be directly matched to those in the naming standard, the examination item information can be matched with the names in the naming standard based on the similarity between the examination item information and the names in the naming standard.

[0140] In one embodiment of this application, the unstructured features include textual information and image information; wherein the textual information includes medical record information of the tested person and / or inspection text information, and the image information includes inspection image information.

[0141] In this embodiment, the unstructured features may include textual information and image information. The textual information may be complex and cannot be directly categorized into a preset category. The image information requires further processing to obtain structured medical association features.

[0142] The textual information may include the test subject's medical records and / or examination text information. The test subject's medical records may be medical records kept by medical personnel for the test subject. These records may include information such as the test subject's clinical symptoms, epidemiological history, family history, and past medical history. The examination text information may be the examination report text, described in natural language, after the test subject underwent the required examinations. Because both the medical records and examination text information can be complex texts described in natural language, structured features cannot be directly obtained; further processing is required to obtain medically relevant features.

[0143] The image information may include examination image information. This examination image information can be medical images obtained after the subject undergoes an examination, such as chest X-rays, CT scans, enhanced CT scans, and MRI images. Since the examination image information is in the form of images, structured features cannot be directly obtained; further processing is required to obtain medically relevant features.

[0144] In one embodiment of this application, the personnel classification information includes virus category information and / or severity information.

[0145] In this application embodiment, the personnel classification information may include virus category information and / or severity information. The virus category information can be used to indicate whether the tested person is likely to have a viral disease. As an example of this application, the personnel classification information can be: having a viral disease, having other diseases, or being a healthy person. The severity information can be used to identify the severity of the tested person's illness if they have a viral disease or other diseases. As an example of this application, the severity information can be: mild case, severe case, or critical case.

[0146] Step 203: Extract medical-related features from the unstructured features;

[0147] In this embodiment of the application, since the information contained in the unstructured features is relatively complex and cannot be directly classified into a preset category, the unstructured features can be processed to extract medical-related features that may be basically associated with virulence from the unstructured features.

[0148] The medical association features can be those obtained from the unstructured features and can be used to classify the test subjects based on medical information. For example, information such as medications taken and symptoms of the test subjects can be extracted from medical records, and information such as lesion sites and lesion sizes can be extracted from examination images.

[0149] In one embodiment of this application, the step of extracting medical-related features from the unstructured features includes:

[0150] S11, extract at least one medical entity word from the text information;

[0151] In this embodiment, text matching, entity recognition, or other methods can be used to extract at least one medical entity term from the textual information. The medical entity term can be a word with specific meaning in the medical field, such as a person's name, place name, institution name, or medical terminology. Medical terminology can include symptom terms, disease terms, surgical terms, body part terms, epidemiological terms, prescription drug names, and historical examination and testing items.

[0152] In practical implementation, a medical knowledge graph can be pre-defined. This knowledge graph can include multiple medical entity terms and the relationships between them. Text matching can be used to extract words similar to those in the knowledge graph from the text information, thereby obtaining medical entity terms. Based on the relationships between medical entity terms in the knowledge graph, the relationships between the extracted medical entity terms can be determined. Alternatively, a text classification model trained on the knowledge graph can be used to perform entity recognition on the text information. After inputting the text information into the text classification model, the model can determine the medical entity terms in the text information and the relationships between them, and provide the corresponding classification for each medical entity term, such as time, personal name, or medical terminology.

[0153] S12, using preset naming standards and the medical entity terms, determine medical association features.

[0154] In this embodiment, since different hospitals and doctors may use different naming methods to name the same thing, a naming standard can be preset. After extracting medical entity terms, the standard expression of the medical entity terms can be determined based on the similarity between the medical entity terms and the naming standard. Subsequently, the medical association features corresponding to the medical entity terms can be determined based on the standard expression of the medical entity terms and the association relationships between the medical entity terms.

[0155] As an example of this application, the medical entity terms "lung CT image" and "glass opacity" are identified in the text information, and the two are related. Based on a preset naming standard, the standard expression for "lung CT image" is "lung radioactive image". Subsequently, based on the relationship between "lung radioactive image" and "glass opacity", "lung radioactive image" can be used as a category, and "glass opacity" can be used as the value of the category "lung radioactive image" to generate a structured medical association feature "lung radioactive image: glass opacity".

[0156] In one embodiment of this application, the medical association features include at least one of lesion location information, lesion size information, and lesion category information;

[0157] In this embodiment, the lesion location information can be the location information of the lesion in the body of the tested person. The lesion size information can be at least one of the following: area information, volume information, and proportion information of the lesion in the entire organ tissue. The lesion category information can be the category to which the lesion in the body of the tested person belongs, such as cancer, viral disease, etc.

[0158] The step of extracting medical-related features from the unstructured features includes:

[0159] S21, determine the location of the lesion from the examined image information to obtain lesion location information; and / or,

[0160] In this embodiment of the application, the location of the lesion can be determined based on the characteristics of the lesion image in the examination image information, thereby obtaining the lesion location information.

[0161] In a practical implementation, various preset sliding windows of different sizes can be used to select a portion of the examined image information as candidate regions. These candidate regions can then be input into a preset classifier, which outputs the probability that the candidate region is the location of a lesion. Based on the classifier's output, it can be determined whether the candidate region is indeed the location of a lesion. If the candidate region is indeed the location of a lesion, lesion location information recording the candidate region as the lesion location can be generated. The classifier can be trained using training samples that have already been labeled with categories.

[0162] S22, extract lesion site image information from the examined image information; determine the size of the lesion site based on the lesion site image information to obtain the lesion size information; and / or,

[0163] In this embodiment of the application, if there is a lesion in the examination image information, the image information of the lesion site can be extracted, and the size of the lesion site can be determined based on the image information of the lesion site to obtain the lesion size information, so as to judge the severity of the lesion based on the lesion size information.

[0164] In a practical implementation, the edges of the lesion can be determined based on color differences in the examination image information, and the image of the lesion can be extracted from the examination image information to obtain the lesion image information. Subsequently, the area of ​​the lesion can be determined based on the pixel area of ​​the lesion image information and the scaling ratio between the examination image information and the actual examination area to obtain the area information of the lesion.

[0165] When the examination performed on the subject is an imaging examination using a layer-by-layer scanning method, multiple examination images can be obtained. In this case, the lesion site image information can be extracted from each examination image information, and the area information of the lesion site image information can be determined. Subsequently, based on the area information of the lesion site corresponding to each examination image information and the thickness of each layer during the layer-by-layer scanning process, the volume of the lesion site can be determined to obtain the volume information.

[0166] S23, using a preset image classification model, determine the lesion category information corresponding to the examination image information.

[0167] In the embodiments of this application, the lesion may be caused by different diseases. For example, the lesion may be a tumor, and the corresponding disease could be cancer. As another example, if the lesion is a ground-glass opacity, the corresponding disease could be viral pneumonia or ordinary pneumonia.

[0168] To determine the possible disease category corresponding to the lesion site, a preset image classification model can be used to determine the disease type that the examination image information can correspond to, thereby obtaining lesion category information. The lesion type information may include at least one disease category corresponding to the lesion site.

[0169] In a specific implementation, the image classification model can be a multi-label classification model. The image classification model can be trained using training samples already labeled with disease categories. The examination image information can be input into the image classification model, and the image classification model can output at least one disease category corresponding to the examination image information, thereby obtaining the lesion category information.

[0170] Step 204: Select at least one target training feature from the structured features and the medical association features;

[0171] In this embodiment, since the number of structured features and medical-related features can be large, using all features to train the classification model may lead to a deterioration in the model's performance. Therefore, features that have a greater impact on the classification of the tested individuals can be selected from the structured features and medical-related features as target training features.

[0172] In one embodiment of this application, the step of selecting at least one target training feature from the structured features and the medical association features includes:

[0173] S31, determine the importance of the structured features and the medical association features;

[0174] In this embodiment of the application, in order to select features that have a greater impact on the classification of the test subject from the structured features and the medical association features, the importance of the structured features and the medical association features can be determined.

[0175] In specific implementations, methods such as screening based on statistical indicators, correlation coefficients, mutual information methods, random forests, and gradient boosting trees can be used to determine whether the structured features and the medical association features can significantly affect the classification results of the tested individuals, thereby determining the importance of the structured features and the medical association features. This application does not impose any restrictions on this.

[0176] As an example of this application, when determining the importance of the structured features and the medically relevant features based on statistical indicators, a feature can be selected sequentially from the structured features and the medically relevant features, and it can be determined whether the feature tends to be independent of the personnel classification information of the tested person. If the feature tends to be independent of the personnel classification information of the tested person, then the importance of the feature is low. If the feature does not tend to be independent of the personnel classification information of the tested person, then the importance of the feature is high.

[0177] As another example of this application, when determining the importance of the structured features and the medical association features based on the gradient boosting tree, the role of the structured features and the medical association features in the process of constructing the gradient boosting tree can be determined. If a feature is used as a partitioning attribute more often in the process of constructing the gradient boosting tree, then the feature can be considered to have higher importance.

[0178] S32, based on the importance, select at least one target training feature from the structured features and the medical association features.

[0179] In this embodiment, at least one target training feature with higher importance can be selected from the structured features and the medical association features based on the importance level. Training the classification model using the target training feature makes it easier to obtain a classification model with strong generalization ability and high accuracy.

[0180] Step 205: Using the target training features and personnel classification information of the tested personnel, the preset classification model to be trained is trained to obtain a classification model based on virus features.

[0181] In the embodiments of this application, the target training features and personnel classification information of the tested personnel can be used to train a preset classification model to be trained, thereby obtaining a classification model for classifying personnel based on virus features.

[0182] The classification model to be trained can be a nearest neighbor algorithm model, a regression model, a tree model (such as a decision tree model, a random forest model, an Adaboost (Adaptive Boosting) model, an Xgboost (Extreme Gradient Boosting) model, a Catboost (Gradient Boosting and Categorical Features) model, a GBDT (Gradient Boosting Decision Tree) model, etc., and this application does not impose any restrictions on it.

[0183] In a specific implementation, the subjects can be divided into a training set and a test set. The target training features of the subjects in the training set are used as the input to the classification model to be trained, and the personnel classification information is used as the output of the classification model to be trained. The classification model is then trained. Subsequently, the classification accuracy of the classification model can be tested using the target training features and personnel classification information of the subjects in the test set. If the accuracy is greater than a preset threshold, the classification model is considered complete and the classification model is obtained.

[0184] In a specific implementation, the target training features of the tested personnel can be used as input to the classification model to be trained, and the personnel classification information can be used as output to iteratively train the classification model. Subsequently, the loss function of the classification model to be trained can be calculated, and based on the loss function, it can be determined whether the classification model to be trained has converged. If the classification model to be trained has converged, it can be considered that the training of the classification model has been completed, and the classification model is obtained.

[0185] In a specific implementation, the personnel classification information output by the classification model to be trained may include virus category information and / or severity information. When the classification model to be trained is used to perform a single classification task, only the virus category information can be used as the output of the classification model, in which case the model can output labels such as viral disease patient, other disease patient, and normal person. Alternatively, both the virus category information and severity information can be used as the output of the classification model, in which case the model can output labels such as mild viral disease patient, severe viral disease patient, critical viral disease patient, mild other disease patient, severe other disease patient, critical other disease patient, and normal person. When the classification model to be trained is used to perform a multi-output classification task, the virus category information and severity information can be used as two independent outputs, in which case the model can output two independent labels, for example, viral disease patient and mild patient, other disease patient and critical patient.

[0186] The virus-feature-based classification model training method of this application embodiment obtains the medical record information of at least one subject; uses the medical record information to extract virus features and personnel classification information; extracts medically relevant features from the unstructured features; selects at least one target training feature from the structured features and the medically relevant features; and trains a preset classification model to be trained using the target training feature and personnel classification information of the subject to obtain a virus-feature-based classification model. Therefore, the target training feature and personnel classification information of the subject can be used to train the classification model to be trained, thereby obtaining a classification model that can classify personnel based on virus features, providing auxiliary opinions for medical personnel to classify examinees, and improving the efficiency and accuracy of medical personnel in classifying examinees.

[0187] Reference Figure 3 The diagram illustrates a flowchart of an embodiment of a method for obtaining virus reference information according to this application, which may specifically include the following steps:

[0188] Step 301: Obtain the characteristics of the virus to be tested from the person to be tested;

[0189] In this embodiment, a virus-based classification model can be used to provide medical personnel with auxiliary opinions for classifying individuals to be tested. This allows for the acquisition of the virus characteristics of the individuals to be tested, providing information that can be used as input to the virus-based classification model.

[0190] The viral features to be tested can be characteristics associated with viral diseases obtained from the person being tested. Specifically, the viral features to be tested can include basic physical information of the person being tested, such as name, height, and weight. The viral features can also include image, audio, and video information related to the person being tested and the virus.

[0191] Step 302: Input the virus features of the person to be tested into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested; wherein, the virus feature-based classification model obtains the virus features and personnel classification information of at least one person to be tested; using the virus features and personnel classification information of the person to be tested, the preset classification model to be trained is trained to obtain the virus feature-based classification model.

[0192] In this embodiment, the characteristics of the virus to be tested can be input into a preset virus-feature-based classification model to obtain the virus reference information of the person to be tested. The virus reference information may include category reference information and the probability information corresponding to that category reference information, so that medical personnel can know which category the person to be tested may belong to and the probability of belonging to that category.

[0193] The category reference information can be used to indicate whether the person being tested is likely to have a viral disease. As an example of this application, the category reference information can be: having a viral disease, having other diseases, or being a healthy person.

[0194] In this embodiment of the application, the virus feature-based classification model obtains the virus features and personnel classification information of at least one person being tested; using the virus features and personnel classification information of the person being tested, a preset classification model to be trained is trained to obtain the virus feature-based classification model.

[0195] The classification model based on virus features can be a nearest neighbor algorithm model, a regression model, a tree model (such as a decision tree model, a random forest model, an Adaboost (Adaptive Boosting) model, an Xgboost (Extreme Gradient Boosting) model, a Catboost (Gradient Boosting and Categorical Features) model, a GBDT (Gradient Boosting Decision Tree) model, etc., and this application does not limit it.

[0196] The method for obtaining virus reference information according to embodiments of this application acquires the virus characteristics of a person to be tested; the virus characteristics of the person to be tested are input into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested. Therefore, the virus reference information output by the virus feature-based classification model can be used to assist medical personnel in classifying the person to be tested, improving the accuracy of the classification.

[0197] Reference Figure 4 The diagram illustrates a flowchart of an embodiment of a method for obtaining virus reference information according to this application, which may specifically include the following steps:

[0198] Step 401: Obtain the virus characteristics of the person to be tested; the virus characteristics include structured characteristics and unstructured characteristics.

[0199] In this embodiment, a virus-based classification model can be used to provide medical personnel with auxiliary opinions for classifying individuals to be tested. This allows for the acquisition of the virus characteristics of the individuals to be tested, providing information that can be used as input to the virus-based classification model.

[0200] The virus features to be tested can be features obtained from the individuals being tested and associated with viral diseases. These virus features can include structured features and unstructured features. The structured features can belong to at least one predefined category, such as name, height, and weight. The unstructured features can be features that do not belong to any predefined category. The unstructured features may not be easily represented by simple numerical values ​​or text and are difficult to clearly categorize into a predefined category; for example, complex text information, image information, audio information, and video information.

[0201] In one embodiment of this application, the unstructured features to be tested include text-type information and image-type information to be tested; wherein, the text-type information to be tested includes medical record information of the person to be tested, and / or examination text information to be tested, and the image-type information to be tested includes examination image information to be tested.

[0202] In this embodiment, the unstructured feature to be tested may include text-based information and image-based information. The text-based information may be complex and cannot be directly categorized into a preset category. The image-based information requires further processing to obtain structured medical association features.

[0203] The text information to be tested may include the medical record information of the person being tested, and / or examination text information. The medical record information of the person being tested may be medical records recorded by medical personnel for the person being tested. It may contain information such as the person being tested's clinical symptoms, epidemiological history, family history, and past medical history. The examination text information may be the examination report text described in natural language after the person being tested underwent the examination. Because the medical record information and examination text information to be tested can be complex text described in natural language, the structured features to be tested cannot be obtained directly, and further processing is required to obtain the medical association features to be tested.

[0204] The image information to be tested may include examination image information. This examination image information can be medical images obtained after the person undergoes an examination, such as chest X-rays, CT scans, enhanced CT scans, and MRI images. Since the examination image information is in the form of images, the structured features to be tested cannot be directly obtained; further processing is required to obtain the relevant medical features.

[0205] Step 402: Extract the medical-related features from the unstructured features to be tested;

[0206] In this embodiment of the application, since the information contained in the unstructured feature to be tested is relatively complex and cannot be directly classified into a preset category, the unstructured feature to be tested can be processed to extract medical association features that may be associated with viral diseases.

[0207] The medical association features can be those obtainable from the unstructured features and used to classify the test subjects. For example, information such as medications taken and symptoms of the test subjects extracted from medical records, and information such as lesion sites and lesion sizes of the test subjects extracted from examination images.

[0208] In practical implementation, natural language processing can be used to identify medically relevant features contained in complex text. Alternatively, a pre-defined image classification model can be used to classify the examination images, and the classification results can be used as medically relevant features. The size of lesions in the examination images of the examinee can also be identified to obtain the size information of the lesions.

[0209] In one embodiment of this application, the step of extracting the medical-related features from the unstructured features to be tested includes:

[0210] S41, extract at least one medical entity word from the text information to be tested;

[0211] In this embodiment, text matching, entity recognition, or other methods can be used to extract at least one medical entity term from the text information to be tested. The medical entity term can be a word with specific meaning in the medical field, such as a person's name, place name, institution name, or medical terminology. Medical terminology can include symptom terms, disease terms, surgical terms, body part terms, epidemiological terms, prescription drug names, and historical examination and testing items.

[0212] In a practical implementation, a medical knowledge graph can be pre-defined. This knowledge graph can include multiple medical entity terms and the relationships between them. Text matching can be used to extract words similar to those in the knowledge graph from the text to be tested, thus obtaining medical entity terms. Based on the relationships between these medical entity terms in the knowledge graph, the relationships between the extracted medical entity terms can be determined. Alternatively, a text classification model trained on the knowledge graph can be used to perform entity recognition on the text to be tested. After inputting the text to be tested into the text classification model, the model can determine the medical entity terms in the text and the relationships between them, and provide the corresponding classification for each medical entity term, such as time, personal name, or medical terminology.

[0213] S42, using a preset naming standard, determine the medical association features to be tested corresponding to the medical entity words.

[0214] In this embodiment, since different hospitals and doctors may use different naming methods to name the same thing, a naming standard can be preset. After extracting medical entity terms, the standard expression of the medical entity terms can be determined based on the similarity between the medical entity terms and the naming standard. Subsequently, the medical association features to be tested corresponding to the medical entity terms can be determined based on the standard expression of the medical entity terms and the association relationships between the medical entity terms.

[0215] In one embodiment of this application, the medical association feature to be tested includes at least one of the following: information on the location of the lesion to be tested, information on the size of the lesion to be tested, and information on the category of the lesion to be tested.

[0216] In this embodiment, the information about the lesion site to be tested can be the location information of the lesion in the body of the person being tested. The information about the size of the lesion to be tested can be at least one of the following: area information, volume information, and proportion information of the lesion site in the body of the person being tested. The information about the category of the lesion to be tested can be the category to which the lesion in the body of the person being tested belongs, such as cancer, viral disease, etc.

[0217] The step of extracting the medical-related features from the unstructured features to be tested includes:

[0218] S51, determine the location of the lesion in the image information to be examined, and obtain the information of the lesion location; and / or,

[0219] In this embodiment of the application, the location of the lesion can be determined based on the characteristics of the lesion image in the image information to be examined, thereby obtaining the lesion location information.

[0220] In a practical implementation, various preset sliding windows of different sizes can be used to select a portion of the image information to be examined, as candidate regions. These candidate regions can then be input into a preset classifier, which outputs the probability that the candidate region is the location of a lesion. Based on the classifier's output, it can be determined whether the candidate region is the location of a lesion. If the candidate region is the location of a lesion, information about the lesion site can be generated, recording that the candidate region is the location of a lesion. The classifier can be trained using training samples that have already been labeled with categories.

[0221] S52, extract the image information of the lesion site to be tested from the image information to be tested; determine the size of the lesion site based on the image information of the lesion site to be tested, and obtain the size information of the lesion site to be tested; and / or,

[0222] In this embodiment of the application, if there is a lesion in the examination image information, the image information of the lesion site to be tested can be extracted, and the size of the lesion site can be determined based on the image information of the lesion site to be tested to obtain the size information of the lesion to be tested, so as to judge the severity of the lesion based on the size information of the lesion to be tested.

[0223] In a practical implementation, the edge of the lesion can be determined based on the color differences in the image information to be examined, and the image of the lesion can be extracted from the image information to obtain the image information of the lesion to be examined. Subsequently, the area of ​​the lesion to be examined can be determined based on the pixel area of ​​the image information of the lesion to be examined and the scaling ratio between the image information to be examined and the actual examination area, thus obtaining the area information of the lesion.

[0224] When the examination performed on the person being examined is an imaging examination using a layer-by-layer scanning method, multiple images of the examination can be obtained. In this case, the image information of the lesion site to be examined can be extracted from each image, and the area information of the lesion site can be determined. Subsequently, based on the area information of the lesion site corresponding to each image and the thickness of each layer during the layer-by-layer scanning process, the volume of the lesion site can be determined to obtain the volume information.

[0225] S53, using a preset image classification model, determine the lesion category information corresponding to the image information to be examined.

[0226] In the embodiments of this application, the lesion may be caused by different diseases. For example, the lesion may be a tumor, and the corresponding disease could be cancer. As another example, if the lesion is a ground-glass opacity, the corresponding disease could be viral pneumonia or ordinary pneumonia.

[0227] To determine the possible disease category corresponding to the lesion site, a preset image classification model can be used to determine the disease type that the image information to be examined can correspond to, thereby obtaining the lesion category information. The lesion type information may include at least one disease category corresponding to the lesion site.

[0228] In a specific implementation, the image classification model can be a multi-label classification model. The image classification model can be trained using training samples already labeled with disease categories. The image information to be examined can be input into the image classification model, and the image classification model can output at least one disease category corresponding to the image information to be examined, thereby obtaining the lesion category information to be examined.

[0229] Step 403: Input the structured features to be tested and the medical association features to be tested into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested; wherein, the virus feature-based classification model obtains the virus features and personnel classification information of at least one person to be tested, the virus features including structured features and unstructured features; extract medical association features from the unstructured features; select at least one target training feature from the structured features and the medical association features; and train the preset classification model to be trained using the target training features and personnel classification information of the person to be tested.

[0230] In this embodiment, the structured features to be tested and the medical association features to be tested can be input into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested. The virus reference information may include category reference information and the probability information corresponding to that category reference information, so that medical personnel can know which category the person to be tested may belong to and the probability of belonging to that category.

[0231] The category reference information can be used to indicate whether the person being tested is likely to have a viral disease. As an example of this application, the category reference information can be: having a viral disease, having other diseases, or being a healthy person.

[0232] In this embodiment of the application, the virus feature-based classification model is obtained by acquiring the virus features and personnel classification information of at least one subject. The virus features include structured features and unstructured features. Among the unstructured features, medical association features are extracted. Among the structured features and the medical association features, at least one target training feature is selected. The target training feature and personnel classification information of the subject are used to train a preset classification model to be trained.

[0233] The classification model based on virus features can be a nearest neighbor algorithm model, a regression model, a tree model (such as a decision tree model, a random forest model, an Adaboost (Adaptive Boosting) model, an Xgboost (Extreme Gradient Boosting) model, a Catboost (Gradient Boosting and Categorical Features) model, a GBDT (Gradient Boosting Decision Tree) model, etc., and this application does not limit it.

[0234] In one embodiment of this application, the virus reference information includes classification reference information and / or reference basis information.

[0235] In this embodiment, the classification reference information may include the category information and probability information corresponding to the person to be tested, output by the virus feature-based classification model. This provides a reference for medical personnel to classify the person to be tested.

[0236] In this embodiment, the reference information can be the basis information for the classification reference information output by the virus feature-based classification model. Medical personnel can then use this reference information to understand which characteristics of the individual being tested are used to determine the classification reference information output by the virus feature-based classification model. This allows the virus feature-based classification model to better assist medical personnel in classifying the individuals being tested.

[0237] In one embodiment of this application, the step of inputting the virus characteristics of the person to be tested into a preset virus characteristic-based classification model to obtain the virus reference information of the person to be tested includes:

[0238] S61, input the structured features to be tested and the medical association features to be tested into a preset classification model based on virus features to obtain the classification reference information;

[0239] In this embodiment of the application, the structured features to be tested and the medical association features to be tested can be input into a preset classification model based on virus features. The classification model based on virus features can output the classification reference information to provide a reference for medical personnel to classify the individuals to be tested.

[0240] In one embodiment of this application, the classification reference message may specifically include category reference information and / or severity reference information.

[0241] The category reference information can be used to represent the category information and probability information corresponding to the person being tested. As an example of this application, the category reference information may include patients with viral diseases, patients with other diseases, and normal individuals, as well as the probability that the person being tested belongs to each category.

[0242] The severity reference information can be used to indicate the severity of the illness of the person being tested. As an example of this application, the severity reference information may include mild patients, severe patients, and critical patients.

[0243] In a specific implementation, the classification reference information output by the virus-feature-based classification model may only include the category reference information. Therefore, the virus-feature-based classification model can provide a reference for whether the person being tested is likely to be classified as a viral disease patient, allowing medical personnel to determine which category the person being tested might belong to and the probability of belonging to that category.

[0244] In a specific implementation, the classification reference information output by the virus-feature-based classification model can simultaneously include both category reference information and severity reference information. Therefore, the virus-feature-based classification model can provide a reference for whether the person being tested is classified as a viral disease patient, and also for the severity of the disease to which the person being tested should be classified. Medical personnel can determine which category the person being tested might belong to, the probability of belonging to that category, and further, the severity of the person's disease, facilitating subsequent treatment services.

[0245] S62, determine the reference information according to the classification process of the classification model; or, determine the reference information according to the preset importance of the structured features to be tested and the medical association features to be tested.

[0246] In this embodiment of the application, in order to understand which characteristics of the person to be tested are used to determine the classification reference information output by the virus feature-based classification model, further reference information can be determined to better assist medical personnel in classifying the person to be tested. As an example of this application, the reference information may include fever for 7 days, travel to an epidemic area in the past two weeks, glassy shadows on lung radiographic imaging, lesion size accounting for 25%, and abnormalities in multiple blood routine indicators.

[0247] In a specific implementation, the method for determining the reference information can be determined based on the type of the virus feature-based classification model.

[0248] When the virus-based classification model is a tree model, since the classification process of the tree model is visual, the structured features and / or medical association features to be tested that the tree model relies on at each node of the classification process can be determined. Therefore, reference information can be generated based on the structured features and / or medical association features to be tested that the classification process of the classification model relies on.

[0249] When the virus-based classification model is non-interpretable, it's impossible to directly obtain how the model determines the classification reference information. In this case, the reference information can be obtained by arranging the structured features and medical association features of the tested individual in descending order of importance based on their preset importance.

[0250] The preset importance is determined during the training of the classification model based on whether the structured features and medical association features can significantly influence the classification results of the subjects. Specifically, the importance can be determined using methods such as screening based on statistical indicators, correlation coefficients, mutual information methods, random forests, and gradient boosting trees.

[0251] The method for obtaining virus reference information according to embodiments of this application acquires the virus characteristics of the person to be tested; extracts the medically relevant features from the unstructured features; inputs the structured features and the medically relevant features into a preset virus feature-based classification model to obtain the classification reference information; determines the reference basis information according to the classification process of the classification model; or, determines the reference basis information according to the preset importance of the structured features and the medically relevant features. Therefore, the virus reference information output by the virus feature-based classification model can be used to assist medical personnel in classifying the person to be tested, improving the accuracy of the classification.

[0252] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of this application.

[0253] Reference Figure 5 The diagram illustrates a structural block diagram of an embodiment of a classification model training device based on virus features, which may specifically include the following modules:

[0254] The test feature acquisition module 501 is used to acquire the virus features and personnel classification information of at least one test subject. The virus features include structured features and unstructured features.

[0255] The training module 502 is used to train a preset classification model based on virus features by using the virus features and personnel classification information of the tested personnel.

[0256] In one embodiment of this application, the training module includes:

[0257] The first extraction submodule is used to extract medical-related features from the unstructured features;

[0258] The target feature selection submodule is used to select at least one target training feature from the structured features and the medical association features;

[0259] The training submodule is used to train a preset classification model to obtain a classification model based on virus features by using the target training features and personnel classification information of the tested personnel.

[0260] In one embodiment of this application, the measured feature acquisition module 501 includes:

[0261] The medical record acquisition submodule is used to acquire the medical record information of at least one subject.

[0262] The detected feature acquisition submodule is used to extract the virus features and personnel classification information using the medical record information.

[0263] In one embodiment of this application, the structured features include basic information of the person being tested, and / or, inspection item information.

[0264] In one embodiment of this application, the unstructured features include textual information and image information; wherein the textual information includes medical record information of the tested person and / or inspection text information, and the image information includes inspection image information.

[0265] In one embodiment of this application, the first extraction submodule includes:

[0266] The first entity word extraction unit is used to extract at least one medical entity word from the text information.

[0267] The first association feature determination unit is used to determine medical association features using a preset naming standard and the medical entity terms.

[0268] In one embodiment of this application, the medical association features include at least one of lesion location information, lesion size information, and lesion category information;

[0269] The first extraction submodule steps include:

[0270] The first location determination unit is used to determine the location of the lesion in the examination image information and obtain lesion location information;

[0271] And / or,

[0272] The first size determination unit is used to extract lesion site image information from the examination image information; and determine the size of the lesion site based on the lesion site image information to obtain the lesion size information.

[0273] And / or,

[0274] The first category determination unit is used to determine the lesion category information corresponding to the examination image information using a preset image classification model.

[0275] In one embodiment of this application, the target feature selection submodule includes:

[0276] An importance determination unit is used to determine the importance of the structured features and the medical association features;

[0277] The target feature selection unit is used to select at least one target training feature from the structured features and the medical association features according to the importance.

[0278] In one embodiment of this application, the personnel classification information includes virus category information and / or severity information.

[0279] Reference Figure 6 The diagram illustrates a structural block diagram of an embodiment of a virus reference information acquisition device according to this application, which may specifically include the following modules:

[0280] The test feature acquisition module 601 is used to acquire the test virus features of the test personnel; the test virus features include test structured features and test unstructured features;

[0281] The reference information acquisition module 602 is used to input the virus features of the person to be tested into a preset virus feature-based classification model to obtain the virus reference information of the person to be tested; wherein, the virus feature-based classification model is obtained by acquiring the virus features and personnel classification information of at least one person to be tested; the virus features and personnel classification information of the person to be tested are used to train the preset classification model to be trained to obtain the virus feature-based classification model.

[0282] Optionally, the virus features to be tested include structured features to be tested and unstructured features to be tested;

[0283] The reference information acquisition module includes:

[0284] The second extraction submodule is used to extract the medical-related features to be tested from the unstructured features to be tested;

[0285] The reference information acquisition submodule is used to input the structured features to be tested and the medical association features to be tested into a preset classification model based on virus features to obtain the virus reference information of the person to be tested.

[0286] In one embodiment of this application, the virus reference information includes classification reference information and / or reference basis information.

[0287] In one embodiment of this application, the classification reference information includes category reference information and / or severity reference information.

[0288] In one embodiment of this application, the reference information acquisition submodule includes:

[0289] The classification reference information acquisition unit is used to input the structured features to be tested and the medical association features to be tested into a preset classification model based on virus features to obtain the classification reference information.

[0290] The reference information acquisition unit is used to determine the reference information according to the classification process of the classification model; or, to determine the reference information according to the preset importance of the structured feature to be tested and the medical association feature to be tested.

[0291] In one embodiment of this application, the unstructured features to be tested include text-type information and image-type information to be tested; wherein, the text-type information to be tested includes medical record information of the person to be tested, and / or examination text information to be tested, and the image-type information to be tested includes examination image information to be tested.

[0292] In one embodiment of this application, the second extraction submodule includes:

[0293] The second entity word extraction unit is used to extract at least one medical entity word from the text information to be tested;

[0294] The second association feature determination unit is used to determine the medical association features to be tested using a preset naming standard and the medical entity words.

[0295] In one embodiment of this application, the medical association feature to be tested includes at least one of the following: information on the location of the lesion to be tested, information on the size of the lesion to be tested, and information on the category of the lesion to be tested.

[0296] The second extraction submodule includes:

[0297] The second part determination unit is used to determine the location of the lesion in the image information to be examined, and to obtain the lesion location information to be examined.

[0298] And / or,

[0299] The second size determination unit is used to extract the image information of the lesion site to be tested from the image information to be tested; and to determine the size of the lesion site based on the image information of the lesion site to be tested, thereby obtaining the size information of the lesion site to be tested.

[0300] And / or,

[0301] The second category determination unit is used to determine the category information of the lesion to be tested corresponding to the image information to be tested using a preset image classification model.

[0302] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0303] This application also provides an apparatus, including:

[0304] One or more processors; and

[0305] One or more machine-readable media having instructions stored thereon, which, when executed by the one or more processors, cause the device to perform the methods described in the embodiments of this application.

[0306] This application also provides one or more machine-readable media storing instructions that, when executed by one or more processors, cause the processors to perform the methods described in this application.

[0307] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0308] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, embodiments of this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0309] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0310] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0311] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0312] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0313] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0314] The above provides a detailed description of a virus feature-based classification model training method, a virus reference information acquisition method, a virus feature-based classification model training device, and a virus reference information acquisition device provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for training a classification model based on virus features, characterized in that, The method comprises the following steps: obtaining virus characteristics and personnel classification information of at least one tested person; the personnel classification information comprises virus category information and severity information; the virus characteristics comprise structured characteristics and unstructured characteristics; the structured characteristics comprise basic information of the tested person and examination item information; the unstructured characteristics comprise text information and image information; wherein the examination item information is a unified naming standard determined based on classification results of classification of examination item information in examination item lists of multiple hospitals based on name similarity between different examination item information in the examination item lists; training a preset to-be-trained classification model based on the virus characteristics and the personnel classification information of the tested person to obtain a classification model based on virus characteristics; wherein the method comprises the following steps: extracting medical correlation characteristics from the unstructured characteristics; selecting at least one target training feature from the structured characteristics and the medical correlation characteristics; training the preset to-be-trained classification model based on the target training feature and the personnel classification information of the tested person to obtain the classification model based on virus characteristics; the step of selecting at least one target training feature from the structured characteristics and the medical correlation characteristics comprises the following steps: determining the importance of the structured characteristics and the medical correlation characteristics; and selecting at least one target training feature from the structured characteristics and the medical correlation characteristics according to the importance.

2. The method of claim 1, wherein, The step of obtaining the virus characteristics and the personnel classification information of at least one tested person comprises the following steps: obtaining medical record information of at least one tested person; extracting the virus characteristics and the personnel classification information based on the medical record information.

3. The method of claim 1, wherein, The text information comprises medical record information of the tested person and / or examination text information, and the image information comprises examination image information.

4. The method according to claim 1 or 3, characterized in that, The step of extracting medical correlation characteristics from the unstructured characteristics comprises the following steps: extracting at least one medical entity word from the text information; determining medical correlation characteristics based on a preset naming standard and the medical entity word.

5. The method of claim 3, wherein, The medical correlation characteristics comprise at least one of lesion site information, lesion size information and lesion category information. The step of extracting medical correlation characteristics from the unstructured characteristics comprises the following steps: determining a lesion site in the examination image information to obtain lesion site information; and / or extracting lesion site image information from the examination image information; determining the size of the lesion site based on the lesion site image information to obtain the lesion size information; and / or determining lesion category information corresponding to the examination image information based on a preset image classification model.

6. A method of acquiring virus reference information, characterized by, The method comprises the following steps: obtaining virus characteristics of a to-be-tested person; Input the to-be-tested virus feature of the to-be-tested personnel into a preset virus feature-based classification model to obtain virus reference information of the to-be-tested personnel; wherein, the virus feature-based classification model is obtained by acquiring virus features and personnel classification information of at least one tested personnel; the virus features and the personnel classification information of the tested personnel are used to train a preset to-be-trained classification model to obtain the virus feature-based classification model; the virus features include structured features and unstructured features; the structured features include basic information of the tested personnel and examination item information; the unstructured features include text information and image information; wherein, the examination item information is obtained by collecting examination item lists of multiple hospitals, classifying the examination item information based on name similarity between different examination item information in the examination item lists, and determining a unified naming standard based on the classification result; The step of training the preset to-be-trained classification model by using the virus features and the personnel classification information of the tested personnel to obtain the virus feature-based classification model includes: extracting medical correlation features from the unstructured features; selecting at least one target training feature from the structured features and the medical correlation features; and training the preset to-be-trained classification model by using the target training features and the personnel classification information of the tested personnel to obtain the virus feature-based classification model; the step of selecting at least one target training feature from the structured features and the medical correlation features includes: determining the importance of the structured features and the medical correlation features; and selecting at least one target training feature from the structured features and the medical correlation features according to the importance; the personnel classification information includes virus category information and severity information.

7. The method of claim 6, wherein, The to-be-tested virus features include to-be-tested structured features and to-be-tested unstructured features. The step of inputting the to-be-tested virus features of the to-be-tested personnel into the preset virus feature-based classification model to obtain the virus reference information of the to-be-tested personnel includes: extracting to-be-tested medical correlation features from the to-be-tested unstructured features; inputting the to-be-tested structured features and the to-be-tested medical correlation features into the preset virus feature-based classification model to obtain the virus reference information of the to-be-tested personnel.

8. The method of claim 7, wherein, The virus reference information includes classification reference information and / or reference basis information.

9. The method of claim 8, wherein, The classification reference information includes category reference information and / or severity reference information.

10. The method according to claim 8 or 9, characterized in that, The step of inputting the to-be-tested structured features and the to-be-tested medical correlation features into the preset virus feature-based classification model to obtain the virus reference information of the to-be-tested personnel includes: inputting the to-be-tested structured features and the to-be-tested medical correlation features into the preset virus feature-based classification model to obtain the classification reference information; determining the reference basis information according to a classification process of the classification model or according to a preset importance of the to-be-tested structured features and the to-be-tested medical correlation features.

11. The method of claim 7, wherein, The to-be-tested unstructured features include to-be-tested text information and to-be-tested image information; the to-be-tested text information includes to-be-tested personnel medical record information and / or to-be-tested examination text information, and the to-be-tested image information includes to-be-tested examination image information.

12. The method of claim 7 or 11, wherein, The step of extracting to-be-tested medical correlation features from the to-be-tested unstructured features includes: extracting at least one medical entity word from the to-be-tested text information; determining to-be-tested medical correlation features by using a preset naming standard and the medical entity word.

13. The method of claim 11, wherein, The to-be-tested medical correlation features include at least one of to-be-tested lesion site information, to-be-tested lesion size information and to-be-tested lesion category information. The step of extracting to-be-tested medical correlation features from the to-be-tested unstructured features includes: determining a lesion site in the to-be-tested examination image information to obtain the to-be-tested lesion site information; and / or, extracting to-be-tested lesion site image information from the to-be-tested examination image information, and determining the size of the lesion site based on the to-be-tested lesion site image information to obtain the to-be-tested lesion size information; and / or, determining to-be-tested lesion category information corresponding to the to-be-tested examination image information by using a preset image classification model. 14.A device for training a classification model based on virus features, characterized in that, The method includes: a measured feature acquisition module configured to acquire virus features and personnel classification information of at least one measured person; the virus features include structured features and unstructured features; the structured features include measured personnel basic information and examination item information; the unstructured features include text information and image information; the examination item information is a unified naming standard determined based on classification results of classification of examination item information in examination item lists of multiple hospitals based on name similarity between different examination item information in the examination item lists; a training module configured to train a preset to-be-trained classification model based on the virus features and the personnel classification information of the measured person to obtain a classification model based on virus features; the personnel classification information includes virus category information and severity information; the training module includes: a first extraction submodule configured to extract medical correlation features from the unstructured features; a target feature selection submodule configured to select at least one target training feature from the structured features and the medical correlation features; a training submodule configured to train the preset to-be-trained classification model based on the target training features and the personnel classification information of the measured person to obtain the classification model based on virus features; the target feature selection submodule includes: an importance determination unit configured to determine the importance of the structured features and the medical correlation features; and a target feature selection unit configured to select at least one target training feature from the structured features and the medical correlation features according to the importance.

15. The apparatus of claim 14, wherein, The measured feature acquisition module includes: a medical record acquisition submodule configured to acquire medical record information of at least one measured person; a measured feature acquisition submodule configured to extract the virus features and the personnel classification information by using the medical record information.

16. The apparatus of claim 14, wherein, The text information includes medical record information of the person to be tested, and / or examination text information, and the image information includes examination image information.

17. The apparatus of claim 14 or 16, wherein, The first extraction submodule includes: A first entity word extraction unit is configured to extract at least one medical entity word from the text information. A first association feature determination unit is configured to determine medical association features based on a preset naming standard and the medical entity word.

18. The apparatus of claim 16, wherein, The medical association features include at least one of lesion site information, lesion size information, and lesion category information. The first extraction submodule includes: A first site determination unit is configured to determine a lesion site in the examination image information to obtain the lesion site information. And / or, A first size determination unit is configured to extract lesion site image information from the examination image information, determine the size of the lesion site based on the lesion site image information, and obtain the lesion size information. And / or, A first category determination unit is configured to determine lesion category information corresponding to the examination image information based on a preset image classification model.

19. A device for acquiring virus reference information, characterized in that, The method includes: A to-be-tested feature acquisition module is configured to acquire to-be-tested virus features of a person to be tested. A reference information acquisition module is configured to input the to-be-tested virus features of the person to be tested into a preset classification model based on virus features to obtain virus reference information of the person to be tested. The classification model based on virus features is obtained by acquiring virus features and personnel classification information of at least one person to be tested, training a preset to-be-trained classification model based on the virus features and the personnel classification information, and obtaining the classification model based on virus features. The personnel classification information includes virus category information and severity information. The virus features include structured features and unstructured features. The structured features include basic information and examination item information of the person to be tested. The unstructured features include text information and image information. The examination item information is obtained by collecting examination item lists of multiple hospitals, classifying the examination item information based on name similarity between different examination item information in the examination item lists, and determining a unified naming standard based on a classification result. Training the preset to-be-trained classification model based on the virus features and the personnel classification information to obtain the classification model based on virus features includes extracting medical association features from the unstructured features, selecting at least one target training feature from the structured features and the medical association features, and training the preset to-be-trained classification model based on the target training feature and the personnel classification information to obtain the classification model based on virus features. The step of selecting at least one target training feature from the structured features and the medical association features includes determining the importance of the structured features and the medical association features, and selecting at least one target training feature from the structured features and the medical association features based on the importance.

20. The apparatus of claim 19, wherein, The to-be-tested virus features include to-be-tested structured features and to-be-tested unstructured features. The reference information obtaining module comprises: A second extraction submodule for extracting a to-be-tested medical correlation feature from the to-be-tested unstructured feature; A reference information obtaining submodule for inputting the to-be-tested structured feature and the to-be-tested medical correlation feature into a preset classification model based on a virus feature to obtain virus reference information of the to-be-tested person.

21. The apparatus of claim 20, wherein, The virus reference information comprises classification reference information and / or reference basis information.

22. The apparatus of claim 21, wherein, The classification reference information comprises category reference information and / or severity reference information.

23. The apparatus of claim 21 or 22, wherein, The reference information obtaining submodule comprises: A classification reference information obtaining unit for inputting the to-be-tested structured feature and the to-be-tested medical correlation feature into a preset classification model based on a virus feature to obtain the classification reference information; A reference basis information obtaining unit for determining the reference basis information according to a classification process of the classification model or according to a preset importance of the to-be-tested structured feature and the to-be-tested medical correlation feature.

24. The apparatus of claim 20, wherein The to-be-tested unstructured feature comprises to-be-tested text information and to-be-tested image information; wherein the to-be-tested text information comprises to-be-tested person medical record information and / or to-be-tested examination text information, and the to-be-tested image information comprises to-be-tested examination image information.

25. The apparatus of claim 20 or 24, wherein, The second extraction submodule comprises: A second entity word extraction unit for extracting at least one medical entity word from the to-be-tested text information; A second correlation feature determination unit for determining a to-be-tested medical correlation feature by using a preset naming standard and the medical entity word.

26. The apparatus of claim 24, wherein, The to-be-tested medical correlation feature comprises at least one of to-be-tested lesion site information, to-be-tested lesion size information and to-be-tested lesion category information; The second extraction submodule comprises: A second site determination unit for determining a lesion site in the to-be-tested examination image information to obtain the to-be-tested lesion site information; And / or A second size determination unit for extracting to-be-tested lesion site image information from the to-be-tested examination image information, determining a size of the lesion site based on the to-be-tested lesion site image information to obtain the to-be-tested lesion size information; And / or A second category determination unit for determining to-be-tested lesion category information corresponding to the to-be-tested examination image information by using a preset image classification model.

27. An apparatus comprising: comprise: one or more processors; and one or more machine-readable media having instructions stored thereon that, when executed by the one or more processors, cause the apparatus to perform one or more of the methods of claims 1-5 or 6-13.

28. one or more machine-readable media having instructions stored thereon that, when executed by one or more processors, cause the processors to perform one or more of the methods of claims 1-5 or 6-13.

Images