Model training method, disease risk prediction method, device, equipment and medium

By using the preset prediction model in the training of the disease risk prediction model, the problem of excessive workload of the marker is solved, the labeling accuracy and efficiency are improved, and the target risk level is accurately predicted.

CN114093516BActive Publication Date: 2024-11-15CAS & MEDICALSYSTEM AI RES LAB SUZHOU CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202111420480.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-26
Publication Date
2024-11-15
Estimated Expiration
2041-11-26

AI Technical Summary

Technical Problem

During the training of the disease risk prediction model, the labeler needs to label a large number of training samples, resulting in excessive workload and excessive workload.

Method used

By obtaining the label field and the preset prediction model, the label field is processed and used to determine the label risk level of the target disease. The preset prediction model is a pre-trained neural network model, which is used to predict the preset risk level of the target disease. The label will be marked Enter the preset prediction model in the field to obtain the preset risk level and send it to the disease risk marking client, so that the markers can mark the preset risk level as a reference, reducing the unreferenced marking and saving workload.

Benefits of technology

The workload of labeling personnel during model training is significantly reduced, the labeling accuracy and efficiency are improved, and the target prediction model used to predict the target risk level of target diseases is trained through multiple sets of training samples.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114093516B_ABST
    Figure CN114093516B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention discloses a model training method, a disease risk prediction method, an apparatus, a device and a medium. The model training method includes: obtaining a labeling field and a preset prediction model, wherein the labeling field is a field obtained after processing the sample case data for determining the labeling risk degree of the target disease, and the preset prediction model is a pre-trained neural network model for predicting the preset risk degree of the target disease; inputting each labeling field into the preset prediction model to obtain the preset risk degree, and sending each labeling field and the preset risk degree to the disease risk labeling client; receiving the labeling risk degree returned by the disease risk labeling client, and training the original prediction model based on the training samples including each labeling field and the labeling risk degree, to obtain a target prediction model for predicting the target risk degree of the target disease. The technical solution of the embodiment of the present invention can effectively reduce the workload of labeling personnel.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of data processing technology, and in particular to a model training method, a disease risk prediction method, an apparatus, a device and a medium. Background Art

[0002] At present, with the rapid development of machine learning technology, machine learning models have gradually been applied to disease risk prediction, such as predicting the risk of cardiovascular and cerebrovascular complications. In order to obtain better model prediction capabilities, labelers are required to label a large number of training samples during the model training process. Obviously, this is too much work for labelers and the workload is too heavy. Summary of the invention

[0003] The embodiments of the present invention provide a model training method, a disease risk prediction method, an apparatus, a device and a medium, which solve the problem of excessive workload of labeling personnel during the model training process.

[0004] In a first aspect, an embodiment of the present invention provides a model training method, which may include:

[0005] Obtaining annotated fields and preset prediction models, wherein the annotated fields are fields obtained after processing sample case data for determining annotated risk levels of target diseases, and the preset prediction model is a pre-trained neural network model for predicting preset risk levels of target diseases;

[0006] Input each annotation field into a preset prediction model to obtain a preset risk level, and send each annotation field and the preset risk level to a disease risk annotation client;

[0007] Receive the annotated risk level returned by the disease risk annotation client, and use each annotated field and the annotated risk level as a set of training samples;

[0008] The original prediction model is trained based on multiple groups of training samples to obtain a target prediction model that can be used to predict the target risk level of the target disease.

[0009] In a second aspect, an embodiment of the present invention further provides a disease risk prediction method, which may include:

[0010] Obtaining a prediction field and a target prediction model generated according to the model generation method provided by any embodiment of the present invention, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease;

[0011] Each prediction field is input into the target prediction model, and the target risk level is obtained according to the output result of the target prediction model.

[0012] In a third aspect, an embodiment of the present invention further provides a model training device, which may include:

[0013] A preset prediction model acquisition module, used to acquire annotated fields and preset prediction models, wherein the annotated fields are fields obtained after processing sample case data for determining the annotated risk level of the target disease, and the preset prediction model is a pre-trained neural network model for predicting the preset risk level of the target disease;

[0014] A preset risk level sending module is used to input each annotation field into a preset prediction model to obtain a preset risk level, and send each annotation field and the preset risk level to a disease risk annotation client;

[0015] The training sample acquisition module is used to receive the annotated risk level returned by the disease risk annotation client, and use each annotated field and the annotated risk level as a set of training samples;

[0016] The target prediction model obtaining module is used to train the original prediction model based on multiple groups of training samples to obtain a target prediction model for predicting the target risk level of the target disease.

[0017] In a fourth aspect, an embodiment of the present invention further provides a disease risk prediction device, which may include:

[0018] A target prediction model acquisition module, used to acquire prediction fields and a target prediction model generated according to the model generation method provided in any embodiment of the present invention, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease;

[0019] The target risk degree obtaining module is used to input each prediction field into the target prediction model and obtain the target risk degree according to the output result of the target prediction model.

[0020] In a fifth aspect, an embodiment of the present invention further provides an electronic device, which may include:

[0021] one or more processors;

[0022] A memory for storing one or more programs;

[0023] When one or more programs are executed by one or more processors, the one or more processors implement the model training method or disease risk prediction method provided by any embodiment of the present invention.

[0024] In a sixth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a model training method or a disease risk prediction method as provided in any embodiment of the present invention.

[0025] The technical solution of the embodiment of the present invention obtains the annotation field and the preset prediction model, wherein the annotation field is a field for determining the annotation risk degree of the target disease obtained after processing the sample case data, and the preset prediction model is a pre-trained neural network model for predicting the preset risk degree of the target disease; each annotation field is input into the preset prediction model to obtain the preset risk degree, and each annotation field and the preset risk degree are sent to the disease risk annotation client, so that the annotation personnel can annotate with the preset risk degree as a reference, instead of annotating without any reference, thereby saving the workload of the annotation personnel; further, the annotation risk degree returned by the disease risk annotation client is received, and each annotation field and the annotation risk degree are used as a group of training samples; the original prediction model is trained based on multiple groups of training samples to obtain a target prediction model for predicting the target risk degree of the target disease. The above technical solution, by sending the preset risk degree that can be used as a reference to the disease risk annotation client, so that the annotation personnel can annotate on this basis, thereby greatly reducing the workload of the annotation personnel during the model training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of a model training method in Embodiment 1 of the present invention;

[0027] Figure 2 is a flowchart of an optional example of a model training method in Embodiment 1 of the present invention;

[0028] Figure 3 is a flow chart of a disease risk prediction method in Embodiment 2 of the present invention;

[0029] Figure 4 is a flowchart of an optional example of a disease risk prediction method in Embodiment 2 of the present invention;

[0030] Figure 5 It is a structural block diagram of a model training device in Embodiment 3 of the present invention;

[0031] Figure 6 is a structural block diagram of a disease risk prediction device in Embodiment 4 of the present invention;

[0032] Figure 7 It is a structural schematic diagram of an electronic device in Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It is also necessary to explain that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0034] Embodiment 1

[0035] Figure 1 It is a flow chart of a model training method provided in the first embodiment of the present invention. This embodiment can be applied to the case of reducing the labeling workload in the model training process. The method can be executed by the model training device provided in the embodiment of the present invention, and the device can be implemented by software and / or hardware. The device can be integrated on an electronic device, and the electronic device can be a model training server.

[0036] See also Figure 1 The method of the embodiment of the present invention specifically includes the following steps:

[0037] S110, obtaining annotated fields and a preset prediction model, wherein the annotated fields are fields obtained after processing sample case data for determining the annotated risk level of the target disease, and the preset prediction model is a pre-trained neural network model for predicting the preset risk level of the target disease.

[0038] Among them, the sample case data can be case data involved in the model training process, which can be case data of subjects suffering from the target disease, or case data of subjects not suffering from the target disease; it can include various types of data, such as basic data, physical examination data, family history data, disease history data, diagnosis data, etc., which are not specifically limited here. The target disease can be a disease whose target risk level is to be predicted during the model application process, and the target risk level can indicate the severity of the target disease that has been suffered, the possible degree of suffering from the target disease in the future, etc.; it can be expressed in various ways, such as scores, Chinese characters (such as mild, moderate, severe), English letters (such as A+, A-, B), etc., which are not specifically limited here. The annotation field can be a field obtained after processing the sample case data for determining the annotation risk level of the target disease, which can be a field pre-set by the annotation personnel according to the target disease or obtained after a large number of experiments, such as age, weight, height, whether there is a family history of cardiovascular and cerebrovascular diseases, whether to drink, whether to smoke, work intensity, blood routine examination, etc., which are not specifically limited here. The annotated risk level may be the risk level of the target disease annotated by the annotator based on the annotated fields. It should be noted that the essence of the risk level and the target risk level are both risk levels. The naming here is only for distinguishing the risk levels in different scenarios, and it is not a specific limitation on the substantive content. The preset prediction model may be a pre-trained neural network model for predicting the preset risk level of the target disease, that is, the preset risk level may be the risk level predicted by the preset prediction model based on the annotated fields. Like the annotated risk level and the target risk level, it is also named only for distinguishing the risk levels in different scenarios, and it is not a specific limitation on the substantive content.

[0039] S120, input each annotation field into a preset prediction model to obtain a preset risk level, and send each annotation field and the preset risk level to a disease risk annotation client.

[0040] Among them, each annotation field is input into the preset prediction model, and the preset risk degree is obtained according to the output result of the preset prediction model, thereby achieving the effect of pre-annotating the risk degree of the target disease. Each annotation field and the preset risk degree are sent to the disease risk annotation client, so that the annotator can annotate according to each annotation field and the preset risk degree displayed by the disease risk annotation client. It should be noted that in addition to obtaining each annotation field, the annotator can also obtain the preset risk degree. In other words, it can be annotated on the basis of each annotation field and the preset risk degree, especially on the basis of the preset risk degree. Compared with the scheme of annotating only according to each annotation field, the preset risk degree is used as the annotation reference result, which allows the annotator to annotate with it as a reference, rather than annotating without any reference, thereby saving the workload of the annotator and improving the annotation accuracy of the annotator.

[0041] In practical applications, it is optional, because the annotation is performed at the granularity of case data, and the number of case data of a certain object may be one, two or more. When the generation time of the case data is used as the benchmark, the case data generated later may have reflected whether the object suffers from the target disease. Therefore, the case data generated earlier can be annotated based on the case data generated later, thereby further improving the annotation accuracy. Specifically, the annotation fields and the corresponding preset risk levels obtained after processing the case data of the same object can be sent to the disease risk annotation client together, that is, the annotation fields and preset risk levels corresponding to the case data of the same object are sent to the disease risk annotation client, so that the annotation personnel can annotate the annotation fields corresponding to the case data generated earlier according to the annotation fields and preset risk levels corresponding to the case data generated later.

[0042] S130: Receive the annotated risk level returned by the disease risk annotation client, and use each annotated field and the annotated risk level as a set of training samples.

[0043] Among them, the annotated risk level is the risk level obtained by the annotator. After receiving the annotated risk level returned by the disease risk annotation client, it and each annotated field can be used as a set of training samples.

[0044] S140. Train the original prediction model based on multiple groups of training samples to obtain a target prediction model for predicting a target risk level of a target disease.

[0045] Among them, the original prediction model can be a neural network model to be trained that can realize the prediction function, which can be the same or different neural network model as the preset prediction model, and is not specifically limited here. The original prediction model is trained based on multiple groups of training samples to obtain a target prediction model that can be used to predict the target risk level of the target disease. In practical applications, optionally, in order to reduce the model training time, the original prediction model can be constructed based on the extreme gradient boosting (eXtreme GradientBoosting, XGBoost) algorithm. Compared with conventional neural network models, the neural network model constructed based on the XGBoost algorithm has the characteristics of short training time and the ability to quickly obtain important information in the annotation field, and it requires fewer hyperparameters to be adjusted, and is easier to optimize during the model training process, which is very suitable for situations where rapid model deployment and application are required.

[0046] The technical solution of the embodiment of the present invention obtains the annotation field and the preset prediction model, wherein the annotation field is a field for determining the annotation risk degree of the target disease obtained after processing the sample case data, and the preset prediction model is a pre-trained neural network model for predicting the preset risk degree of the target disease; each annotation field is input into the preset prediction model to obtain the preset risk degree, and each annotation field and the preset risk degree are sent to the disease risk annotation client, so that the annotation personnel can annotate with the preset risk degree as a reference, instead of annotating without any reference, thereby saving the workload of the annotation personnel; further, the annotation risk degree returned by the disease risk annotation client is received, and each annotation field and the annotation risk degree are used as a group of training samples; the original prediction model is trained based on multiple groups of training samples to obtain a target prediction model for predicting the target risk degree of the target disease. The above technical solution, by sending the preset risk degree that can be used as a reference to the disease risk annotation client, so that the annotation personnel can annotate on this basis, thereby greatly reducing the workload of the annotation personnel during the model training process.

[0047] On this basis, an optional technical solution is to input each annotation field into a preset prediction model to obtain a preset risk level, and send each annotation field and the preset risk level to the disease risk annotation client, which may include: for each annotation field of each sample case data in each sample case data derived from the same object, input each annotation field of each sample case data into a preset prediction model to obtain a preset risk level; determine the preset risk trend of each preset risk level according to the generation time of each sample case data, and send each annotation field, each preset risk level, and the preset risk trend of each sample case data to the disease risk annotation client. Among them, as mentioned above, the number of case data of the same object may be one, two or more, and not only the case data generated later is instructive for the annotation process of the case data generated earlier, but also the changing trend of each preset risk level corresponding to these sequentially generated case data (i.e., the preset risk trend) is also instructive for the annotation process of each case data. Therefore, the preset risk levels corresponding to each case data of the same object can be obtained, and then the preset risk trends of each preset risk level can be obtained based on the generation time of each case data. Therefore, while sending the annotation fields and preset risk levels corresponding to each case data to the disease risk annotation client, the preset risk trends are also sent together. This allows the annotation personnel to perform annotations based on the preset risk trends as a reference, further reducing the workload of the annotation personnel.

[0048] Another optional technical solution, the above-mentioned model prediction method may also include: obtaining a case field, wherein the case field is a field obtained after processing the sample case data and is used to filter out the annotation field from each case field; sending each case field to the disease risk annotation client, and receiving the annotation field returned by the disease risk annotation client. Among them, the case field may be a field obtained after the model training device processes the case data, some case fields may be annotation fields that the annotator needs to refer to when annotating, and some may be irrelevant fields that do not need to be referenced (i.e., fields that are not related to the target disease). Therefore, after obtaining each case field, they can be sent to the disease risk annotation client so that the annotator can filter out the annotation field therefrom. It should be noted that since the annotation field is the case field involved in the model requirement process, only these annotation fields will be involved in the model application process, that is, the annotation field only needs to be input into the target prediction model for prediction, and the remaining irrelevant fields do not need to be input.

[0049] On this basis, optionally, the above-mentioned model prediction method may also include: obtaining sample case data; performing natural language processing on the sample case data to obtain case fields; and / or performing desensitization processing on the sample case data and / or case fields, and updating according to the desensitization processing results; and / or normalizing the sample case data from the same object to the same object in the future; and / or determining the missing degree and / or logical degree of the sample case data, if the sample case data is used as data to be deleted according to the determination result, the data to be deleted is sent to the disease risk labeling client, and the data to be deleted is processed according to the deletion processing result returned by the disease risk labeling client. Among them, since the sample case data may include large sections of unstructured language description information, taking the discharge diagnosis information as an example, it may describe the various vital sign parameters, the discharge medication, the discharge time and other contents. In order to convert it into content that the computer can understand, the sample case data can be processed based on natural language processing (natural language processing, NLP) to obtain structured case fields. Since there may be sensitive information in the sample case data and / or case fields, such as name, birthday, mobile phone number, ID card number, etc., it can be desensitized, thereby better protecting the patient's personal privacy. As mentioned above, since the number of case data of the same object may be one, two or more, the sample case data from the same object can be normalized to the same object, so as to analyze the case data of the same object as a whole. In order to ensure the data quality of the sample case data, for each sample case data, its missing degree and / or logical degree can be determined. The higher the missing degree, the worse the integrity of the sample case data, and the lower the logical degree, the higher the possibility that there is a problem with the sample case data, such as age is a negative number, hypertension is above 300, etc., so it can be determined whether it is data to be deleted based on the missing degree and / or logical degree of the sample case data. On this basis, since the computer's determination process is mechanized, there may be some special cases in medicine. Therefore, in order to ensure data quality while making better use of each sample case data, the data to be deleted can be sent to the disease risk annotation client, so that the annotation personnel can determine whether these data to be deleted are really sample case data that cannot be applied to the model training process, and then determine whether to delete or retain these data to be deleted based on the deletion processing results returned by the disease risk annotation client. The above steps can be understood as the governance process of sample case data. In actual applications, one or more of them can be selected for execution according to actual needs, and no specific restrictions are made here.

[0050] Another optional technical solution, the above-mentioned model prediction method may also include: obtaining sample case data and a pre-set data processing strategy, wherein the data processing strategy includes a field mapping strategy and / or a data format conversion strategy; performing data processing on the sample case data based on the data processing strategy, and updating the sample case data according to the data processing result. Among them, since the source channels of each sample case data may be different, they may come from the same or different hospital information systems (HIS), such as medical imaging information systems (PACS), clinical information systems (CIS), radiology information systems (RIS), etc., and these HIS may come from the same or different manufacturers, which means that the sample case data derived from these HIS may differ in field content (such as what fields are included, what each field represents, etc.), data format (such as what format is used to store the sample case data), etc. On this basis, in order to ensure the effective application of subsequent sample case data, and to avoid the long data collection cycle caused by the need to coordinate with multiple manufacturers due to differences in field content, data format, etc., the sample case data can be processed based on the data processing strategy, which can be a field mapping strategy (i.e., what field in a certain HIS is mapped to what field), a data format conversion strategy, etc.; it can be pre-set according to the source channel, that is, a corresponding data processing strategy is set for each source channel, thereby obtaining standardized sample case data. In practical applications, optionally, the above data processing process can be implemented in various forms such as data interface, Excel table, etc., which are not specifically limited here.

[0051] In order to better understand the specific implementation process of the above steps, the model training method of this embodiment is exemplarily described below with reference to specific examples. Figure 2In terms of data collection, standardized sample case data is collected based on the data interface docking; in terms of data processing and annotation, the sample case data is structured and normalized to obtain case fields, and these case fields are sent to the disease risk annotation client, and the annotation fields returned by it are received; the annotation fields are pre-annotated based on the preset prediction model to obtain the preset risk level, and each annotation field and the preset risk level are sent to the disease risk annotation client so that the annotation personnel (such as doctors) can annotate based on them; after receiving the annotation review result returned by the disease risk annotation client, if the annotation review result is passed, the model is trained based on the annotation risk level and each annotation field annotated by the doctor, otherwise the pre-annotation is performed again. In terms of model training, in order to train a target prediction model with better prediction performance, 80% of each training sample can be taken as a training set, and the rest can be used as a test set. At the same time, in order to ensure the effectiveness of the test, the training samples from the same object will not be assigned to the training set and the test set at the same time. After several candidate prediction models are obtained based on the training set, the accuracy of these candidate prediction models can be tested by rolling back the training samples based on the test set, and then the target prediction model can be selected from these candidate prediction models based on the accuracy. In addition, in order to increase the training speed, multiple Nvidia1080Ti graphics cards can be used for parallel training, which can also facilitate the migration of the trained target prediction model to the disease risk prediction server without graphics cards for operation.

[0052] Embodiment 2

[0053] Figure 3 It is a flow chart of a disease risk prediction method provided in the second embodiment of the present invention. This embodiment is applicable to the case of predicting disease risk, especially to the case of predicting disease risk based on a target prediction model obtained with a smaller amount of annotation work. The method can be performed by the disease risk prediction device provided in the embodiment of the present invention, which can be implemented by software and / or hardware, and the device can be integrated in an electronic device, which can be a disease risk prediction server.

[0054] See also Figure 3 The method of this embodiment may specifically include the following steps:

[0055] S210, obtaining a prediction field and a target prediction model generated according to the model generation method provided by any embodiment of the present invention, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease.

[0056] Among them, the prediction field can be a field used to predict the target risk level of the target disease obtained after processing the target case data. It should be noted that the prediction field and the annotation field are essentially the same. The naming here is only to distinguish the model application stage and the model training stage, and it is not a specific limitation of the substantive content; the situation of sample case data and target case data is similar, and will not be repeated here. In practical applications, optionally, the acquired target case data may be a large section of unstructured language description information, which can be structured here to obtain a case field that can be understood by a computer; it may also be a case field that has been structured. Optionally, the acquired and / or processed case field may be the prediction field, or it may be an irrelevant field that does not need to be involved in the model application stage. At this time, the prediction field can be extracted from each case field.

[0057] S220: Input each prediction field into the target prediction model, and obtain the target risk level according to the output result of the target prediction model.

[0058] Among them, since the target prediction model can be a pre-trained neural network model that can predict the target risk level of the target disease, each prediction field can be input into the target prediction model, and the target risk level can be obtained according to the output result of the target prediction model, which can be used as a reference factor for medical staff when diagnosing the target disease, thereby timely discovering and early preventing and treating the target disease, which can greatly reduce the risk of patients developing the target disease and reduce the waste of medical resources.

[0059] The technical solution of the embodiment of the present invention obtains prediction fields that can be used to predict the target risk level of the target disease after processing the target case data, and then inputs each prediction field into the target prediction model to obtain the target risk level according to the output result of the target prediction model. The above technical solution achieves the effect of accurately predicting the target risk level.

[0060] In order to better understand the specific implementation process of the above steps, the disease risk prediction method of this embodiment is exemplarily described below with reference to specific examples. Figure 4 , obtain the target case data, perform structured processing and prediction field extraction processing on it in turn to obtain the prediction field; input the prediction field into the target prediction model to obtain the target risk level of the target disease.

[0061] Embodiment 3

[0062] Figure 5This is a structural block diagram of a model training device provided in Embodiment 3 of the present invention, and the device is used to execute the model training method provided in any of the above embodiments. The device and the model training method in the above embodiments belong to the same inventive concept, and the details not described in detail in the embodiment of the model training device can refer to the embodiment of the above model training method. Figure 5 The device may specifically include: a preset prediction model acquisition module 310, a preset risk level sending module 320, a training sample acquisition module 330 and a target prediction model acquisition module 340.

[0063] The preset prediction model acquisition module 310 is used to acquire the annotated field and the preset prediction model, wherein the annotated field is a field obtained after processing the sample case data for determining the annotated risk level of the target disease, and the preset prediction model is a pre-trained neural network model for predicting the preset risk level of the target disease;

[0064] The preset risk level sending module 320 is used to input each annotation field into the preset prediction model to obtain the preset risk level, and send each annotation field and the preset risk level to the disease risk annotation client;

[0065] The training sample obtaining module 330 is used to receive the annotated risk level returned by the disease risk annotation client, and use each annotated field and the annotated risk level as a set of training samples;

[0066] The target prediction model obtaining module 340 is used to train the original prediction model based on multiple groups of training samples to obtain a target prediction model for predicting the target risk level of the target disease.

[0067] Optionally, the preset risk level sending module 320 may include:

[0068] A preset risk level obtaining unit is used to input each labeled field of each sample case data in each sample case data from the same object into a preset prediction model to obtain a preset risk level;

[0069] The preset risk trend sending unit is used to determine the preset risk trend of each preset risk level according to the generation time of each sample case data, and send each marked field, each preset risk level, and preset risk trend of each sample case data to the disease risk marking client.

[0070] Optionally, the above-mentioned model training device may further include:

[0071] A case field acquisition module is used to acquire case fields, wherein the case field is a field obtained after processing the sample case data and used to filter out the marked field from each case field;

[0072] The annotation field receiving module is used to send each case field to the disease risk annotation client and receive the annotation fields returned by the disease risk annotation client.

[0073] On this basis, optionally, the above-mentioned model training device may further include:

[0074] A sample case data acquisition module is used to acquire sample case data;

[0075] A natural language processing module is used to perform natural language processing on sample case data to obtain case fields; and / or,

[0076] A desensitization processing module, used to perform desensitization processing on sample case data and / or case fields, and update according to the desensitization processing results; and / or,

[0077] A normalization processing module, used to normalize the sample case data from the same subject to the same subject in the future; and / or,

[0078] The module for processing data to be deleted is used to determine the degree of missingness and / or logical degree of sample case data. If the sample case data is determined as data to be deleted based on the determination result, the data to be deleted will be sent to the disease risk labeling client, and the data to be deleted will be processed according to the deletion processing result returned by the disease risk labeling client.

[0079] Optionally, the above-mentioned model training device may further include:

[0080] The data processing module is used to obtain sample case data and a pre-set data processing strategy, process the sample case data based on the data processing strategy, and update the sample case data according to the data processing result, wherein the data processing strategy includes a field mapping strategy and / or a data format conversion strategy.

[0081] Optionally, the original prediction model includes a neural network model to be trained that is constructed based on the extreme gradient boosting algorithm.

[0082] The model training device provided in the third embodiment of the present invention obtains the annotation field and the preset prediction model through the preset prediction model acquisition module, wherein the annotation field is a field for determining the annotation risk degree of the target disease obtained after processing the sample case data, and the preset prediction model is a pre-trained neural network model for predicting the preset risk degree of the target disease; each annotation field is input into the preset prediction model through the preset risk degree sending module to obtain the preset risk degree, and each annotation field and the preset risk degree are sent to the disease risk annotation client, so that the annotation personnel can annotate with the preset risk degree as a reference, rather than annotating without any reference, thereby saving the workload of the annotation personnel; further, the annotation risk degree returned by the disease risk annotation client is received through the training sample acquisition module, and each annotation field and the annotation risk degree are used as a group of training samples; the target prediction model acquisition module trains the original prediction model based on multiple groups of training samples to obtain a target prediction model for predicting the target risk degree of the target disease. The above device sends the preset risk degree that can be used as a reference to the disease risk annotation client, so that the annotation personnel can annotate on this basis, thereby greatly reducing the workload of the annotation personnel during the model training process.

[0083] The model training device provided in the embodiment of the present invention can execute the model training method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0084] It is worth noting that in the embodiment of the above-mentioned model training device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0085] Embodiment 4

[0086] Figure 6 This is a structural block diagram of a disease risk prediction device provided in Embodiment 4 of the present invention. The device is used to execute the disease risk prediction method provided in any of the above embodiments. The device and the disease risk prediction method in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the disease risk prediction device, please refer to the embodiments of the above disease risk prediction method. Figure 6 The device may specifically include: a target prediction model acquisition module 410 and a target risk level acquisition module 420.

[0087] The target prediction model acquisition module 410 is used to acquire the prediction field and the target prediction model generated by the model generation method provided by any embodiment of the present invention, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease;

[0088] The target risk level obtaining module 420 is used to input each prediction field into the target prediction model and obtain the target risk level according to the output result of the target prediction model.

[0089] The disease risk prediction device provided in the fourth embodiment of the present invention obtains the prediction fields for predicting the target risk degree of the target disease obtained after processing the target case data through the cooperation between the target prediction model acquisition module and the target risk degree acquisition module, and then inputs each prediction field into the target prediction model, and obtains the target risk degree according to the output result of the target prediction model. The above device achieves the effect of accurately predicting the target risk degree.

[0090] The disease risk prediction device provided in the embodiment of the present invention can execute the disease risk prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0091] It is worth noting that in the embodiment of the above-mentioned disease risk prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0092] Embodiment 5

[0093] Figure 7 A schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention is shown in FIG. Figure 7 As shown, the electronic device includes a memory 510, a processor 520, an input device 530, and an output device 540. The number of processors 520 in the electronic device can be one or more. Figure 7 A processor 520 is taken as an example; the memory 510, the processor 520, the input device 530 and the output device 540 in the electronic device can be connected via a bus or other means. Figure 7 The connection via bus 550 is taken as an example.

[0094] The memory 510, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as the program instructions / modules corresponding to the model training method in the embodiment of the present invention (e.g., the preset prediction model acquisition module 310, the preset risk level sending module 320, the training sample acquisition module 330 and the target prediction model acquisition module 340 in the model training device), or, such as the program instructions / modules corresponding to the disease risk prediction method in the embodiment of the present invention (e.g., the target prediction model acquisition module 410 and the target risk level acquisition module 420 in the disease risk prediction device). The processor 520 executes various functional applications and data processing of the electronic device by running the software programs, instructions and modules stored in the memory 510, that is, realizes the above-mentioned model training method or disease risk prediction method.

[0095] The memory 510 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 510 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 510 may further include a memory remotely arranged relative to the processor 520, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0096] The input device 530 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 540 may include a display device such as a display screen.

[0097] Embodiment 6

[0098] Embodiment 6 of the present invention provides a storage medium containing computer executable instructions, wherein the computer executable instructions are used to execute a model training method when executed by a computer processor, the method comprising:

[0099] Obtaining annotated fields and preset prediction models, wherein the annotated fields are fields obtained after processing sample case data for determining annotated risk levels of target diseases, and the preset prediction model is a pre-trained neural network model for predicting preset risk levels of target diseases;

[0100] Input each annotation field into a preset prediction model to obtain a preset risk level, and send each annotation field and the preset risk level to a disease risk annotation client;

[0101] Receive the annotated risk level returned by the disease risk annotation client, and use each annotated field and the annotated risk level as a set of training samples;

[0102] The original prediction model is trained based on multiple groups of training samples to obtain a target prediction model for predicting the target risk level of the target disease.

[0103] Of course, the storage medium containing computer executable instructions provided in an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in the model training method provided in any embodiment of the present invention.

[0104] Embodiment 7

[0105] Embodiment 7 of the present invention provides a storage medium containing computer executable instructions, wherein the computer executable instructions are used to execute a disease risk prediction method when executed by a computer processor, the method comprising:

[0106] Obtaining a prediction field and a target prediction model generated according to the model generation method provided by any embodiment of the present invention, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease;

[0107] Each prediction field is input into the target prediction model, and the target risk level is obtained according to the output result of the target prediction model.

[0108] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0109] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A model training method, characterized in that: include: Acquire a labeled field and a preset prediction model, wherein the labeled field is a field obtained after processing sample case data for determining a labeled risk level of a target disease, and the preset prediction model is a pre-trained neural network model for predicting a preset risk level of the target disease; Input each of the annotated fields into the preset prediction model to obtain the preset risk level, and send each of the annotated fields and the preset risk level to a disease risk annotation client; Receive the annotated risk level returned by the disease risk annotation client, and use each of the annotated fields and the annotated risk level as a set of training samples; The original prediction model is trained based on the multiple groups of training samples to obtain a target prediction model for predicting the target risk level of the target disease.

2. The method according to claim 1, characterized in that The step of inputting each of the annotation fields into the preset prediction model to obtain the preset risk level, and sending each of the annotation fields and the preset risk level to a disease risk annotation client comprises: For each of the annotated fields of each of the sample case data in each of the sample case data originating from the same object, input each of the annotated fields of each of the sample case data into the preset prediction model to obtain the preset risk level; The preset risk trend of each preset risk degree is determined according to the generation time of each sample case data, and each annotation field, each preset risk degree, and the preset risk trend of each sample case data are sent to the disease risk annotation client.

3. The method according to claim 1, characterized in that Also includes: Acquire a case field, wherein the case field is a field obtained after processing the sample case data and used to filter out the marked field from each of the case fields; Each of the case fields is sent to the disease risk annotation client, and the annotation fields returned by the disease risk annotation client are received.

4. The method according to claim 3, characterized in that Also includes: Acquiring the sample case data; Performing natural language processing on the sample case data to obtain the case field; and / or, Performing desensitization processing on the sample case data and / or the case fields, and updating according to the desensitization processing result; and / or, In the future, the sample case data from the same subject will be normalized to the same subject; and / or, Determine the missing degree and / or logical degree of the sample case data. If the sample case data is used as data to be deleted according to the determination result, send the data to be deleted to the disease risk labeling client, and process the data to be deleted according to the deletion processing result returned by the disease risk labeling client.

5. The method according to claim 1, characterized in that Also includes: Acquire the sample case data and a preset data processing strategy, perform data processing on the sample case data based on the data processing strategy, and update the sample case data according to the data processing result; The data processing strategy includes a field mapping strategy and / or a data format conversion strategy.

6. The method according to claim 1, characterized in that The original prediction model includes a neural network model to be trained that is constructed based on the extreme gradient boosting algorithm.

7. A disease risk prediction method, characterized in that: include: Obtaining a prediction field and a target prediction model generated according to the method of any one of claims 1 to 6, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease; Each of the prediction fields is input into the target prediction model, and the target risk level is obtained according to the output result of the target prediction model.

8. A model training device, characterized in that: include: A preset prediction model acquisition module, used to acquire a labeled field and a preset prediction model, wherein the labeled field is a field obtained after processing the sample case data for determining the labeled risk level of the target disease, and the preset prediction model is a pre-trained neural network model for predicting the preset risk level of the target disease; A preset risk level sending module, used for inputting each of the annotation fields into the preset prediction model to obtain the preset risk level, and sending each of the annotation fields and the preset risk level to a disease risk annotation client; A training sample obtaining module, used for receiving the annotated risk level returned by the disease risk annotation client, and taking each of the annotated fields and the annotated risk level as a set of training samples; The target prediction model obtaining module is used to train the original prediction model based on multiple groups of training samples to obtain a target prediction model for predicting the target risk level of the target disease.

9. A disease risk prediction device, characterized in that: include: A target prediction model acquisition module, used to acquire a prediction field and a target prediction model generated according to the method of any one of claims 1 to 6, wherein the prediction field is a field obtained after processing the target case data and used to predict the target risk level of the target disease; The target risk level obtaining module is used to input each of the prediction fields into the target prediction model and obtain the target risk level according to the output result of the target prediction model.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the model training method as described in any one of claims 1-6, or the disease risk prediction method as described in claim 7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the model training method as described in any one of claims 1 to 6, or the disease risk prediction method as described in claim 7.

Citation Information

Patent Citations

  • Data processing method, apparatus, server and system

    CN106951925A

  • Model training method and device and information prediction method and device

    CN110110811A

  • Disease risk prediction method and device, equipment and computer readable medium

    CN112435747A

  • Classification model training method and device and coronary heart disease auxiliary screening method and device

    CN112562860A

  • Model training method and device based on electronic medical record, equipment and storage medium

    CN112614562A