Method, device, electronic device and medium for building a label prediction model
By establishing a label prediction model for multi-task learning, using the labeling system of diseases, categories, locations and medical records, and combining multi-task learning methods, the problem of data not sharing and low accuracy in prediction results in the existing technology is solved, and the improvement is achieved Accurate label prediction.
Patent Information
- Application Number
- CN202011022926.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-09-25
- Publication Date
- 2024-12-17
- Estimated Expiration
- 2040-09-25
AI Technical Summary
The prior art has problems such as data sharing and low accuracy in the prediction results when building tag prediction models. Especially in multi-task learning scenarios, it is difficult to achieve high-accurate tag prediction after the single-task learning model is summarized.
By establishing a disease and category labeling system, a disease and site labeling system, and a medical record and disease labeling system, combining multi-task learning methods, using the first, second and third samples as training samples, we will build a label prediction for multi-task learning. Model, achieve commonality and correlation between different samples while learning.
The multi-task learning label prediction model built can output more comprehensive and accurate prediction labels, which improves the technical defects of data not sharing and low accuracy of prediction results after the single-task learning model is constructed separately, and improves the accuracy of label prediction. sex.
Smart Images
Figure CN112117009B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence, specifically natural language processing, knowledge graphs and big data, and can be applied to the field of smart medical scenarios, and in particular to methods, devices, electronic devices and computer-readable storage media for building label prediction models. Background Art
[0002] With the advancement of electronic informatization in all walks of life, more and more data are beginning to be aggregated online. Combined with the concept of machine learning, a large amount of user data can be effectively used to build various types of models to achieve multiple purposes including data prediction and content analysis.
[0003] The prior art usually adopts a method of dispersing different purposes into multiple single prediction models, that is, predicting parameters of different purposes respectively through different single prediction models. Summary of the invention
[0004] The embodiments of the present application provide a method, an apparatus, an electronic device, and a computer-readable storage medium for constructing a label prediction model.
[0005] In a first aspect, an embodiment of the present application proposes a method for constructing a label prediction model, comprising: establishing a disease and category label system based on a first sample that records the correspondence between a specific disease and a disease category; establishing a disease and part label system based on a second sample that records the correspondence between a specific disease and a human body part; establishing a medical record and disease label system based on a third sample that records the correspondence between historical medical records and historically diagnosed diseases; and constructing a label prediction model based on the disease and category label system, the disease and part label system, and the medical record and disease label system.
[0006] In the second aspect, an embodiment of the present application proposes a device for constructing a label prediction model, including: a disease and category label system establishment unit, configured to establish a disease and category label system based on a first sample that records the correspondence between a specific disease and a disease category; a disease and part label system establishment unit, configured to establish a disease and part label system based on a second sample that records the correspondence between a specific disease and a human body part; a medical record and disease label system establishment unit, configured to establish a medical record and disease label system based on a third sample that records the correspondence between historical medical records and historically diagnosed diseases; a label prediction model construction unit, configured to construct a label prediction model based on the disease and category label system, the disease and part label system, and the medical record and disease label system.
[0007] In a third aspect, an embodiment of the present application provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that when the at least one processor executes, it is possible to implement the method for constructing a label prediction model as described in any implementation method in the first aspect.
[0008] In a fourth aspect, an embodiment of the present application provides a non-transitory computer-readable storage medium storing computer instructions, which are used to enable a computer to implement the method for constructing a label prediction model as described in any implementation method in the first aspect when executed.
[0009] The method, device, electronic device and computer-readable storage medium for constructing a label prediction model provided in the embodiments of the present application first establish a disease and category label system based on a first sample that records the correspondence between a specific disease and a disease category; then establish a disease and part label system based on a second sample that records the correspondence between a specific disease and a body part; then establish a medical record and disease label system based on a third sample that records the correspondence between historical medical records and historically diagnosed diseases; finally, construct a label prediction model based on the disease and category label system, the disease and part label system and the medical record and disease label system.
[0010] The present application provides a solution for constructing a multi-task learning prediction model through the above-mentioned technical scheme, that is, taking the first sample, the second sample, and the third sample as training samples at the same time, so as to simultaneously learn the potential commonalities and correlations between different samples under different task objectives, so that the constructed multi-task learning label prediction model can output more comprehensive and accurate prediction labels, thereby improving the technical defects of the prior art of constructing multiple single-task learning models separately and then aggregating them, such as non-sharing of data and low accuracy of prediction results.
[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:
[0013] Figure 1 is an exemplary system architecture in which the present application may be applied;
[0014] Figure 2A flowchart of a method for constructing a label prediction model provided in an embodiment of the present application;
[0015] Figure 3 A flowchart of another method for constructing a tag prediction model provided in an embodiment of the present application;
[0016] Figure 4 A flowchart of a method for constructing a label prediction model in an application scenario provided by an embodiment of the present application;
[0017] Figure 5 A structural block diagram of a device for building a label prediction model provided in an embodiment of the present application;
[0018] Figure 6 A schematic diagram of the structure of an electronic device suitable for executing a method for building a tag prediction model provided in an embodiment of the present application. DETAILED DESCRIPTION
[0019] The present application will be 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 relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.
[0020] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0021] Figure 1 An exemplary system architecture 100 is shown to which embodiments of the method, apparatus, electronic device, and computer-readable storage medium for building a tag prediction model of the present application can be applied.
[0022] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0023] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various applications for realizing information communication between the terminal devices 101, 102, 103 and server 105 can be installed, such as data transmission applications, prediction model building applications, instant messaging applications, etc.
[0024] Terminal devices 101, 102, 103 and server 105 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, etc.; when terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above, which can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server; when the server is software, it can be implemented as multiple software or software modules, or as a single software or software module, and no specific limitation is made here.
[0025] The server 105 can provide various services through various built-in applications. Taking the prediction model building application that can provide the service of establishing a label prediction model as an example, the server 105 can achieve the following effects when running this type of application: first, obtain a first sample recording the correspondence between a specific disease and a disease category, a second sample recording the correspondence between a specific disease and a human body part, and a third sample recording the correspondence between historical medical records and historical diagnosed diseases from terminal devices 101, 102, and 103 through network 104; then, according to the first sample, the second sample, and the third sample, respectively establish a disease and category label system, a disease and part label system, and a medical record and disease label system; finally, according to the disease and category label system, the disease and part label system, and the medical record and disease label system, build a label prediction model. Furthermore, after the server 105 builds a usable label prediction model, it can subsequently provide label prediction services based on the label prediction model to the outside world, such as receiving actual electronic medical records sent by users from terminal devices 101, 102, and 103 through network 104, inputting the actual electronic medical records into the label prediction model, and returning the output actual disease category labels and actual human body part labels to the user.
[0026] It should be noted that the first sample recording the correspondence between specific diseases and disease categories, the second sample recording the correspondence between specific diseases and human body parts, and the third sample recording the correspondence between historical medical records and historically diagnosed diseases can be obtained from the terminal devices 101, 102, and 103 through the network 104, or can be pre-stored locally in the server 105 in various ways. Therefore, when the server 105 detects that these data have been stored locally (for example, a backup is retained in the local storage unit of the server), it can choose to directly obtain these data locally. In this case, the exemplary system architecture 100 may also not include the terminal devices 101, 102, 103 and the network 104.
[0027] Since learning and training a label prediction model that can output multiple category labels requires the use of a large number of corresponding samples for training, it needs to occupy more computing resources of the device and have stronger computing power. Therefore, the method for constructing a label prediction model provided in the subsequent embodiments of this application is generally performed by a server 105 with stronger computing power and more computing resources. Correspondingly, the device for constructing the label prediction model is generally also set in the server 105.
[0028] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.
[0029] Please refer to Figure 2 , Figure 2 A flowchart of a method for constructing a tag prediction model provided in an embodiment of the present application, wherein process 200 includes the following steps:
[0030] Step 201: establishing a disease and category label system based on the first sample recording the corresponding relationship between the specific disease and the disease category;
[0031] This step is intended to be performed by the execution subject of the method for building a label prediction model (e.g. Figure 1 The server 105 shown establishes a disease and category label system based on the first sample that records the corresponding relationship between the specific disease and the disease category.
[0032] Among them, the first sample that records the correspondence between specific diseases and disease categories can be obtained through a variety of channels, such as extracted from medical books, obtained from data annotations of experienced physicians, and obtained from knowledge graphs related to medical diseases published by authoritative institutions, etc. The appropriate acquisition method can be selected according to the actual situation, and no specific limitation is made here. Furthermore, according to the different sources of the first sample, the above-mentioned execution subject should also select the corresponding acquisition method. For example, for public resources on the Internet, it can be directly obtained from the Internet through web crawlers or downloads, and for resources stored offline, it can be obtained in a legal manner by applying for authorization.
[0033] It should be understood that a specific disease may belong to different disease categories at the same time, and each disease category may contain multiple different specific diseases. Therefore, the correspondence between the specific disease and the disease category described in this step is actually a complex correspondence including one-to-one, one-to-many, many-to-one, and many-to-many. Therefore, when choosing a recording method, you should try to choose a recording method that can record many-to-many correspondences in a concise manner.
[0034] The following is a data that records the corresponding relationship between specific diseases and disease categories in a table format and can be used to extract the first sample. Please see Table 1 below:
[0035] Table 1 Correspondence between specific diseases and disease categories
[0036]
[0037] Table 1 above records the inclusion relationship between some disease categories and some of the specific diseases included in tabular form. Taking the disease category of "immune disease" as an example, Table 1 specifically includes four specific diseases: "myositis", "lupus erythematosus", "antiphospholipid syndrome", and "immunodeficiency disease", which is a one-to-many correspondence. The constructed disease and category label system can be a label representation of the above content, thereby establishing a system using the correspondence between category labels and specific disease labels.
[0038] Specifically, "immune disease" can be directly used as the category label of the disease in this category, and "myositis" can be used as the label of a specific disease. For simplicity, the above labels can be represented as short and unique specific strings or numbers, which can achieve the same effect and increase the storage and retrieval performance of the corresponding relationship. In addition, the reason why it is built as a label system is to improve efficiency by taking advantage of the characteristics of the label, which is an encapsulated content representation form, to facilitate calling and post-processing.
[0039] Step 202: establishing a disease and body part labeling system according to the second sample recording the corresponding relationship between the specific disease and the body part;
[0040] This step is intended for the execution subject to establish a disease and category label system based on the first sample that records the correspondence between the specific disease and the disease category. Except for the difference in the correspondence recorded in the acquired sample, the rest is described in step 201.
[0041] The following is a data in a table format that records the corresponding relationship between specific diseases and human body parts and can be used to extract the second sample. Please see Table 2 below:
[0042] Table 2 Correspondence between specific diseases and human body parts
[0043] Human body parts Specific diseases heart Neonatal congenital heart disease, heart damage, pericardial effusion… Duodenum Duodenal mass, duodenal intussusception, duodenal Crohn's disease... Ear Congenital auricular malformation, otogenic brain abscess, auricular pseudocyst... Hands Tinea manuum, tinea pedis, finger sprains and strains… Stomach Gastric carcinoid, gastric tumor, gastric antrum and gastric body malignant tumor… Brain Cerebellar cyst, brainstem infarction, brainstem glioma, cerebellar infarction… anus Anal polyps, anogenital warts, anal cancer, anal fissures… throat Herpetic pharyngitis, laryngeal squamous cell carcinoma, pharyngeal malformation…
[0044] Table 2 above records the inclusion relationship between some human body parts and some of the specific diseases included in the form of a table. Taking the "ear" human body part as an example, Table 2 specifically includes four specific diseases: "tinea manuum", "tinea pedis", "finger sprain", and "strain", which is a one-to-many correspondence. The constructed disease and part label system can be a label representation of the above content, so as to establish a system based on the correspondence between the part label and the specific disease label.
[0045] Furthermore, the human body parts can be further subdivided into multiple levels. For example, the human body parts can be divided into skin, fascia, muscle, blood vessel, lymph, nerve, bone, and joint. They can also be divided according to the two-level part system shown in Table 3 below:
[0046] Table 3 Secondary classification system of human body parts
[0047]
[0048]
[0049] It can be seen from this that the correspondence between the parts and specific diseases shown in Table 2 is only the result of one of the parts division methods. Different forms of division of human body parts can be obtained according to actual application scenarios, and no specific limitation is made here.
[0050] Step 203: Establishing a medical record and disease label system according to the third sample recording the correspondence between historical medical records and historical diagnosed diseases;
[0051] Different from the two labeling systems established in step 201 and step 202, this step is intended to enable the above-mentioned executor to establish a medical record and disease labeling system based on a third sample that records the correspondence between historical medical records and historical diagnosed diseases, so as to use a large amount of sample data to enable the machine and model to find the relationship between the medical record content described in natural language and the actual disease diagnosis results.
[0052] Specifically, whether it is a paper medical record or an electronic medical record, the content writer usually expresses it based on his or her own understanding of his or her own physical condition. Considering that the cognition of different users is different, in order to fully explore the potential correlation between historical medical records and diagnosed diseases as much as possible, and find commonalities from special cases based on this potential correlation, the medical record content can also be processed in a variety of ways, including but not limited to medical entity extraction, symptom element extraction, text error correction, confidence determination, etc., and because text is not convenient for feature calculations, you can also choose a suitable method to convert the text into other forms of expression to participate in feature matching and other operations, such as vectors, high-dimensional features, etc.
[0053] It should be noted that there is no causal relationship between step 201, step 202 and step 203 in this embodiment, and they can be executed separately or in parallel. This embodiment only provides a seemingly serial execution method for the convenience of process flow.
[0054] Step 204: Construct a label prediction model based on the disease and category label system, the disease and part label system, and the medical record and disease label system.
[0055] Based on step 201, step 202 and step 203, this step aims to construct a label prediction model according to the disease and category label system, the disease and part label system and the medical record and disease label system by the above-mentioned execution subject.
[0056] That is, this step uses the correspondence between specific diseases and disease categories, the correspondence between specific diseases and human body parts, and the correspondence between historical medical records and historically diagnosed diseases as training data to construct a multi-task learning model, namely the label prediction model. The reason why it is called a label prediction model is that it is a prediction model built based on the label system and the output result is a label of the corresponding content.
[0057] The method for constructing a label prediction model provided in the embodiment of the present application actually provides a solution for constructing a multi-task learning prediction model, that is, the first sample, the second sample, and the third sample are used as training samples at the same time, so that the potential commonalities and correlations between different samples can be learned simultaneously under different task objectives. The multi-task learning label prediction model constructed in this way can be more comprehensive and accurate in predicting labels based on actual medical records output, and can improve as much as possible the technical defects of the prior art of separately constructing multiple single-task learning models and then aggregating them, such as the non-sharing of data and the low accuracy of prediction results.
[0058] Please refer to Figure 3 , Figure 3 A flowchart of another method for constructing a tag prediction model provided in an embodiment of the present application, wherein process 300 includes the following steps:
[0059] Step 301: Establishing a disease and category label system based on the first sample recording the corresponding relationship between the specific disease and the disease category;
[0060] Step 302: establishing a disease and body part labeling system according to the second sample recording the corresponding relationship between the specific disease and the body part;
[0061] Step 303: Establishing a medical record and disease label system according to the third sample recording the correspondence between historical medical records and historical diagnosed diseases;
[0062] In some other embodiments of the present application, a specific method for establishing a medical record and disease label system is also provided, which mainly focuses on how to extract and process information from historical medical records, and specifically includes the following steps:
[0063] 1) Extract natural language text from historical medical records and identify suspected diseases based on historically diagnosed diseases;
[0064] 2) Perform feature dimensionality reduction processing on natural language text to obtain dimensionality-reduced text features;
[0065] The purpose of dimensionality reduction is to convert the original high-dimensional features into low-dimensional features, thereby reducing the difficulty of mining the feature content and exposing some complex features hidden at the bottom. Specifically, the dimensionality reduction process can be implemented using an Embedding layer.
[0066] 3) Using the preset neural network to process the reduced-dimensional text features to obtain processed medical record features;
[0067] For example, the reduced-dimensional text features can be processed by convolutional neural networks (CNNs). CNN is a type of feedforward neural network with deep structure and convolutional calculations, and is one of the representative algorithms of deep learning. CNN is usually composed of convolutional layers, pooling layers, and fully connected layers. Convolutional neural networks have representation learning capabilities and can perform translation-invariant classification of input information according to their hierarchical structure. This step aims to mine more features that can be used to find commonalities with suspected diseases through the convolution and pooling processing of data by CNN.
[0068] 4) Based on the labeled representation of suspected diseases corresponding to the processed medical record characteristics, a medical record and disease labeling system is established.
[0069] Tagging can be understood as a way of encapsulating content so that it can be called and matched in the form of tags.
[0070] The above-mentioned data processing method of successively performing dimensionality reduction processing and neural network processing on the natural language text contained in the medical record content can mine the medical record content as deeply as possible, so as to find the correlation between the medical record content and the suspected disease as comprehensively and accurately as possible.
[0071] Step 304: construct a label prediction model based on the disease and category label system, the disease and part label system, and the medical record and disease label system;
[0072] The above steps 301-304 are similar to Figure 2Steps 201-204 shown are consistent, and for the same contents, please refer to the corresponding parts of the previous embodiment, which will not be repeated here.
[0073] The above parts all belong to how to construct the label prediction model. The subsequent steps will describe how to actually use the constructed label prediction model. It should be understood that the usage method given in the subsequent steps does not require repeating the above label prediction model construction steps before each use, unless a new application scenario is replaced that is not suitable for the previously constructed label prediction model.
[0074] Step 305: receiving the incoming actual electronic medical record;
[0075] Step 306: using the label prediction model to determine the actual disease category label and the actual human body part label corresponding to the actual electronic medical record;
[0076] This step is intended for the above-mentioned execution subject to use the pre-built label prediction model to determine the actual disease category label and actual human body part label corresponding to the actual electronic medical record. In terms of specific implementation, the actual electronic medical record can be imported into the label prediction model as input data, and the actual disease category label and actual human body part label output by the label prediction model are received.
[0077] It should be understood that the reason why the label prediction model can output the actual disease category label and the actual human body part label is that in the process of building the label prediction model, the ability to determine the suspected disease corresponding to the medical record content is learned from the medical record and diagnosis disease label system. On the basis of knowing the suspected disease, the corresponding actual disease category is determined with the help of the affiliation between the specific disease and the disease category learned in the disease and category label system, and the corresponding human body part is determined with the help of the correspondence between the specific disease and the human body part learned in the disease and part label system. Furthermore, the determined labels representing the corresponding disease category and human body part labels can also be used to reversely prove the accuracy of the suspected disease.
[0078] Step 307: Determine the corresponding user attribute label according to the natural language text of the actual electronic medical record;
[0079] This step aims to enable the above-mentioned execution subject to determine the corresponding user attribute labels based on the natural language text of the actual electronic medical record, wherein the user attribute labels include gender labels, family genetic disease labels, work environment labels, etc., which are used to characterize the user's personal attributes. These attribute labels will be used as reference influencing factors to participate in other data processing, such as data statistics, whether there is a family disease, whether it is hereditary, etc.
[0080] Step 308: Return the actual disease category label, the actual human body part label, and the user attribute label as an auxiliary label according to the preset path.
[0081] Based on step 306 and step 307, this step aims to return the actual disease category label, the actual human body part label, and the user attribute label as an auxiliary label according to the preset path. The preset path includes but is not limited to emails, text messages, interface pop-ups, instant messaging applications, etc.
[0082] On the basis of the previous embodiment, this embodiment not only provides a specific implementation method for establishing a medical record and diagnosis disease label system, but also, by performing various processing on the medical record content, is able to more comprehensively find the correlation between the medical record and the suspected disease; at the same time, through steps 305-308, it also provides a method for using the constructed label prediction model to output the actual label corresponding to the actual electronic medical record.
[0083] It should be noted that the technical solution of outputting the actual disease category label and the actual human body part label through the label prediction model given in step 306, and the technical solution of analyzing the user attribute label extracted from the natural language text from the medical record content given in step 307, can form two different label return solutions separately, that is, two analyses are performed at the same time and two types of labels are returned at the same time as in this embodiment, or in another embodiment, only the technical solution of step 306 is included, and only the actual disease category label and the actual human body part label output by the label prediction model are subsequently returned. The purpose of adding step 307 is to be able to perform some other statistics more conveniently and comprehensively based on the attached labels.
[0084] In some other embodiments of the present application, a statistical method for tags including but not limited to the following may be:
[0085] Count the number of abnormal labels when the actual output quantity exceeds the preset quantity within the preset time period;
[0086] Determine whether there is a short-term epidemic outbreak based on the number of abnormal labels.
[0087] For example, in one month, more than 80% of the labels output by electronic medical records contain the disease category label of acute upper respiratory tract infection, far exceeding the average of 40%. To a certain extent, it can be considered that there is an acute influenza outbreak. Furthermore, targeted errors can be made for such situations, such as pushing preventive measures.
[0088] It should be understood that the output labels can be used for a variety of purposes, such as estimating the number of patients and their conditions in a certain area through statistics, arranging drug stores or drug preparation institutions to prepare drugs in a coordinated manner, reducing waste, etc.
[0089] On the basis of any of the above embodiments, in order to ensure the long-term availability of the label prediction model, the user can also be allowed to make corrections and error identifications to the output prediction labels based on the actual situation, and timely adjust the parameters of the model based on the correction information and error identification, so that the continuously corrected model has better prediction accuracy. A specific implementation method can be: receiving label error information input for labels with incorrect predictions, and adjusting the parameters of the label prediction model based on the label error information.
[0090] To deepen understanding, this application also provides a specific implementation solution in combination with a specific application scenario, please refer to Figure 4 The data processing flow diagram for achieving label prediction effect is shown:
[0091] First of all Figure 4 The meaning of the English words in the text are explained as follows:
[0092] NLU is the abbreviation of Natural Language Understanding, which is translated into Chinese as natural language understanding. It aims to explain the corresponding processing of the content of the input medical record; text here refers to the natural text of the medical record obtained after natural language processing, such as coughing for 3 days and dizziness for 2 days; feat refers to the medical record features extracted from the results of natural language processing (such as symptoms, signs, diseases, etc.), corresponding to the above example, there are two feats "cough" and "dizziness"; feat-part refers to the part information to which the feat belongs, such as the part information of dizziness is the head; sex refers to the patient's gender information; popul refers to the patient's population attribute information, such as adults, children, and pregnant women; emb refers to text distributed semantic encoding Embedding processing, which is a processing method for reducing the dimensionality of features; avg pooling value average pooling processing; CNN refers to convolutional neural network, Figure 4 The left part in the middle shows a typical CNN network hierarchy, where filter refers to the filter, Conv refers to the convolutional layer, relu is a loss function, max-pooling refers to the maximum pooling operation; fc refers to the fully connected layer; one-hot refers to the one-hot encoding of text features, such as the one-hot encoding of gender: the one-hot encoding of male is (1, 0), and the one-hot encoding of female is (0, 1); "+" refers to the concatenation of multiple features expressed in vector form; Sigmoid refers to the loss function used by the model.
[0093] The overall usage process is:
[0094] 1) The server receives the incoming medical record natural text;
[0095] 2) The server processes the medical record natural text through natural language processing technology, and parses the medical record natural text (text), medical record features (feat), part information of medical record features (feat-part), gender of medical record patients, and population attributes of medical record patients;
[0096] 3) The server uses CNN to encode the natural text of the medical record (text), uses emb to encode the medical record features (feat) and the part information of the medical record features (feat-part), and uses one-hot to encode the patient's gender (sex) and the patient's population attribute information (popul);
[0097] 4) The server concatenates (+) various medical record information in vector form;
[0098] 5) The server implements label prediction based on the pre-built structured label system, and outputs the predicted first-level part labels, second-level part labels, whole-body part labels and disease category labels.
[0099] In the overall process, the Sigmoid loss function is used to learn the network parameters.
[0100] Further references Figure 5 As an implementation of the methods shown in the above figures, the present application provides an embodiment of a device for constructing a label prediction model. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.
[0101] like Figure 5 As shown, the apparatus 500 for constructing a label prediction model in this embodiment may include: a disease and category label system establishment unit 501, a disease and part label system establishment unit 502, a medical record and disease label system establishment unit 503, and a label prediction model construction unit 504. Among them, the disease and category label system establishment unit 501 is configured to establish a disease and category label system according to a first sample recording the corresponding relationship between a specific disease and a disease category; the disease and part label system establishment unit 502 is configured to establish a disease and part label system according to a second sample recording the corresponding relationship between a specific disease and a human body part; the medical record and disease label system establishment unit 503 is configured to establish a medical record and disease label system according to a third sample recording the corresponding relationship between a historical medical record and a historically diagnosed disease; the label prediction model construction unit 504 is configured to construct a label prediction model according to the disease and category label system, the disease and part label system, and the medical record and disease label system.
[0102] In the present embodiment, in the apparatus 500 for constructing a label prediction model, the specific processing and technical effects of the disease and category label system establishment unit 501, the disease and part label system establishment unit 502, the medical record and disease label system establishment unit 503, and the label prediction model construction unit 504 can be referred to respectively. Figure 2 The relevant descriptions of steps 201 - 204 in the corresponding embodiment are not repeated here.
[0103] In some optional implementations of this embodiment, the medical record and disease label system establishing unit 503 may be further configured to:
[0104] Extract natural language text from historical medical record information and identify suspected diseases based on historically diagnosed diseases;
[0105] Perform feature dimensionality reduction processing on natural language text to obtain dimensionality-reduced text features;
[0106] Use the preset neural network to process the reduced-dimensional text features to obtain processed medical record features;
[0107] Based on the labeled representation of suspected diseases corresponding to the processed medical record characteristics, a medical record and disease labeling system is established.
[0108] In some optional implementations of this embodiment, the apparatus 500 for constructing a label prediction model may further include:
[0109] an actual electronic medical record receiving unit configured to receive an incoming actual electronic medical record;
[0110] an actual label determination unit, configured to determine an actual disease category label and an actual human body part label corresponding to an actual electronic medical record using a label prediction model;
[0111] The actual label returning unit is configured to return the actual disease category label and the actual human body part label according to a preset path.
[0112] In some optional implementations of this embodiment, the apparatus 500 for constructing a label prediction model may further include:
[0113] A user attribute label determination unit is configured to determine a corresponding user attribute label according to the natural language text of the actual electronic medical record; wherein the user attribute label includes a gender label, a family genetic disease label, and a work environment label;
[0114] The user attribute tag returning unit is configured to return the user attribute tag as an attached reference tag.
[0115] In some optional implementations of this embodiment, the apparatus 500 for constructing a label prediction model may further include:
[0116] The abnormal quantity label counting unit is configured to count abnormal quantity labels whose actual output quantity exceeds a preset quantity within a preset time period;
[0117] The short-term epidemic outbreak determination unit is configured to determine whether there is an outbreak of a short-term epidemic according to the quantity anomaly label.
[0118] In some optional implementations of this embodiment, the apparatus 500 for constructing a label prediction model may further include:
[0119] A label error information receiving unit, configured to receive label error information input for a predicted incorrect label;
[0120] The model parameter adjustment unit is configured to adjust the parameters of the label prediction model according to the label error information.
[0121] This embodiment exists as an apparatus embodiment corresponding to the above-mentioned method embodiment. The apparatus for constructing a label prediction model provided by this embodiment provides a scheme for constructing a multi-task learning prediction model, that is, the first sample, the second sample, and the third sample are used as training samples at the same time, so that the potential commonalities and correlations between different samples can be learned simultaneously under different task objectives, so that the constructed multi-task learning label prediction model outputs more comprehensive and accurate prediction labels, thereby improving the technical defects of the prior art of constructing multiple single-task learning models separately and then aggregating them, such as non-sharing of data and low accuracy of prediction results.
[0122] According to an embodiment of the present application, the present application also provides an electronic device and a computer-readable storage medium.
[0123] Figure 6 A block diagram of an electronic device suitable for implementing a method for building a tag prediction model in an embodiment of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or required herein.
[0124] like Figure 6As shown, the electronic device includes: one or more processors 601, a memory 602, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected using different buses and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the electronic device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In other embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 6 A processor 601 is taken as an example.
[0125] The memory 602 is a non-transitory computer-readable storage medium provided in the present application. The memory stores instructions executable by at least one processor to enable at least one processor to execute the method for building a label prediction model provided in the present application. The non-transitory computer-readable storage medium of the present application stores computer instructions, which are used to enable a computer to execute the method for building a label prediction model provided in the present application.
[0126] The memory 602 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as the program instructions / modules corresponding to the method for building a tag prediction model in the embodiment of the present application (for example, the attached Figure 5 The processor 601 executes various functional applications and data processing of the server by running the non-transient software programs, instructions and modules stored in the memory 602, that is, the method for constructing a label prediction model in the above method embodiment is implemented.
[0127] The memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store various types of data created by the electronic device when executing the method for building a tag prediction model. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 602 may optionally include a memory remotely disposed relative to the processor 601, and these remote memories may be connected to an electronic device suitable for executing the method for building a tag prediction model 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.
[0128] The electronic device suitable for executing the method for building a tag prediction model may also include: an input device 603 and an output device 604. The processor 601, the memory 602, the input device 603 and the output device 604 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.
[0129] The input device 603 can receive input digital or character information, and generate key signal input related to user settings and function control of the electronic device suitable for executing the method for building a tag prediction model, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator rod, one or more mouse buttons, a trackball, a joystick and other input devices. The output device 604 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display and a plasma display. In some embodiments, the display device may be a touch screen.
[0130] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0131] These computer programs (also referred to as programs, software, software applications, or code) include machine instructions for programmable processors and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or means (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.
[0132] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).
[0133] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.
[0134] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.
[0135] The technical solution of the embodiment of the present application provides a solution for constructing a multi-task learning prediction model, that is, taking the first sample, the second sample, and the third sample as training samples at the same time, so as to simultaneously learn the potential commonalities and correlations between different samples under different task objectives, so that the constructed multi-task learning label prediction model outputs more comprehensive and accurate prediction labels, thereby improving the technical defects of the prior art of constructing multiple single-task learning models separately and then aggregating them, such as non-sharing of data and low accuracy of prediction results.
[0136] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution disclosed in this application can be achieved, and this document is not limited here.
[0137] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.
Claims
1. A method for constructing a label prediction model, comprising: Establishing a disease and category labeling system based on the first sample that records the corresponding relationship between specific diseases and disease categories; Establishing a disease and body part labeling system based on a second sample recording the corresponding relationship between specific diseases and body parts; Establish a medical record and disease labeling system based on the third sample that records the correspondence between historical medical records and historically diagnosed diseases; A label prediction model is constructed based on the disease and category label system, the disease and part label system, and the medical record and disease label system, including: using the first sample, the second sample, and the third sample as training samples at the same time to construct a multi-task learning model, wherein the multi-task learning model can simultaneously learn the potential commonalities and associations between different samples under different task objectives, wherein the multi-task learning model is the label prediction model, and the label prediction model is used to determine the actual disease category label and the actual human body part label corresponding to the actual electronic medical record input.
2. The method according to claim 1, wherein: The method of establishing a medical record and disease labeling system based on the third sample recording the correspondence between historical medical records and historically diagnosed diseases includes: extracting natural language text from the historical medical record information, and determining suspected diseases based on the historically diagnosed diseases; Performing feature dimensionality reduction processing on the natural language text to obtain dimensionality-reduced text features; Using a preset neural network to process the reduced-dimensional text features to obtain processed medical record features; According to the labeled representation of the suspected disease corresponding to the processed medical record features, the medical record and disease labeling system is established.
3. The method according to claim 1, further comprising: receiving the actual electronic medical record; Determine the actual disease category label and the actual human body part label corresponding to the actual electronic medical record using the label prediction model; The actual disease category label and the actual human body part label are returned according to a preset path.
4. The method according to claim 3, further comprising: Determine the corresponding user attribute tag according to the natural language text of the actual electronic medical record; wherein the user attribute tag includes a gender tag, a family genetic disease tag, and a work environment tag; The user attribute tag is returned as an attached reference tag.
5. The method according to claim 3 or 4, further comprising: Count the number of abnormal labels when the actual output quantity exceeds the preset quantity within the preset time period; Determine whether there is a short-term epidemic outbreak based on the abnormal quantity label.
6. The method according to claim 1, further comprising: Receive label error information for the wrongly predicted label; The parameters of the label prediction model are adjusted according to the label error information.
7. A device for constructing a label prediction model, comprising: A disease and category label system establishing unit is configured to establish a disease and category label system according to a first sample recording a correspondence between a specific disease and a disease category; a disease and part label system establishing unit, configured to establish a disease and part label system according to a second sample recording a corresponding relationship between a specific disease and a human body part; A medical record and disease label system establishment unit is configured to establish a medical record and disease label system according to a third sample recording a correspondence between historical medical records and historically diagnosed diseases; The label prediction model construction unit is configured to construct a label prediction model according to the disease and category label system, the disease and part label system and the medical record and disease label system, including: using the first sample, the second sample and the third sample as training samples at the same time to construct a multi-task learning model, wherein the multi-task learning model can simultaneously learn the potential commonalities and associations between different samples under different task objectives, wherein the multi-task learning model is the label prediction model, and the label prediction model is used to determine the actual disease category label and the actual human body part label corresponding to the actual electronic medical record input.
8. The device according to claim 7, wherein: The medical record and disease label system establishment unit is further configured to: extracting natural language text from the historical medical record information, and determining suspected diseases based on the historically diagnosed diseases; Performing feature dimensionality reduction processing on the natural language text to obtain dimensionality-reduced text features; Using a preset neural network to process the reduced-dimensional text features to obtain processed medical record features; According to the labeled representation of the suspected disease corresponding to the processed medical record features, the medical record and disease labeling system is established.
9. The apparatus according to claim 7, further comprising: an actual electronic medical record receiving unit, configured to receive the actual electronic medical record; an actual label determination unit, configured to determine an actual disease category label and an actual human body part label corresponding to the actual electronic medical record using the label prediction model; The actual label returning unit is configured to return the actual disease category label and the actual human body part label according to a preset path.
10. The apparatus according to claim 9, further comprising: A user attribute label determination unit is configured to determine a corresponding user attribute label according to the natural language text of the actual electronic medical record; wherein the user attribute label includes a gender label, a family genetic disease label, and a work environment label; The user attribute tag returning unit is configured to return the user attribute tag as an attached reference tag.
11. The device according to claim 8 or 9, further comprising: The abnormal quantity label counting unit is configured to count abnormal quantity labels whose actual output quantity exceeds a preset quantity within a preset time period; The short-term epidemic outbreak determination unit is configured to determine whether there is an outbreak of a short-term epidemic according to the abnormal quantity label.
12. The apparatus according to claim 7, further comprising: A label error information receiving unit, configured to receive label error information input for a predicted incorrect label; A model parameter adjustment unit is configured to adjust the parameters of the label prediction model according to the label error information.
13. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for constructing a label prediction model described in any one of claims 1-6.
14. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method for constructing a label prediction model according to any one of claims 1 to 6.