Health assessment model training method, health assessment method and related device

Abstract

The embodiments of the present application provide a training method for a health assessment model, a health assessment method and related devices, which belong to the field of artificial intelligence technology; first training data is constructed by matching case data with health status information at the disease level; the original health assessment model is trained with the first training data; second training data is constructed by adding additional status information to the case data; the original correlation model is trained with the second training data; alignment training is performed based on the first health assessment model and the first correlation model to obtain a trained health assessment model; the case data is matched with the disease information, additional status information and coding information of the health status information in the disease system, which is conducive to improving the accuracy of the health assessment; there is no need to construct an additional intermediate mapping conversion process, and the two tasks of ICD coding and GBD health matching are completed at the same time, thereby improving the efficiency of the model processing task.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a training method for a health assessment model, a health assessment method, and related devices. Background Art

[0002] The disease system defined by the Global Burden of Disease study (GBD) uses unified methods and standards to assess the impact of various diseases and injuries on human health, including death, disability and loss of quality of life. It can provide valuable information and guidance for medical program developers and health professionals to improve global human health and prevent the occurrence of diseases.

[0003] However, the diseases defined in the current GBD disease system, the health consequences caused by diseases, the impact on life expectancy, and other information cannot be directly and completely matched with standardized disease definitions. This makes it impossible to directly link the GBD disease system with patient disease records, reducing the efficiency and effectiveness of health assessments. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a training method for a health assessment model, a health assessment method and related devices, which are conducive to improving the accuracy of health assessment.

[0005] To achieve the above objectives, a first aspect of an embodiment of the present application proposes a training method for a health assessment model, comprising:

[0006] Obtain case data;

[0007] Constructing first training data based on matching the case data with health status information in the disease system at the disease level;

[0008] Training an original health assessment model for generating health status according to cases using the first training data to obtain a first health assessment model;

[0009] Adding additional status information to the case data in the first training data to construct second training data;

[0010] Training an original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model;

[0011] Alignment training is performed based on the first health assessment model and the first correlation model, and parameters of the first health assessment model are adjusted to obtain a trained health assessment model.

[0012] In some embodiments, the health status information includes disease information and corresponding first coded information, and the case data includes second coded information; and constructing first training data based on matching the case data with the health status information in the disease system at the disease level includes:

[0013] Searching and obtaining first target disease information from the disease information, wherein first coding information of the first target disease information is similar to second coding information of the case data;

[0014] According to a first matching degree between the first target disease information and the case data, second target disease information is matched from the first target disease information, and first training data is constructed based on the second target disease information and the case data.

[0015] In some embodiments, the first matching degree includes the semantic similarity between the first target disease information and the case data and the number of first target disease information corresponding to each case data.

[0016] In some embodiments, the health status information includes disease information and corresponding additional status information; and adding the additional status information to the case data in the first training data to construct the second training data includes:

[0017] selecting a target prompt word template from preset prompt word templates according to the type of the health status information;

[0018] Filling the disease information and the additional status information into the target prompt word template as prompt words to obtain a target instruction;

[0019] The case data in the first training data is expanded according to the target template, additional status information is added to the case data in the first training data to obtain extended case data, and the extended case data is divided into positive samples and negative samples according to the correlation between the extended case data and the health status information, and the second training data is constructed by the positive samples and the negative samples.

[0020] In some embodiments, the training of the original correlation model for predicting the correlation between the case and the health status by the second training data to obtain the first correlation model includes:

[0021] Obtaining a relevance score based on the extended case data and the health status information;

[0022] Regression training is performed on the original correlation model according to the correlation score to obtain a first correlation model.

[0023] In some embodiments, the correlation score between the extended case data and the health status information is expressed as: In the formula, Score new Score represents the correlation score between the extended case data and the health status information. old represents the correlation score between the case data in the first training data and the health status information, the direction of ± depends on the correlation between the extended case data and the health status information, ε represents a preset floating score value, N is the number of extended case data, GBD i Represents the sample of the i-th health status information, EHR j represents the sample of the j-th extended case data, and BertScore represents the semantic similarity.

[0024] To achieve the above objectives, a second aspect of the embodiments of the present application provides a health assessment method, including:

[0025] Obtain data on cases to be evaluated;

[0026] Inputting the case data to be evaluated into a trained health assessment model to perform health assessment and obtain an assessment result, wherein the assessment result corresponds to the disease information and additional status information of the health status information of the disease system;

[0027] The trained health assessment model is obtained by training according to the health assessment model training method described in the first aspect of the embodiment of the present application.

[0028] To achieve the above objectives, a third aspect of an embodiment of the present application provides a training device for a health assessment model, comprising:

[0029] Input module, used to obtain case data;

[0030] A first data construction module is used to match the case data with the health status information in the disease system at the disease level to construct first training data;

[0031] A first model training module is configured to train an original health assessment model for generating health status according to a case using the first training data to obtain a first health assessment model;

[0032] A second data construction module is used to add additional status information to the case data in the first training data to construct second training data;

[0033] A second model training module is used to train an original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model;

[0034] The third model training module is used to perform alignment training based on the first health assessment model and the first correlation model, and adjust the parameters of the first health assessment model to obtain a trained health assessment model.

[0035] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes an electronic device, characterized in that the electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the training method of the health assessment model described in the first aspect of the embodiment of the present application and the health assessment method described in the second aspect of the embodiment of the present application.

[0036] To achieve the above-mentioned purpose, the fifth aspect of the embodiment of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the training method of the health assessment model described in the first aspect of the embodiment of the present application and the health assessment method described in the second aspect of the embodiment of the present application.

[0037] The training method, health assessment method and related device of the health assessment model proposed in this application construct first training data by matching case data with health status information in the disease system at the disease level; train the original health assessment model used to generate health status based on the case with the first training data to obtain a first health assessment model; add additional status information to the case data in the first training data to construct second training data; train the original correlation model used to predict the correlation between the case and the health status with the second training data to obtain a first correlation model; align training is performed based on the first health assessment model and the first correlation model, and the parameters of the first health assessment model are adjusted to obtain a trained health assessment model; match the disease information, additional status information and ICD-10 code of the case data with the health status information in the disease system, so that the health assessment model can not only accurately assess the disease information based on the case data, but also accurately assess the additional status information based on the case data, which is beneficial to provide valuable information and guidance to doctors and is beneficial to subsequent further health assessment; and there is no need to construct an additional intermediate mapping conversion process, and it can be directly applied to clinical diagnosis text records, and simultaneously complete the two tasks of health matching of ICD codes and GBD, thereby improving the efficiency of model processing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a step diagram of a training method for a health assessment model provided in an embodiment of the present application;

[0039] Figure 2 It is a sub-step diagram of step S200 provided in an embodiment of the present application;

[0040] Figure 3 It is a sub-step diagram of step S400 provided in an embodiment of the present application;

[0041] Figure 4 It is a sub-step diagram of step S500 provided in an embodiment of the present application;

[0042] Figure 5 is a step diagram of the health assessment method provided in an embodiment of the present application;

[0043] Figure 6 is a structural diagram of a training device for a health assessment model provided in an embodiment of the present application;

[0044] Figure 7 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0046] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0048] The disease system defined by the Global Burden of Disease study (GBD) uses unified methods and standards to assess the impact of various diseases and injuries on human health, including death, disability, and loss of quality of life. This can provide valuable information and guidance to medical plan developers and health professionals to improve global human health and prevent the occurrence of disease. However, the current GBD disease system's definition of diseases, health consequences caused by diseases, and impact on life expectancy cannot be directly and fully matched with standardized disease definitions. This makes it impossible to directly link the GBD disease system with patient disease records, reducing the efficiency and effectiveness of health assessments.

[0049] In order to solve the above problems, the embodiments of the present application provide a training method for a health assessment model, a health assessment method and related devices, aiming to improve the accuracy of health assessment.

[0050] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0051] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0052] The training method of the health assessment model, the health assessment method and the related device provided in the embodiments of the present application relate to the field of artificial intelligence technology. The training method of the health assessment model and the health assessment method provided in the embodiments of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application of the training method and health assessment method of the health assessment model, etc., but is not limited to the above forms.

[0053] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0054] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of such data will comply with relevant regulations. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0055] The training method, health assessment method and related device of the health assessment model provided in the embodiments of the present application are specifically described through the following embodiments. First, the training method of the health assessment model in the embodiments of the present application is described.

[0056] Reference Figure 1 ,The training method of the health assessment model includes the following steps:

[0057] Step S100, obtaining case data;

[0058] Step S200, matching case data with health status information in the disease system at the disease level to construct first training data;

[0059] Step S300: training an original health assessment model for generating health status according to cases using first training data to obtain a first health assessment model;

[0060] Step S400, adding additional status information to the case data in the first training data to construct the second training data;

[0061] Step S500, training an original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model;

[0062] Step S600 : performing alignment training based on the first health assessment model and the first correlation model, and adjusting the parameters of the first health assessment model to obtain a trained health assessment model.

[0063] In step S100 of some embodiments, the user inputs case data through an input device such as a keyboard or scanner. The case data can be an electronic health record (EHR) or text data converted into digital format by performing image recognition on paper medical records using a camera, scanner, or other device.

[0064] Medical records are stored in hospital databases. Electronic medical records effectively protect patient privacy through access control, encryption, anonymization, and de-identification. Only after obtaining consent from both the patient and the hospital and obtaining read permission can medical records be accessed from the hospital database.

[0065] Reference Figure 2 In some embodiments, in step S200, first training data is constructed based on matching case data with health status information in the disease system at the disease level, including the following steps:

[0066] Step S210, searching and obtaining first target disease information from disease information;

[0067] Step S220 , according to the first matching degree between the first target disease information and the case data, matching the second target disease information from the first target disease information, and constructing the first training data according to the second target disease information and the case data.

[0068] Specifically, the disease system is a set of disease systems defined by the Global Burden of Disease study (GBD), which also includes a weight system for specific health conditions caused by the disease and its impact on life expectancy. The Global Burden of Disease Study is an international research collaboration project composed of global health and disease experts. It aims to estimate the incidence, mortality and disability rates of different diseases and injuries worldwide, as well as the overall burden of these diseases and injuries on human health. The Global Burden of Disease Study uses unified methods and standards to assess the impact of various diseases and injuries on human health, including death, disability and loss of quality of life. The disease system defined by the Global Burden of Disease Study can estimate the health status information of the population in a certain region and further assess healthy life expectancy (average life expectancy after deducting disease).

[0069] For example, Table 1 describes the health status information in the disease system of the Global Burden of Disease Study.

[0070] Table 1 GBD health status information table

[0071]

[0072]

[0073] It is understandable that Table 1 only exemplifies part of the health status information in the disease system of the Global Burden of Disease Study.

[0074] The health status information in the disease system includes disease information, additional status information and corresponding first coding information.

[0075] Use named entity recognition technology to extract health status information from disease systems. For example, extract health status information from the description "Beta-thalassemia major, with severe anemia." The disease information is: beta-thalassemia, severe anemia, and the first code information is: D56.100, D64.903. The first code information is the ICD-10 code.

[0076] The case data includes second coding information, which is the diagnosis code of the physician in the electronic medical record.

[0077] In step S210 of some embodiments, first target disease information is retrieved from disease information based on the first coding information and the second coding information.

[0078] When the first coding information corresponding to the disease information is the same as the second coding information of the case data, the disease information is determined as the first target disease information.

[0079] For example, the second code of the case data is D56.100, the first code of disease information A is D56.100, the first code of disease information B is D56.100, and the first code of disease information C is D55.200. The first codes of disease information A and disease information B are the same as the second code of the case data, both D56.100. Disease information A and disease information B are identified as the first target disease information. The first code of disease information C is different from the second code of the case data, so disease information C is excluded.

[0080] In step S220 of some embodiments, a first matching degree between the first target disease information and the case data is calculated, where the first matching degree includes the semantic similarity between the first target disease information and the case data and the number of first target disease information corresponding to each case data.

[0081] Specifically, the first matching degree is expressed as: Score = Sim + #Hit; where Sim represents the semantic similarity between the first target disease information and the case data, and #Hit represents the number of first target disease information matched to a single case data. A higher first matching degree score indicates a higher degree of overlap between the case data's diagnosis and the health status information of the disease system at the disease level; a lower first matching degree score indicates a lower degree of overlap between the case data's diagnosis and the health status information of the disease system at the disease level.

[0082] The semantic similarity is calculated based on a pre-trained language model, which can use Bidirectional Encoder Representations from Transformers (BERT).

[0083] The process of calculating semantic similarity using BERT is as follows. During the pre-training phase, the BERT model is first pre-trained on a large-scale text corpus to learn language patterns and contextual relationships. BERT randomly masks some words in the input sentence and then attempts to predict these masked words. This allows the model to learn the contextual relationships between words within the sentence. BERT also learns the relationship between sentences by predicting whether two sentences are consecutive. This helps the model understand the logical connections between sentences. The input text is segmented and converted into a series of tokens. These tokens are mapped into a pre-trained word embedding space to obtain word embedding vectors. To preserve the position information of words in the sequence, positional encoding is added to the word embedding vectors. The BERT model processes the data using the Transformer encoder layers, each of which incorporates a self-attention mechanism to capture the contextual relationships between words in the sentence. In this self-attention mechanism, the representation of each word depends not only on its own word vector but also on the representations of all other words, thus achieving bidirectional semantic understanding. To calculate the semantic similarity between two sentences, BERT concatenates the two sentences together to form a sequence and adds a special [CLS] marker at the beginning of the sequence. The final hidden state of the [CLS] tag is used as the representation of the entire sentence pair for subsequent similarity calculations. In the similarity calculation stage, the word embedding vector is further processed through the pooling layer to output the hidden state of each word unit. For example, average pooling or maximum pooling is used to extract a fixed-size representation from all word embedding vectors. The hidden state corresponding to the [CLS] tag is usually used as the representation of the sentence. A similarity measurement method is used to calculate the similarity between the [CLS] vectors of two text fragments. Commonly used similarity measurement methods include cosine similarity, Euclidean distance, etc. For example, cosine similarity is used to calculate semantic similarity. Cosine similarity is expressed as: Where A·B represents the dot product of vector A and vector B, and ||A|| and ||B|| are the norms of vector A and vector B, respectively.

[0084] Match the second target disease information from the first target disease information. Specifically, the first target disease information is sorted according to the first matching score, and the first n first target disease information with the highest first matching scores are selected as the second target disease information. For example, if the first matching score of disease information A is 10, the first matching score of disease information B is 7, the first matching score of disease information D is 8, and the first matching score of disease information E is 9, the first target disease information is sorted from high to low according to the first matching score, and the first three first target disease information with the highest first matching scores are selected as the second target disease information. In this case, disease information A, disease information E, and disease information D are selected as the second target disease information.

[0085] The first training data set is constructed based on the second target disease information and case data. For example, disease information A is severe β-thalassemia with mild anemia; disease information E is severe β-thalassemia with severe infection and severe anemia; and disease information D is severe β-thalassemia without anemia. The first coded information corresponding to disease information A, disease information E, and disease information D is D56.100 and D64.901. The case data is severe β-thalassemia with mild anemia. The case data, disease information A, disease information E, and disease information D form a set of training data, namely the first training data.

[0086] In step S300 of some embodiments, the original health assessment model is a large language model based on a Transformer architecture. The original health assessment model is trained using first training data, and case data, disease information of health status information, and first coding information are used as inputs of the original health assessment model. The original health assessment model predicts the disease information and first coding information of the health status information based on the case data, compares the predicted disease information and first coding information of the health status information with the disease information and first coding information of the actual health status information, obtains a loss function based on the comparison result, and adjusts the parameters of the original health assessment model based on the loss function until the original health assessment model converges or reaches a preset training number threshold, and obtains a first health assessment model from the original health assessment model.

[0087] However, the primary training data is constructed based on mappings between ICD-10 codes at the disease level. In addition to disease information, health status information also includes various additional status details related to the disease, including concomitant symptoms, complications, denials, severity, and historical interventions. Training the primary health model only optimizes consistency at the disease level, not the various additional status details within the disease level. Therefore, optimizing the consistency of these additional status details within the disease level is necessary.

[0088] The original health assessment model can be quickly trained using existing large models in general fields, which is conducive to model training in low-resource environments. It has low dependence on the labeling of input training data and saves training costs.

[0089] Reference Figure 3 In some embodiments, in step S400, the health status information includes additional status information corresponding to the disease information. Adding the additional status information to the case data in the first training data to construct the second training data includes the following steps:

[0090] Step S410, selecting a target prompt word template from preset prompt word templates according to the type of health status information;

[0091] Step S420, filling the disease information and additional status information into the target prompt word template as prompt words to obtain the target instruction;

[0092] Step S430, expanding the case data in the first training data according to the target template, adding additional status information to the case data in the first training data to obtain extended case data, dividing the extended case data into positive samples and negative samples according to the correlation between the extended case data and the health status information, and constructing the second training data from the positive samples and the negative samples.

[0093] Prompt word templates are designed according to the type of health status information; for example, prompt word template 1 is "Please rewrite the case according to the disease {...} but without complications {...}, and do not change other diagnoses; prompt word template 2 is "Please rewrite the case according to the disease {...} with complications {...}, and do not change other diagnoses; prompt word template 3 is "Please rewrite the original case according to the treatment of the disease {...} {...}, and other diagnoses remain unchanged; prompt word template 4 is "The default is the disease {...} that has been treated {...}, and the original case is rewritten as necessary, and other diagnoses remain unchanged; prompt word template 5 is "Please set the disease {...} to the degree {...} and rewrite it, and do not change other diagnoses; prompt word template 6 is "Please set the disease {...} to the degree {...} and rewrite it, and do not change other diagnoses.

[0094] In step S410 of some embodiments, a target prompt word template is selected from preset prompt word templates based on the type of health status information. For example, for the health status information "AIDS, no anemia," prompt word templates 1 and 2 may be selected as target prompt word templates; for the health status information "AIDS, not immunotherapy," prompt word templates 3 and 4 may be selected as target prompt word templates; and for the health status information "severe iron deficiency anemia," prompt word templates 5 and 6 may be selected as target prompt word templates.

[0095] In step S420 of some embodiments, the disease information and additional status information are entered as prompt words into a target prompt word template to obtain a target instruction. For example, entering the health status information "AIDS, no anemia" as a prompt word into prompt word template 1 results in a target instruction of "Please rewrite the case based on {AIDS} but without {anemia}, and do not change other diagnoses." Entering the health status information "AIDS, no anemia" as a prompt word into prompt word template 2 results in a target instruction of "Please rewrite the case based on AIDS with {anemia}, and do not change other diagnoses." Entering the health status information "AIDS, not immunotherapy-treated" as a prompt word into prompt word template 3 results in a target instruction of "Please rewrite the original case based on {immunotherapy} for {AIDS}, and keep other diagnoses unchanged." Entering the health status information "AIDS, not immunotherapy-treated" as a prompt word into prompt word template 4 results in a target instruction of "Default to {AIDS}, having received {immunotherapy}, rewrite the original case as necessary, and keep other diagnoses unchanged." The health status information "severe iron deficiency anemia" is entered as a prompt word into prompt word template 5 to obtain the target instruction "Please set iron deficiency anemia to {severe}, rewrite it, and do not change other diagnoses." The health status information "severe iron deficiency anemia" is entered as a prompt word into prompt word template 6 to obtain the target instruction "Please set iron deficiency anemia to {mild}, rewrite it, and do not change other diagnoses."

[0096] In some embodiments, step S430 expands the case data in the first training data according to the target template, adding additional status information to the case data in the first training data to obtain expanded case data. For example, according to the target instruction "Please rewrite the case based on {AIDS} but without {anemia}, and do not change other diagnoses," the case data of "Diagnosis Conclusion: 1. AIDS 2. Diabetes 3. Pneumonia" is expanded to obtain expanded case data of "Diagnosis Conclusion: 1. AIDS, no anemia 2. Diabetes 3. Pneumonia." According to the target instruction "Please rewrite the case based on AIDS with {anemia}, and do not change other diagnoses," the case data of "Diagnosis Conclusion: 1. AIDS 2. Diabetes 3. Pneumonia" is expanded to obtain expanded case data of "Diagnosis Conclusion: 1. AIDS 2. Anemia 3. Diabetes 4. Pneumonia." According to the target instruction "Please rewrite the original case based on {immunotherapy} for the disease {AIDS}, and keep other diagnoses unchanged," the case data of "Diagnosis Conclusion: 1. AIDS" is expanded to obtain expanded case data of "Diagnosis Conclusion: 1. AIDS, no immunotherapy." According to the target instruction "Default to {AIDS} who has received {immunotherapy}, rewrite the original case as necessary, and keep other diagnoses unchanged", the case data of "Diagnosis Conclusion: 1. AIDS" is expanded to obtain the expanded case data of "Diagnosis Conclusion: AIDS, history of immunotherapy". According to the target instruction "Please set iron deficiency anemia to {severe}, rewrite it, and do not change other diagnoses", the case data of "Diagnosis Conclusion: Iron Deficiency Anemia" is expanded to obtain the expanded case data of "Diagnosis Conclusion: Severe Iron Deficiency Anemia". According to the target instruction "Please set iron deficiency anemia to {mild}, rewrite it, and do not change other diagnoses", the case data of "Diagnosis Conclusion: Iron Deficiency Anemia" is expanded to obtain the expanded case data of "Diagnosis Conclusion: Mild Iron Deficiency Anemia".

[0097] The extended case data is divided into positive samples and negative samples based on their correlation with the health status information, and the second training data is constructed from the positive and negative samples. Specifically, if the extended case data is consistent with the definition of the health status information, the correlation score of the extended case data with the health status information is improved, and the extended case data is used as a positive sample; if the extended case data is inconsistent with the definition of the health status information, the correlation score of the extended case data with the health status information is reduced, and the extended case data is used as a negative sample.

[0098] Reference Figure 4 In step S500 of some embodiments, training the original correlation model for predicting the correlation between cases and health status with the second training data to obtain a first correlation model includes the following steps:

[0099] Step S510, obtaining a correlation score based on the extended case data and health status information;

[0100] Step S520 , performing regression training on the original correlation model according to the correlation score to obtain a first correlation model.

[0101] The correlation score between extended case data and health status information is expressed as: In the formula, Score new Represents the correlation score between extended case data and health status information, Score old represents the correlation score between the case data and health status information in the first training data, the direction of ± depends on the correlation between the extended case data and health status information, ε represents the preset floating score value, N is the number of extended case data, GBD i Represents the sample of the i-th health status information, EHR j represents the sample of the j-th extended case data, and BertScore represents the semantic similarity.

[0102] Some extended case data were scored, and the correlation scores were shown in Table 2.

[0103] Table 2 Correlation score table

[0104]

[0105]

[0106] Regression training is performed on the original correlation model based on the correlation score to obtain a first correlation model. The original correlation model can be based on one of the following regression models: a linear regression model, a polynomial regression model, a decision tree regression model, a random forest regression model, a support vector machine regression model, and a neural network regression model. The original correlation model can predict the correlation score between the health status information and the case data based on the input health status information and case data.

[0107] The original correlation model can be quickly trained using existing large models in general fields, which is conducive to model training in low-resource environments. It has low dependence on the labeling of input training data and saves training costs.

[0108] In step S600 of some embodiments, a reinforcement learning mechanism of Reinforcement Learning from Human Feedback (RLHF) is used to perform alignment training based on the first health assessment model and the first correlation model, and the parameters of the first health assessment model are adjusted to obtain a trained health assessment model. Specifically, in the alignment training phase, the first health assessment model and the first correlation model can be trained simultaneously, sharing part or all of the training data; the first health assessment model and the first correlation model can also be trained alternately, first training on the first health assessment model for several rounds and then training on the first correlation model for several rounds, or first training on the first correlation model for several rounds and then training on the first health assessment model for several rounds. The first health assessment model and the first correlation model are aligned by minimizing the difference in feature representation of the intermediate layers of the two models through similarity loss.

[0109] The trained health assessment model can predict the disease information, additional status information and ICD-10 code of the health status information of the disease system based on case data.

[0110] By matching case data with the disease information, additional status information and ICD-10 codes of health status information in the disease system, the health assessment model can not only accurately assess the disease information based on the case data, but also accurately assess the additional status information based on the case data, which is beneficial for providing valuable information and guidance to doctors and is conducive to subsequent further health assessments; and there is no need to build an additional intermediate mapping conversion process, and it can be directly applied to clinical diagnosis text records, while completing the two tasks of ICD coding and GBD health matching, thereby improving the efficiency of the model processing tasks.

[0111] The embodiments of the present application also provide a health assessment method.

[0112] Reference Figure 5 , the health assessment method includes the following steps:

[0113] Step S700, obtaining case data to be evaluated;

[0114] Step S800: Input the case data to be evaluated into the trained health evaluation model to perform health evaluation and obtain evaluation results. The evaluation results correspond to the disease information and additional status information of the health status information of the disease system.

[0115] Among them, the trained health assessment model is obtained by training according to the training method of the health assessment model of the first aspect of the embodiment of the present application.

[0116] In step S700 of some embodiments, the user enters the case data to be evaluated using an input device such as a keyboard or scanner. The case data to be evaluated can be an electronic medical record, or it can be converted from paper medical records into digital text data through image recognition using a camera, scanner, or other device. The case data to be evaluated can also be stored in a hospital database. The case data can only be read from the hospital database after obtaining the consent of both the patient and the hospital and obtaining read permission.

[0117] In step S800 of some embodiments, the case data to be evaluated is input into a trained health assessment model for health assessment; the trained health assessment model extracts features from the case data to be evaluated, classifies the data according to the features, and obtains disease information, additional status information, and ICD coding information of the health status information of the disease system corresponding to the case data to be evaluated.

[0118] Accurately assessing the disease information, additional condition information, and ICD code information of the GBD disease system's health status information based on case data provides valuable information and guidance to doctors, facilitating subsequent health assessments. For example, the healthy life expectancy of the population in a region can be estimated based on the disease information, additional condition information, and ICD code information of the GBD disease system's health status information.

[0119] The embodiments of the present application also provide a training device for a health assessment model.

[0120] Reference Figure 6 The training device of the health assessment model includes: an input module 801, a first data construction module 802, a first model training module 803, a second data construction module 804, a second model training module 805 and a third model training module 806.

[0121] Among them, the input module 801 is used to obtain case data; the first data construction module 802 is used to match the case data with the health status information in the disease system at the disease level to construct the first training data; the first model training module 803 is used to train the original health assessment model used to generate the health status based on the case through the first training data to obtain the first health assessment model; the second data construction module 804 is used to add additional status information to the case data in the first training data to construct the second training data; the second model training module 805 is used to train the original correlation model used to predict the correlation between the case and the health status through the second training data to obtain the first correlation model; the third model training module 806 is used to perform alignment training based on the first health assessment model and the first correlation model, and adjust the parameters of the first health assessment model to obtain the trained health assessment model.

[0122] It is understood that the training device for the health assessment module of the embodiment of the present application applies the above-mentioned training method for the health assessment module. The modules of the training device for the health assessment module of the embodiment of the present application correspond to the steps of the training method for the health assessment module. The training device for the health assessment module and the training method for the health assessment module use the same technical means, solve the same technical problems, and bring about the same technical effects, and will not be repeated here.

[0123] The embodiment of the present application also provides an electronic device. Figure 7 The electronic device includes a memory 902 and a processor 901. The memory 902 stores a computer program. When the processor 901 executes the computer program, the training method and health assessment method of the health assessment model are implemented. The electronic device can be any smart terminal including a tablet computer, an in-vehicle computer, etc.

[0124] The processor 901 can be implemented in the form of a general-purpose CPU (Central Processing Unit, central processing unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application; the memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 is used to call and execute the training method and health assessment method of the health assessment model of the embodiment of the present application.

[0125] The input / output interface 903 is used to realize information input and output; the communication interface 904 is used to realize communication interaction between this device and other devices, and communication can be realized through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.); the bus 905 transmits information between the various components of the device (such as the processor 901, memory 902, input / output interface 903 and communication interface 904); among them, the processor 901, memory 902, input / output interface 903 and communication interface 904 realize communication connection with each other within the device through the bus 905.

[0126] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the training method and health assessment method of the above-mentioned health assessment model.

[0127] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory 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 may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor 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 embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0129] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0130] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0131] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0132] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0133] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0134] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0135] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0136] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0137] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: various media that can store programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0138] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A training method for a health assessment model, characterized in that: include: Obtain case data; Constructing first training data based on matching the case data with health status information in the disease system at the disease level; Training an original health assessment model for generating health status according to cases using the first training data to obtain a first health assessment model; Adding additional status information to the case data in the first training data to construct second training data; Training an original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model; performing alignment training according to the first health assessment model and the first correlation model, and adjusting parameters of the first health assessment model to obtain a trained health assessment model; The health status information includes disease information and corresponding first coding information, and the case data includes second coding information; and the first training data is constructed by matching the case data with the health status information in the disease system at the disease level, including: Searching and obtaining first target disease information from the disease information, wherein first coding information of the first target disease information is similar to second coding information of the case data; According to a first matching degree between the first target disease information and the case data, second target disease information is matched from the first target disease information, and first training data is constructed based on the second target disease information and the case data.

2. The training method of the health assessment model according to claim 1, characterized in that: The first matching degree includes the semantic similarity between the first target disease information and the case data and the number of first target disease information corresponding to each case data.

3. The training method of the health assessment model according to claim 1, characterized in that: The health status information includes additional status information corresponding to the disease information; and the adding of the additional status information to the case data in the first training data to construct the second training data includes: selecting a target prompt word template from preset prompt word templates according to the type of the health status information; Filling the disease information and the additional status information into the target prompt word template as prompt words to obtain a target instruction; The case data in the first training data is expanded according to the target template, additional status information is added to the case data in the first training data to obtain extended case data, and the extended case data is divided into positive samples and negative samples according to the correlation between the extended case data and the health status information, and the second training data is constructed by the positive samples and the negative samples.

4. The training method of the health assessment model according to claim 3, characterized in that: The method of training the original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model includes: Obtaining a relevance score based on the extended case data and the health status information; Regression training is performed on the original correlation model according to the correlation score to obtain a first correlation model.

5. The training method of the health assessment model according to claim 4, characterized in that: The correlation score between the extended case data and the health status information is expressed as: In the formula, Score new Score represents the correlation score between the extended case data and the health status information. old represents the correlation score between the case data in the first training data and the health status information, the direction of ± depends on the correlation between the extended case data and the health status information, ε represents a preset floating score value, N is the number of extended case data, GBD i Represents the sample of the i-th health status information, EHR j represents the sample of the j-th extended case data, and BertScore represents the semantic similarity.

6. A health assessment method, characterized in that: include: Obtain data on cases to be evaluated; Inputting the case data to be evaluated into a trained health assessment model to perform health assessment and obtain an assessment result, wherein the assessment result corresponds to the disease information and additional status information of the health status information of the disease system; The trained health assessment model is obtained by training according to the health assessment model training method according to any one of claims 1 to 5.

7. A training device for a health assessment model, characterized in that: include: Input module, used to obtain case data; A first data construction module is used to match the case data with the health status information in the disease system at the disease level to construct first training data; A first model training module is configured to train an original health assessment model for generating health status according to a case using the first training data to obtain a first health assessment model; A second data construction module is used to add additional status information to the case data in the first training data to construct second training data; A second model training module is used to train an original correlation model for predicting the correlation between cases and health status using the second training data to obtain a first correlation model; a third model training module, configured to perform alignment training based on the first health assessment model and the first correlation model, and adjust parameters of the first health assessment model to obtain a trained health assessment model; The health status information includes disease information and corresponding first coding information, and the case data includes second coding information; and the first training data is constructed by matching the case data with the health status information in the disease system at the disease level, including: Searching and obtaining first target disease information from the disease information, wherein first coding information of the first target disease information is similar to second coding information of the case data; According to a first matching degree between the first target disease information and the case data, second target disease information is matched from the first target disease information, and first training data is constructed based on the second target disease information and the case data.

8. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the training method of the health assessment model described in any one of claims 1 to 5 and the health assessment method described in claim 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the training method of the health assessment model according to any one of claims 1 to 5 and the health assessment method according to claim 6 are implemented.

Images