Report conclusion generation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a report conclusion generation method and device, electronic equipment and a storage medium, and belongs to the technical field of AI and digital medical treatment. The method comprises the steps of obtaining first sample inspection content and a first sample report conclusion; encoding the first sample inspection content according to an original encoder to obtain content features; decoding the content features according to an original decoder to obtain a prediction report conclusion; performing parameter adjustment on the original connector according to the prediction report conclusion and the first sample report conclusion to obtain a target connector; performing parameter freezing training on the original encoder to obtain a target encoder, and performing parameter freezing training on the original decoder to obtain a target decoder; obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder; and performing conclusion generation processing on the target inspection report according to the target conclusion generation model to obtain a target report conclusion. According to the embodiment of the invention, the universality of generating the report conclusion based on the report can be improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and digital medical technology, and in particular to a method and device for generating report conclusions, an electronic device, and a storage medium. Background Art

[0002] At present, information retrieval, question answering, intelligent assistant and other functions can be realized based on question-answering technology. For example, in digital medical scenarios, medical reports can be understood based on question-answering technology to obtain corresponding report conclusions. In related technologies, question-answering technology includes VQA (Visual Question Answering), reading comprehension technology and text summarization technology. Among them, VQA can only understand images; reading comprehension technology and text summarization technology can understand text content, but the understanding effect depends on the quality of the object's questions. In other words, the above-mentioned question-answering technologies have certain limitations in use, which is not conducive to use in scenarios such as digital medicine. Based on this, how to provide a method for generating report conclusions to improve the universality of report conclusions generated based on reports has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The main purpose of the embodiments of the present application is to propose a method and device for generating report conclusions, an electronic device and a storage medium, aiming to improve the universality of report conclusions generated based on reports.

[0004] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for generating a report conclusion, the method comprising:

[0005] Obtain the first sample inspection content and the first sample report conclusion of the first sample inspection report;

[0006] Inputting the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector, and an original decoder;

[0007] encoding the first sample inspection content according to the original encoder to obtain content features;

[0008] Mapping the content features to the original decoder according to the original connector, decoding the content features according to the original decoder, and obtaining a prediction report conclusion;

[0009] According to the conclusion of the prediction report and the conclusion of the first sample report, the parameters of the original connector are adjusted to obtain a target connector; the parameters of the original encoder are frozen and trained to obtain a target encoder, and the parameters of the original decoder are frozen and trained to obtain a target decoder;

[0010] Obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder;

[0011] The acquired target inspection report is processed for conclusion generation according to the target conclusion generation model to obtain the target report conclusion.

[0012] In some embodiments, adjusting parameters of the original connector according to the prediction report conclusion and the first sample report conclusion to obtain a target connector; performing parameter freezing training on the original encoder to obtain a target encoder, and performing parameter freezing training on the original decoder to obtain a target decoder, comprises:

[0013] Freezing model parameters of the original encoder to obtain a frozen encoder, and freezing model parameters of the original decoder to obtain a frozen decoder;

[0014] Adjusting parameters of the original connector according to the conclusion of the prediction report and the conclusion of the first sample report to obtain the target connector;

[0015] Thawing the frozen encoder to obtain a thawed encoder, and adjusting parameters of the frozen encoder according to the prediction report conclusion and the first sample report conclusion to obtain the target encoder;

[0016] The frozen decoder is unfrozen to obtain a target decoder.

[0017] In some embodiments, the thawing of the frozen decoder to obtain a target decoder includes:

[0018] Thawing the frozen decoder to obtain a thawed decoder;

[0019] Obtaining the model type of the unfrozen decoder;

[0020] If the model type is the same as the preset type, using the unfrozen decoder as the target decoder;

[0021] If the model type is different from the preset type, parameters of the unfrozen decoder are adjusted according to the prediction report conclusion and the first sample report conclusion to obtain the target decoder.

[0022] In some embodiments, obtaining the first sample inspection content and the first sample report conclusion of the first sample inspection report includes:

[0023] Performing content segmentation on the first sample inspection report based on a pre-trained target content segmentation model to obtain at least two report images and the first sample report conclusion;

[0024] Data splicing is performed on all the report images according to the pre-constructed single conclusion requirement data to obtain the first sample inspection content.

[0025] In some embodiments, before performing content segmentation on the sample inspection report based on a preset target content segmentation model, the method further includes training the target content segmentation model, including:

[0026] Obtaining a second sample inspection report, wherein the second sample inspection report includes a sample image;

[0027] Performing a conclusion query on the second sample inspection report to obtain a conclusion query result;

[0028] If the conclusion query result indicates that the second sample inspection report does not include a report conclusion, fill in the conclusion of the second sample inspection report to obtain a second sample report conclusion;

[0029] Performing content segmentation on the second sample inspection report according to a preset original content segmentation model to obtain a predicted image and a predicted report conclusion;

[0030] The parameters of the original content segmentation model are adjusted according to the predicted image, the prediction report conclusion, the second sample report conclusion and the sample image to obtain the target content segmentation model.

[0031] In some embodiments, obtaining the first sample inspection content and the first sample report conclusion of the first sample inspection report includes:

[0032] Randomly sampling the first sample inspection report to obtain a candidate report;

[0033] Performing content segmentation on the candidate report based on a pre-trained target content segmentation model to obtain candidate images and candidate report conclusions;

[0034] Performing data splicing on the pre-constructed multi-conclusion requirement data and the candidate image to obtain the first sample inspection content;

[0035] The pre-constructed conclusion connection data and the candidate report conclusion are concatenated to obtain the first sample report conclusion.

[0036] In some embodiments, performing conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion includes:

[0037] Determining report data associated with the target inspection report according to a preset association condition;

[0038] Performing data splicing on the report data and the target inspection report to obtain target processing data;

[0039] The target processing data is processed for conclusion generation according to the target conclusion generation model to obtain the target report conclusion.

[0040] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a device for generating a report conclusion, the device comprising:

[0041] An inspection report acquisition module, used to obtain the first sample inspection content and the first sample report conclusion of the first sample inspection report;

[0042] A data input module, used for inputting the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector and an original decoder;

[0043] An encoding module, used for encoding the first sample inspection content according to the original encoder to obtain content features;

[0044] A decoding module, used for mapping the content features to the original decoder according to the original connector, decoding the content features according to the original decoder, and obtaining a prediction report conclusion;

[0045] A model training module, used to adjust the parameters of the original connector according to the prediction report conclusion and the first sample report conclusion to obtain a target connector; perform parameter freezing training on the original encoder to obtain a target encoder, and perform parameter freezing training on the original decoder to obtain a target decoder;

[0046] A model building model, for obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder;

[0047] The conclusion generation module is used to perform conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion.

[0048] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0049] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0050] The report conclusion generation method and device, electronic device and storage medium proposed in the present application generate a prediction report conclusion corresponding to the first sample inspection content through the original conclusion generation model. The original connector is parameter adjusted through the prediction report conclusion and the first sample report conclusion, and the original encoder and the original decoder are respectively trained to freeze the parameters, so that the target conclusion generation model can be obtained. The target inspection report is concluded according to the target conclusion generation model to obtain the corresponding target report conclusion. It can be seen that the embodiment of the present application can directly generate conclusions based on the target conclusion generation model, reducing the situation in the related art that the quality of conclusion generation depends on the quality of the object question, that is, the embodiment of the present application is not sensitive to the problem. Therefore, the embodiment of the present application can improve the universality of report conclusions based on report generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a flow chart of a method for generating a report conclusion provided in an embodiment of the present application;

[0052] Figure 2 yes Figure 1 Flow chart of step S105 in FIG.

[0053] Figure 3 yes Figure 2 Flow chart of step S204 in FIG.

[0054] Figure 4 yes Figure 1 Flow chart of step S101 in FIG.

[0055] Figure 5 yes Figure 1 Flow chart of step S101 in FIG.

[0056] Figure 6 is a flow chart of a method for training a target content segmentation model provided in an embodiment of the present application;

[0057] Figure 7 yes Figure 1 Flow chart of step S107 in FIG.

[0058] Figure 8 is a schematic diagram of a multi-round dialogue scenario provided in an embodiment of the present application;

[0059] Fig. 9 It is a structural schematic diagram of a device for generating report conclusions provided in an embodiment of the present application;

[0060] Fig.10 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0062] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0064] First, some nouns involved in this application are analyzed:

[0065] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science. AI attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a 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 obtain the best results.

[0066] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is often used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0067] At present, question-answering technology can realize functions such as information retrieval, question answering, and intelligent assistants. For example, in digital medical scenarios, medical reports can be understood based on question-answering technology to obtain corresponding report conclusions. Among the related technologies, question-answering technology includes VQA (Visual Question Answering), reading comprehension technology, and text summarization technology.

[0068] VQA combines computer vision and natural language processing technologies. The input data of VQA includes images and questions about the images, and the output data of VQA includes answers to the questions. Therefore, it can be seen that VQA can only understand images.

[0069] Reading comprehension technology can provide responses based on text content and questions, but it is sensitive to questions. Inappropriate questions may cause the model corresponding to the technology to misunderstand and generate incorrect answers.

[0070] Text summarization technology is an artificial intelligence technology that combines text summarization and question-answering systems. Text summarization technology aims to provide a concise and accurate summary answer based on the question by reading and understanding the question. Text summarization technology is similar to reading comprehension technology, that is, the understanding effect of text summarization technology depends on the quality of the question asked by the object.

[0071] It can be seen that the above-mentioned question-and-answer technologies have certain limitations in use and are not conducive to use in scenarios such as digital medicine.

[0072] Based on this, the embodiments of the present application provide a method and device for generating a report conclusion, an electronic device and a storage medium, aiming to improve the universality of generating a report conclusion based on a report.

[0073] The method and device for generating report conclusions, electronic device and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the method for generating report conclusions in the embodiments of the present application is described.

[0074] 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 obtain the best results.

[0075] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0076] The method for generating report conclusions provided in the embodiment of the present application relates to the field of artificial intelligence technology. The method for generating report conclusions provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a 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 a distributed system composed of multiple physical servers, or can 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 that implements the method for generating report conclusions, etc., but is not limited to the above forms.

[0077] 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, etc. 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, etc. 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 through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0078] It should be noted that in each specific implementation 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 medical reports, 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 these data will comply with relevant laws, regulations, and standards. 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.

[0079] Figure 1 is an optional flowchart of the method for generating a report conclusion provided in an embodiment of the present application, Figure 1 The method may include but is not limited to steps S101 to S107.

[0080] Step S101, obtaining the first sample inspection content and the first sample report conclusion of the first sample inspection report;

[0081] Step S102, inputting the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector, and an original decoder;

[0082] Step S103, encoding the first sample inspection content according to the original encoder to obtain content features;

[0083] Step S104, mapping the content features to the original decoder according to the original connector, decoding the content features according to the original decoder, and obtaining a prediction report conclusion;

[0084] Step S105, adjusting the parameters of the original connector according to the conclusion of the prediction report and the conclusion of the first sample report to obtain a target connector; performing parameter freezing training on the original encoder to obtain a target encoder, and performing parameter freezing training on the original decoder to obtain a target decoder;

[0085] Step S106, obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder;

[0086] Step S107, performing conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion.

[0087] Steps S101 to S107 shown in the embodiment of the present application generate a prediction report conclusion corresponding to the first sample inspection content through the original conclusion generation model. The original connector is parameter adjusted through the prediction report conclusion and the first sample report conclusion, and the original encoder and the original decoder are respectively trained to freeze parameters, so that the target conclusion generation model can be obtained. The target inspection report is concluded according to the target conclusion generation model to obtain the corresponding target report conclusion. It can be seen that the embodiment of the present application can directly generate conclusions based on the target conclusion generation model, reducing the situation in the related art that the quality of conclusion generation depends on the quality of the object question, that is, the embodiment of the present application is not sensitive to the problem. Therefore, the embodiment of the present application can improve the universality of report conclusions based on report generation.

[0088] In step S101 of some embodiments, the first sample inspection report may refer to a report including the first sample inspection content and the first sample report conclusion. Among them, the first sample report conclusion can be obtained based on the first sample inspection content, and the first sample inspection content can refer to the content used to describe a certain matter. When the embodiments of the present application are applied to different scenarios, the types of the first sample inspection reports may be different. For example, when the embodiments of the present application are applied to digital medical scenarios, the first sample inspection report may refer to a physical examination report and an inspection report, etc. At this time, the first sample inspection content may refer to the content used to describe the physical examination matters, such as the first sample inspection content may include various physical examination parameters. The first sample report conclusion may refer to the physical examination conclusion obtained based on various physical examination parameters. When the embodiments of the present application are applied to insurance claims scenarios, the first sample inspection report may refer to a vehicle damage report, etc. At this time, the first sample inspection content may refer to the content describing the vehicle damage, such as the first sample inspection content may include vehicle damage photos, repair lists and repair quotations, etc. The first sample report conclusion may refer to the vehicle claim conclusion drawn based on vehicle damage photos, repair lists and repair quotations, etc.

[0089] It is understandable that the first sample inspection report can be obtained from the corresponding database. For example, when the embodiment of the present application is applied to the insurance claim scenario, the first sample inspection content can be obtained from the database corresponding to the claim platform, and the first sample report conclusion corresponding to the first sample inspection content can be obtained, so that the first sample inspection report can be combined. In other words, the first sample inspection report may originally have the first sample inspection content and the first sample report conclusion, and the first sample inspection report can be directly obtained at this time. Alternatively, it is necessary to first obtain the first sample inspection content and the first sample report conclusion separately, and then combine the first sample inspection content and the first sample report conclusion to obtain the first sample inspection report.

[0090] In step S102 of some embodiments, the original conclusion generation model may refer to a model having the ability to generate corresponding conclusions according to the inspection content. The original conclusion generation model may be a structure of an original encoder-original connector-original decoder. Among them, the original encoder is used to convert the input data into a vector of a fixed size, which captures the main features and information of the input data. The original connector can pass the output data of the original encoder to the original decoder. The original decoder can generate a target sequence, such as can be used to generate corresponding conclusions in the embodiment of the present application. Specifically, the original encoder can map the input data to a space representing understanding, and the original connector is used to map the understanding space output by the original encoder to the input space of the original decoder, that is, to the space to be decoded of the original decoder. The original decoder is used to map the space to be decoded to the natural language text space. Among them, the original connector can be one or more linear mapping layers. Alternatively, the last few layers of the original encoder, or the first few layers of the original decoder can be used as the original connector.

[0091] In step S103 of some embodiments, the first sample inspection content is used as input data of the original encoder, and the original encoder performs feature capture on the first sample inspection content to obtain content features.

[0092] In step S104 of some embodiments, the original connector maps the content features output by the original encoder to the to-be-decoded space of the original decoder. The original decoder maps the content features to the natural language text space to decode the content features, thereby obtaining a prediction report conclusion. It is understandable that the prediction report conclusion may refer to the conclusion predicted by the original conclusion generation model based on the first sample inspection content.

[0093] In step S105 of some embodiments, in order to improve the accuracy of the conclusion generated by the original conclusion generation model, the original conclusion generation model can be trained. Specifically, the parameters of the original connector can be adjusted according to the prediction report conclusion and the first sample report conclusion to improve the mapping ability of the original connector. At the same time, parameter freezing training can also be performed on the original encoder and the original decoder. It can be understood that parameter freezing training can refer to freezing the model parameters of itself when training other structures. When training itself, the model parameters are thawed. In other words, the embodiment of the present application can train the original encoder, the original connector and the original decoder separately.

[0094] The target encoder can be obtained after parameter freezing training of the original encoder, and the target decoder can be obtained after parameter freezing training of the original decoder.

[0095] Reference Figure 2 In some embodiments, step S105 includes but is not limited to steps S201 to S204.

[0096] Step S201, freezing the model parameters of the original encoder to obtain a frozen encoder, and freezing the model parameters of the original decoder to obtain a frozen decoder;

[0097] Step S202, adjusting parameters of the original connector according to the conclusion of the prediction report and the conclusion of the first sample report to obtain a target connector;

[0098] Step S203, unfreeze the frozen encoder to obtain a unfrozen encoder, and adjust parameters of the frozen encoder according to the conclusion of the prediction report and the conclusion of the first sample report to obtain a target encoder;

[0099] Step S204, unfreeze the frozen decoder to obtain a target decoder.

[0100] In steps S201 to S202 of some embodiments, since the original encoder and the original decoder have learned more knowledge, that is, they have certain capabilities, the original connector can be trained first to improve the linkage effect of the original encoder and the original decoder. Specifically, the model parameters of the original encoder are frozen to obtain a frozen encoder, and the model parameters of the original decoder are frozen to obtain a frozen decoder. On this basis, the parameters of the original connector are adjusted according to the conclusions of the prediction report and the conclusions of the first sample report to reduce the error generated when the original connector maps the understanding space output by the original encoder to the input space of the original decoder, so as to obtain the target connector. It can be understood that freezing model parameters can refer to fixing the parameters of all layers or most layers of the model, that is, not updating these fixed parameters. Take the original encoder as an example. The parameters of all layers of the original encoder can be fixed, or the parameters of the first few layers of the original encoder can be fixed, that is, the parameters of the last few layers of the original encoder can be updated. The last few layers of the original encoder can refer to the model layers that are closely connected to the original connector.

[0101] In step S203 of some embodiments, the frozen encoder is thawed to obtain a thawed encoder. The thawed encoder can participate in training, that is, the parameters of the thawed encoder can be adjusted to obtain a target encoder. Specifically, the encoding error of the thawed encoder can be determined according to the prediction report conclusion and the first sample report conclusion, so that the parameters of the thawed encoder can be adjusted according to the encoding error.

[0102] It is understandable that when all layers of the original encoder are parameter frozen, the unfrozen encoder is the original encoder. At this time, there is no need to update the prediction report conclusion, that is, the prediction report conclusion is still based on the conclusion generated by the original encoder.

[0103] When the parameters of most layers of the original encoder are frozen, the unfrozen encoder is different from the original encoder. At this time, the prediction report conclusion needs to be updated, that is, the first sample inspection content is input into the unfrozen encoder to obtain a new prediction report conclusion.

[0104] In step S204 of some embodiments, it can be determined whether the original decoder needs to be trained according to the application scenario. For example, when the original decoder is an LLM (Large Language Model 1) structure, training the original decoder may reduce the emergence ability of the original decoder as a large language model. At this time, the frozen decoder can be directly unfrozen to obtain a target decoder. It can be understood that in this case, the target decoder is the original decoder.

[0105] It is understandable that when the application scenario is a multi-round dialogue scenario, the original decoder can be set to an LLM structure. When the application scenario is a scenario where a conclusion is directly generated based on an inspection report, the frozen decoder can be thawed, and the thawed frozen decoder can be trained to obtain a target decoder. It is understandable that the method for training the thawed frozen decoder is similar to the method for training the thawed encoder in step S203. The training method of the thawed frozen decoder is described below.

[0106] Reference Figure 3 In some embodiments, step S204 includes but is not limited to steps S301 to S304.

[0107] Step S301, thawing the frozen decoder to obtain a thawed decoder;

[0108] Step S302, obtaining the model type of the unfreezing decoder;

[0109] Step S303, if the model type is the same as the preset type, the unfrozen decoder is used as the target decoder;

[0110] Step S304: if the model type is different from the preset type, adjust the parameters of the unfrozen decoder according to the prediction report conclusion and the first sample report conclusion to obtain a target decoder.

[0111] In step S301 of some embodiments, the frozen decoder is thawed to obtain a thawed decoder, and the thawed decoder can participate in training. It is understandable that when all layers of the original decoder are parameter frozen, the thawed decoder is the original decoder. When most layers of the original decoder are parameter frozen, the thawed decoder is different from the original decoder.

[0112] In step S302 of some embodiments, the model type may refer to the structural type of the model, such as the LLM structural type. The model type of the unfrozen decoder is the same as the model type of the original decoder, so the model type of the unfrozen decoder can be determined according to the model type of the original decoder.

[0113] In step S303 of some embodiments, when it is desired not to train a certain type of original decoder, the model type of the original decoder can be used as a preset type. For example, when the application scenario is a multi-round dialogue scenario, the LLM structure type can be set to the preset type to avoid training the thawed decoder of the LLM type. In other words, when the model type of the thawed decoder is the same as the preset type, such as the LLM structure type, the thawed decoder can be directly used as the target decoder.

[0114] In step S304 of some embodiments, when the model type of the unfrozen decoder is different from the preset type, a decoding error of the unfrozen decoder is determined according to the prediction report and the conclusion of the first sample report. Parameters of the unfrozen decoder are adjusted according to the decoding error to obtain a target decoder.

[0115] It is understandable that when all layers of the original decoder are parameter frozen, the unfrozen decoder is the original decoder. At this time, the prediction report conclusion can be updated, such as the first sample inspection content can be used as input data, and the prediction report conclusion can be updated according to the output data of the target encoder-target connector-original decoder.

[0116] When the parameters of most layers of the original encoder are frozen, the unfrozen encoder is different from the original encoder. At this time, the prediction report conclusion can be updated, such as using the first sample inspection content as input data and updating the prediction report conclusion according to the output data of the target encoder-target connector-unfrozen decoder.

[0117] In step S106 of some embodiments, the target encoder, the target connector, and the target decoder are connected to obtain a target conclusion generation model.

[0118] In step S107 of some embodiments, the target inspection report may refer to a report for which conclusion generation is to be performed in an actual application scenario. The target inspection report is used as input data of the target conclusion generation model to obtain a target report conclusion corresponding to the inspection content in the target inspection report. It is understandable that in the initial state, the target inspection report may include or exclude a conclusion. When the target inspection report does not include a conclusion, the target report conclusion may be directly used as the conclusion of the target inspection report. When the target inspection report includes an original conclusion, the target report conclusion may be compared with the original conclusion, and the final conclusion of the target inspection report may be obtained based on the comparison result, thereby improving the accuracy of the target inspection report conclusion.

[0119] It is understandable that the target conclusion generation model trained according to the above embodiment can directly generate the corresponding target report conclusion according to the target inspection report, wherein the target inspection report can be a single-page report or a multi-page report.

[0120] In some embodiments, when the application scenario is a multi-round dialogue scenario, the form of the first sample inspection report can be adjusted so that the target conclusion generation model obtained by training based on the first sample inspection report has multi-round dialogue capabilities. The adjustment method of the first sample inspection report is described as follows.

[0121] It is understandable that the embodiment of the present application does not specifically limit the form of the first sample inspection report, that is, the first sample inspection report can be in text form, image form, OCR (Optical Character Recognition) form, or a combination of text and image, etc. For ease of explanation, the following description is based on the example of the first sample inspection report in image form.

[0122] Reference Figure 4 In some embodiments, step S101 includes but is not limited to steps S401 to S402.

[0123] Step S401, performing content segmentation on a first sample inspection report based on a pre-trained target content segmentation model to obtain at least two report images and a first sample report conclusion;

[0124] Step S402: perform data splicing on all report images according to pre-constructed single conclusion requirement data to obtain the first sample inspection content.

[0125] In step S401 of some embodiments, the target content segmentation model may refer to a model that has the ability to segment content from conclusions. The target content segmentation model may be pre-trained. The first sample inspection report is used as input data of the target content segmentation model to obtain the report content and the first sample report conclusion. It is understandable that the first sample inspection report may be a multi-page report, and the report image may refer to an image corresponding to the report content portion in each page of the report. Therefore, at least two report images may be segmented from the first sample inspection report.

[0126] In step S402 of some embodiments, the single conclusion requirement data may represent data that only requires a single conclusion, such as: Please tell me a general conclusion. The single conclusion requirement data is spliced ​​with all report images to obtain a sample with the "multiple content single conclusion" characteristic, that is, the first sample inspection content is obtained. Therefore, when the original conclusion generation model is subsequently trained according to the first sample inspection content, the trained original conclusion generation model can be enabled to have the ability to generate a single conclusion based on multiple contents.

[0127] It is understandable that in order to make the trained original conclusion generation model insensitive to the problem, multiple different forms of single conclusion requirement data can be set.

[0128] Take the first sample inspection report as an example of a physical examination report in a digital medical scenario. By performing content segmentation on the first sample inspection report according to the target content segmentation model, two report images, image 1 and image 2, and the first sample report conclusions corresponding to image 1 and image 2 can be obtained. For example, image 1 and image 2 represent kidney images, and the first sample report conclusion can be "Conclusion: A stone of size xx can be seen in the left kidney." Image 1 and image 2 are spliced ​​according to the single conclusion requirement data, and the first sample inspection content obtained can be "Please tell me the conclusions of the following report: (image 1), (image 2)".

[0129] Reference Figure 5 In some other embodiments, step S101 may include but is not limited to steps S501 to S504.

[0130] Step S501, randomly sampling the first sample inspection report to obtain a candidate report;

[0131] Step S502, performing content segmentation on the candidate report based on the pre-trained target content segmentation model to obtain candidate images and candidate report conclusions;

[0132] Step S503, performing data splicing on the pre-constructed multi-conclusion requirement data and the candidate images to obtain the first sample inspection content;

[0133] Step S504, concatenating the pre-constructed conclusion connection data and the candidate report conclusion to obtain the first sample report conclusion.

[0134] In step S501 of some embodiments, the first sample inspection report may be a single-page report. A plurality of first sample inspection reports are randomly selected, and the selected first sample inspection reports are used as candidate reports. It is understandable that the number of the selected candidate reports may be adaptively set according to actual needs, and this embodiment of the present application does not specifically limit this.

[0135] In step S502 of some embodiments, content segmentation is performed on each candidate report based on a target content segmentation model to obtain a candidate image corresponding to the content portion of the candidate report and a candidate report conclusion corresponding to the content.

[0136] In step S503 of some embodiments, the multi-conclusion requirement data may represent data that requires multiple conclusions, such as: Please tell me what these conclusions are. The multi-conclusion requirement data is spliced ​​with all candidate images to obtain a sample with the "multi-content multi-conclusion" feature, that is, the first sample inspection content is obtained. Therefore, when the original conclusion generation model is subsequently trained according to the first sample inspection content, the trained original conclusion generation model can be enabled to have the ability to generate multiple conclusions based on multiple contents.

[0137] It can be understood that in order to make the trained original conclusion generation model insensitive to the problem, multiple different forms of multi-conclusion requirement data can be set.

[0138] Taking the first sample examination report as a physical examination report in a digital medical scenario as an example, each first sample examination report can represent a physical examination report on a part. For example, there are three physical examination reports in total, the first physical examination report is a physical examination report on the kidneys, the second physical examination report is a physical examination report on the liver, and the third physical examination report is a physical examination report on the chest. The first physical examination report and the second physical examination report are randomly selected as candidate reports. According to the target content segmentation model, the first physical examination report is segmented to obtain the candidate image 1 corresponding to the kidney B-ultrasound and the candidate report conclusion of "a stone of size xx can be seen in the left kidney". According to the target content segmentation model, the second physical examination report is segmented to obtain the candidate image 2 corresponding to the liver function data part and the candidate report conclusion of "no abnormality in liver function". According to the multi-conclusion requirement data, all candidate images are spliced, and the first sample inspection content obtained can be "Please tell me what the conclusions of candidate image 1 and candidate image 2 are respectively".

[0139] In step S504 of some embodiments, the conclusion connection data may refer to data used to connect all candidate report conclusions to make the word order smooth. For example, the conclusion connection data may refer to conjunctions such as "at the same time", "furthermore", and "and". All candidate report conclusions are spliced ​​according to the conclusion connection data, and the first sample report conclusion obtained may be "a stone of size xx is visible in the left kidney, and no abnormality is found in liver function".

[0140] The advantage of steps S501 to S504 is that, through random sampling, the multiple candidate images for data splicing are not necessarily images with obvious correlation in content. Therefore, when the original conclusion generation model is subsequently trained based on the first sample inspection content, the ability of the original conclusion generation model to derive conclusions can be improved.

[0141] Reference Figure 6 In some embodiments, the method of training the target content segmentation model includes but is not limited to steps S601 to S605.

[0142] Step S601, obtaining a second sample inspection report, wherein the second sample inspection report includes a sample image;

[0143] Step S602, querying the conclusion of the second sample inspection report to obtain a conclusion query result;

[0144] Step S603: if the conclusion query result indicates that the second sample inspection report does not contain a report conclusion, the second sample inspection report is filled with a conclusion to obtain a second sample report conclusion;

[0145] Step S604, performing content segmentation on the second sample inspection report according to a preset original content segmentation model to obtain a predicted image and a predicted report conclusion;

[0146] Step S605, adjusting the parameters of the original content segmentation model according to the predicted image, the prediction report conclusion, the second sample report conclusion and the sample image to obtain a target content segmentation model.

[0147] In step S601 of some embodiments, the method for obtaining the second sample inspection report may be the same as or different from the method for obtaining the first sample inspection report, and the report type of the second sample inspection report may be the same as or different from the report type of the first sample inspection report, which is not specifically limited in the embodiments of the present application. The sample image may refer to an image corresponding to the content portion in the second sample inspection report.

[0148] In step S602 of some embodiments, conclusion query may refer to querying whether the second sample inspection report includes a report conclusion corresponding to the sample image. The conclusion query result may include that the second sample inspection report includes a report conclusion, or that the second sample inspection report does not include a report conclusion.

[0149] In step S603 of some embodiments, when the conclusion query result indicates that the second sample inspection report does not include a report conclusion, the conclusion can be filled in according to the content corresponding to the sample image to obtain the second sample report conclusion. It is understandable that when the conclusion query result indicates that the second sample inspection report includes a report conclusion, the report conclusion in the second sample inspection report can be directly used as the second sample report conclusion.

[0150] In step S604 of some embodiments, the second sample inspection report is used as input data of the original content segmentation model to perform content segmentation on the second sample inspection report according to the original content segmentation model to obtain a predicted image and a predicted report conclusion. The predicted image may refer to an image corresponding to the content portion in the second sample inspection report predicted by the original content segmentation model. The predicted report conclusion may refer to a conclusion extracted from the second sample inspection report by the original content segmentation model.

[0151] In step S605 of some embodiments, the predicted image is compared with the sample image to determine the recognition error of the original content segmentation model for the content part. The predicted report conclusion is compared with the conclusion of the second sample report period to determine the error of the original content segmentation model in extracting the conclusion. According to the above two errors, the parameters of the original content segmentation model are adjusted to improve the segmentation and recognition ability of the original content segmentation model for the report content part and the conclusion part, and the target content segmentation model is obtained.

[0152] Reference Figure 7 , corresponding to the above-mentioned multi-round dialogue scenario, in some embodiments, step S107 includes but is not limited to steps S701 to S703.

[0153] Step S701, determining report data associated with a target inspection report according to a preset association condition;

[0154] Step S702, combining the report data with the target inspection report to obtain target processing data;

[0155] Step S703: the target conclusion generation model performs conclusion generation processing on the target processing data to obtain a target report conclusion.

[0156] In step S701 of some embodiments, the preset association condition may include a time condition. The time condition may refer to the time difference from the acquisition of the target inspection report being less than a preset time threshold. The specific value of the preset time threshold may be adaptively set according to actual needs, such as the preset time threshold may be set to 10 minutes, which is not specifically limited in the embodiments of the present application. For example, in Figure 8 In the multi-round dialogue scenario shown, there are a question object 801 and an answer object 802. The answer object can generate a corresponding report conclusion based on the method provided in the embodiment of the present application. Figure 8 The scenario shown may include a first round of dialogue 803 and a second round of dialogue 804. In the first round of dialogue 803, the questioner 801 may send target inspection report 1 and conclusion requirement data "Please help me take a look at the following report", and the answerer 802 may generate a report conclusion corresponding to the target inspection report 1 according to the conclusion requirement data. In the second round of dialogue 804, the questioner 801 may send target inspection report 2 and conclusion requirement data "Please help me take a look at this report again". If the time when the answerer 802 obtains the target inspection report 2 is less than the time difference between the time when the answerer 802 obtains the target inspection report 2 and the time when the answerer 802 obtains the target inspection report 1 is less than the preset time threshold, the target inspection report 1 may be determined as the report data associated with the target inspection report 2.

[0157] According to actual needs, the association conditions may also include content association conditions, address association conditions, etc., which are not specifically limited in the embodiments of the present application.

[0158] In step S702 of some embodiments, the report data is spliced ​​with the target inspection report, such as merging the target inspection report 1 and the target inspection report 2 to obtain the target processing data.

[0159] In step S703 of some embodiments, the target processing data is used as input data of the target conclusion generation model, and conclusion generation processing is performed according to the target conclusion generation model to obtain the target report conclusion.

[0160] The benefit of steps S701 to S703 is that the target inspection report can be analyzed in combination with the associated report data, thereby improving the accuracy and richness of the target report conclusions.

[0161] The method for generating report conclusions provided in the embodiment of the present application generates a predicted report conclusion corresponding to the first sample inspection content through the original conclusion generation model. The original connector is parameter adjusted through the predicted report conclusion and the first sample report conclusion, and the original encoder and the original decoder are respectively trained to freeze parameters, so that the target conclusion generation model can be obtained. The target inspection report is concluded according to the target conclusion generation model to obtain the corresponding target report conclusion. It can be seen that the embodiment of the present application can directly generate conclusions based on the target conclusion generation model, reducing the situation in which the quality of conclusion generation in the related art depends on the quality of the object question, that is, the embodiment of the present application is insensitive to the problem. Therefore, the embodiment of the present application can improve the universality of report conclusions based on report generation. In addition, compared with the scenario in which the conclusion generation model in the related art is a classification model and can only output a limited number of category conclusions, the target report conclusion of the embodiment of the present application is derived based on natural language, thereby improving the richness of the target report conclusion.

[0162] See also Fig. 9 The embodiment of the present application also provides a device for generating a report conclusion, which can implement the above-mentioned method for generating a report conclusion, and the device includes:

[0163] An inspection report acquisition module 901 is used to acquire the first sample inspection content and the first sample report conclusion of the first sample inspection report;

[0164] A data input module 902 is used to input the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector and an original decoder;

[0165] The encoding module 903 is used to encode the first sample inspection content according to the original encoder to obtain content features;

[0166] A decoding module 904 is used to map the content features to the original decoder according to the original connector, decode the content features according to the original decoder, and obtain a prediction report conclusion;

[0167] The model training module 905 is used to adjust the parameters of the original connector according to the conclusion of the prediction report and the conclusion of the first sample report to obtain a target connector; perform parameter freezing training on the original encoder to obtain a target encoder, and perform parameter freezing training on the original decoder to obtain a target decoder;

[0168] A model building module 906 is used to obtain a target conclusion generation model according to the target encoder, the target connector and the target decoder;

[0169] The conclusion generation module 907 is used to perform conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion.

[0170] The specific implementation of the device for generating the report conclusion is basically the same as the specific implementation of the method for generating the report conclusion described above, and will not be described in detail here.

[0171] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method for generating the above report conclusion when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0172] See also Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0173] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an 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;

[0174] The memory 1002 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 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the method for generating the report conclusion of the embodiment of this application;

[0175] Input / output interface 1003, used to implement information input and output;

[0176] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WI FI, Bluetooth, etc.);

[0177] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0178] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0179] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the method for generating the above-mentioned report conclusion is implemented.

[0180] 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 disposed 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.

[0181] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present 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 the present application are also applicable to similar technical problems.

[0182] Those skilled in the art will appreciate 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.

[0183] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0184] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0185] 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 sequence. 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 of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising 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.

[0186] It should be understood that in the present 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 objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single 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.

[0187] In the several embodiments provided in the present 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 only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, 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.

[0188] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0189] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0190] 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 is essentially 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, and the computer software product is stored in a storage medium, including multiple instructions to enable 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: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0191] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for generating a report conclusion, characterized in that: The method comprises: Obtain the first sample inspection content and the first sample report conclusion of the first sample inspection report; Inputting the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector, and an original decoder; encoding the first sample inspection content according to the original encoder to obtain content features; Mapping the content features to the original decoder according to the original connector, decoding the content features according to the original decoder, and obtaining a prediction report conclusion; According to the conclusion of the prediction report and the conclusion of the first sample report, the parameters of the original connector are adjusted to obtain a target connector; the parameters of the original encoder are frozen and trained to obtain a target encoder, and the parameters of the original decoder are frozen and trained to obtain a target decoder; Obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder; The acquired target inspection report is processed for conclusion generation according to the target conclusion generation model to obtain the target report conclusion.

2. The method according to claim 1, characterized in that adjusting parameters of the original connector according to the prediction report conclusion and the first sample report conclusion to obtain a target connector; The method comprises: performing parameter freezing training on the original encoder to obtain a target encoder, and performing parameter freezing training on the original decoder to obtain a target decoder, comprising: Freezing model parameters of the original encoder to obtain a frozen encoder, and freezing model parameters of the original decoder to obtain a frozen decoder; Adjusting parameters of the original connector according to the conclusion of the prediction report and the conclusion of the first sample report to obtain the target connector; Thawing the frozen encoder to obtain a thawed encoder, and adjusting parameters of the frozen encoder according to the prediction report conclusion and the first sample report conclusion to obtain the target encoder; The frozen decoder is unfrozen to obtain a target decoder.

3. The method according to claim 2, characterized in that The step of unfreezing the frozen decoder to obtain a target decoder comprises: Thawing the frozen decoder to obtain a thawed decoder; Obtaining the model type of the unfrozen decoder; If the model type is the same as the preset type, using the unfrozen decoder as the target decoder; If the model type is different from the preset type, parameters of the unfrozen decoder are adjusted according to the prediction report conclusion and the first sample report conclusion to obtain the target decoder.

4. The method according to claim 1, characterized in that: The obtaining of the first sample inspection content and the first sample report conclusion of the first sample inspection report includes: Performing content segmentation on the first sample inspection report based on a pre-trained target content segmentation model to obtain at least two report images and the first sample report conclusion; Data splicing is performed on all the report images according to the pre-constructed single conclusion requirement data to obtain the first sample inspection content.

5. The method according to claim 4, characterized in that Before performing content segmentation on the sample inspection report based on the preset target content segmentation model, the method further includes training the target content segmentation model, including: Obtaining a second sample inspection report, wherein the second sample inspection report includes a sample image; Performing a conclusion query on the second sample inspection report to obtain a conclusion query result; If the conclusion query result indicates that the second sample inspection report does not include a report conclusion, fill in the conclusion of the second sample inspection report to obtain a second sample report conclusion; Performing content segmentation on the second sample inspection report according to a preset original content segmentation model to obtain a predicted image and a predicted report conclusion; The parameters of the original content segmentation model are adjusted according to the predicted image, the prediction report conclusion, the second sample report conclusion and the sample image to obtain the target content segmentation model.

6. The method according to claim 1, characterized in that The obtaining of the first sample inspection content and the first sample report conclusion of the first sample inspection report includes: Randomly sampling the first sample inspection report to obtain a candidate report; Performing content segmentation on the candidate report based on a pre-trained target content segmentation model to obtain candidate images and candidate report conclusions; Performing data splicing on the pre-constructed multi-conclusion requirement data and the candidate image to obtain the first sample inspection content; The pre-constructed conclusion connection data and the candidate report conclusion are concatenated to obtain the first sample report conclusion.

7. The method according to any one of claims 1 to 6, characterized in that: The step of performing conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion includes: Determining report data associated with the target inspection report according to a preset association condition; Performing data splicing on the report data and the target inspection report to obtain target processing data; The target processing data is processed for conclusion generation according to the target conclusion generation model to obtain the target report conclusion.

8. A device for generating report conclusions, characterized in that: The device comprises: An inspection report acquisition module, used to obtain the first sample inspection content and the first sample report conclusion of the first sample inspection report; A data input module, used for inputting the first sample inspection content into a preset original conclusion generation model, wherein the original conclusion generation model includes an original encoder, an original connector and an original decoder; An encoding module, used for encoding the first sample inspection content according to the original encoder to obtain content features; A decoding module, used for mapping the content features to the original decoder according to the original connector, decoding the content features according to the original decoder, and obtaining a prediction report conclusion; A model training module, used to adjust the parameters of the original connector according to the prediction report conclusion and the first sample report conclusion to obtain a target connector; perform parameter freezing training on the original encoder to obtain a target encoder, and perform parameter freezing training on the original decoder to obtain a target decoder; A model building module, used for obtaining a target conclusion generation model according to the target encoder, the target connector and the target decoder; The conclusion generation module is used to perform conclusion generation processing on the acquired target inspection report according to the target conclusion generation model to obtain the target report conclusion.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images