Medical examination report processing method and device, storage medium and electronic equipment

Abstract

The invention discloses a medical examination report processing method and device, a storage medium and electronic equipment. The method relates to the technical field of medical treatment, and comprises the following steps: obtaining a to-be-processed target medical examination report; performing term extraction on the target medical examination report to obtain target term information in the medical examination report; and according to the target term information and the target inference map, carrying out quality inspection on the target medical examination report to obtain a quality inspection result. Through the method and the device, the problem that the interpretability of the quality inspection result of the medical examination report is relatively poor due to the fact that the machine learning model is a predictive end-to-end black box when quality inspection is performed on the medical examination report through the machine learning model in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a method and device for processing a medical examination report, a storage medium, and an electronic device. Background Art

[0002] Medical examination reports are one of the important bases for diagnosis. In the business of writing medical examination reports, a basic and important task is to conduct quality control on the findings and conclusions in the medical report. After the doctor completes the initial writing of the report, there will be a physician responsible for verification to check the report one by one in several dimensions, such as "lack of diagnosis corresponding to the findings", "lack of diagnosis corresponding to the findings", "conflict between measurement values ​​and diagnosis", and "lack of diagnosis corresponding to the measurement values". After confirmation, it will be provided to the physician and the patient. For example: In the cardiac ultrasound report, the findings mentioned that the pulmonary artery diastolic pressure was 39 mmHg, but the conclusion mistakenly gave a diagnosis of severe pulmonary artery hypertension, which is a "conflict between measurement values ​​and diagnosis" type error, and it should actually be mild pulmonary artery hypertension. In current actual business scenarios, medical examination reports are often quality-checked based on machine learning models, but this method lacks explainability and the effect cannot be controlled.

[0003] Regarding the related technology of using machine learning models to perform quality inspection on medical examination reports, since the machine learning model is an end-to-end black box prediction, which leads to the problem of poor interpretability of the quality inspection results of the medical examination reports, no effective solution has been proposed yet. Summary of the invention

[0004] The main purpose of the present application is to provide a method and device for processing medical examination reports, a storage medium and an electronic device, so as to solve the problem in the related art of quality inspection of medical examination reports through machine learning models. Since the machine learning model is a predictive end-to-end black box, the interpretability of the quality inspection results of the medical examination reports is relatively poor.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for processing medical examination reports is provided. The method comprises: obtaining a target medical examination report to be processed; performing term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; performing quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnosis opinion, and the term information corresponding to what is seen in the historical medical examination report and term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination report by a target model, and the target model is one of the following: an entity extraction model and a term generation model.

[0006] Furthermore, before performing quality inspection on the medical examination report based on the target term information and the target reasoning graph to obtain the quality inspection result, the method also includes: obtaining multiple first historical medical examination reports; performing term extraction on the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnosis opinion in the first historical medical examination report; and constructing the target reasoning graph based on the historical term information.

[0007] Furthermore, term extraction is performed on the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, including: obtaining multiple second historical medical examination reports and constructing a target training set based on the multiple second historical medical examination reports; training an initial model based on the target training set to obtain a target model; and term extraction is performed on the multiple first historical medical examination reports based on the target model to obtain the historical term information.

[0008] Furthermore, constructing a target training set based on the multiple second historical medical examination reports includes: if the target model is an entity extraction model, labeling the information in the multiple second historical medical examination reports according to a preset medical entity relationship database to obtain multiple third historical medical examination reports, and constructing the target training set based on the multiple third historical medical examination reports; if the target model is a term generation model, labeling the terms in the multiple second historical medical examination reports to obtain a target sequence, and constructing the target training set based on the target sequence, wherein the target sequence is composed of term information corresponding to the findings in the second historical medical examination report and term information corresponding to the diagnostic opinion in the second historical medical examination report.

[0009] Furthermore, if the target model is an entity extraction model, term extraction is performed on the multiple first historical medical examination reports based on the target model to obtain the historical term information, including: extracting the entities and the relationships between entities in the first historical medical examination reports through the entity extraction model to obtain entity relationship data; obtaining a preset term extraction library, and performing term extraction on the entity relationship data based on the term extraction library to obtain the historical term information.

[0010] Furthermore, if the target model is a terminology generation model, terminology is extracted from the multiple first historical medical examination reports based on the target model to obtain the historical terminology information, including: preprocessing the multiple first historical medical examination reports to obtain processed first historical medical examination reports; performing feature extraction on the processed first historical medical examination reports through the target model to obtain target feature information; and obtaining the historical terminology information based on the target feature information.

[0011] Furthermore, obtaining a preset term extraction library includes: obtaining a plurality of fourth historical medical examination reports, and obtaining term information corresponding to findings in the fourth historical medical examination reports and term information corresponding to diagnostic opinions in the fourth historical medical examination reports; preprocessing the term information corresponding to findings in the fourth historical medical examination reports and the term information corresponding to diagnostic opinions in the fourth historical medical examination reports to obtain processed term information; and obtaining the preset term extraction library based on the processed term information.

[0012] Furthermore, constructing the target reasoning graph based on the historical term information includes: obtaining a normalized vocabulary, and normalizing the historical term information based on the normalized vocabulary to obtain processed historical term information; determining, based on the processed historical term information, first term information corresponding to what is seen in the first historical medical examination report and second term information corresponding to the diagnosis opinion in the first historical medical examination report; determining a target reasoning link between the first term information corresponding to what is seen and the second term information corresponding to the diagnosis opinion; and constructing the target reasoning graph based on the target reasoning link.

[0013] Furthermore, determining the target reasoning link between the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion includes: for each second term information, determining the total number of times each first term information and the second term information appear in the same first historical medical examination report; based on the total number of times and a preset threshold, determining the first term information matched by the second term information; and determining the target reasoning link based on the second term information and the first term information matched by the second term information.

[0014] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a device for processing medical examination reports is provided. The device includes: a first acquisition unit, which is used to acquire a target medical examination report to be processed; an extraction unit, which is used to perform term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; a quality inspection unit, which is used to perform quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by the term information corresponding to what is seen in the historical medical examination report and the term information corresponding to the diagnosis opinion, and the term information corresponding to what is seen in the historical medical examination report and the term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination report by a target model, and the target model is one of the following: an entity extraction model and a term generation model.

[0015] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a computer-readable storage medium is provided, which stores a program, wherein when the program is running, the device where the storage medium is located is controlled to execute any one of the above-mentioned methods for processing medical examination reports.

[0016] In order to achieve the above-mentioned purpose, according to another aspect of the present application, an electronic device is also provided, the electronic device includes one or more processors and a memory, and the memory is used to store the one or more processors to implement the processing method of the medical examination report described in any one of the above.

[0017] Through the present application, the following steps are adopted: obtaining a target medical examination report to be processed; performing term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnostic opinion in the target medical examination report; performing quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnostic opinion, and the term information corresponding to what is seen in the historical medical examination report and term information corresponding to the diagnostic opinion are obtained by processing the historical medical examination report by a target model, and the target model is one of the following: an entity extraction model and a term generation model, which solves the problem in the related art of quality inspection of medical examination reports through machine learning models, and since the machine learning model is a prediction end-to-end black box, the problem of poor interpretability of the quality inspection results of the medical examination report is caused. In this scheme, the terminology information in the medical examination report is extracted to obtain the target terminology information, and then the target medical examination report is quality-inspected based on the target terminology information through the target reasoning graph to obtain the final quality inspection result. The quality inspection report obtained through the target reasoning graph is interpretable, and the target reasoning graph is constructed based on the terminology information corresponding to the actual historical medical examination reports and the terminology information corresponding to the diagnostic opinions. It can adapt to the characteristics of the medical report itself, thereby achieving the effect of improving the accuracy of quality inspection of medical examination reports. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 The process of the method for processing medical examination reports provided in the embodiment of the present application is as follows Figure 1 ; Figure 2 is a schematic diagram of extracting term information according to an embodiment of the present application; Figure 3 It is a schematic diagram of constructing a target reasoning graph according to an embodiment of the present application; Figure 4 The process of the method for processing medical examination reports provided in the embodiment of the present application is as follows Figure 2 ; Figure 5 is a schematic diagram of a medical examination report processing device provided according to an embodiment of the present application; Figure 6 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims 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 interchanged where appropriate, so that the embodiments of the present application described here. 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.

[0022] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0023] The present invention is described below in conjunction with preferred implementation steps. Figure 1 The process of the method for processing medical examination reports provided in the embodiment of the present application is as follows Figure 1 ,like Figure 1 As shown, the method comprises the following steps:

[0024] Step S101, obtaining a target medical examination report to be processed.

[0025] Optionally, a target medical examination report that needs to be quality-checked is obtained. It should be noted that the target medical examination report can be a report written by a doctor in real time or a report written in non-real time.

[0026] Step S102, extracting terms from the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to findings in the target medical examination report and term information corresponding to diagnostic opinions in the target medical examination report.

[0027] Optionally, the term information corresponding to the findings in the target medical examination report and the term information corresponding to the diagnosis opinion in the target medical examination report are processed to obtain the above-mentioned target term information.

[0028] For example, the term information corresponding to the findings may include decreased contraction amplitude of each segment of the left ventricular wall, normal relationship and inner diameter of the great arteries; the term information corresponding to the conclusion may include severe pulmonary hypertension, moderate mitral valve regurgitation, small amount of aortic valve regurgitation, etc.

[0029] Step S103, performing quality inspection on the target medical examination report based on the target terminology information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by the terminology information corresponding to what is found in the historical medical examination report and the terminology information corresponding to the diagnosis opinion, and the terminology information corresponding to what is found in the historical medical examination report and the terminology information corresponding to the diagnosis opinion are obtained by processing the historical medical examination report by a target model, and the target model is one of the following: an entity extraction model, a terminology generation model.

[0030] Optionally, after obtaining the above-mentioned target term information, the terms seen in the target term information and in the conclusion can also be normalized through the term normalization vocabulary to obtain the normalized terms seen and in the conclusion. It should be noted that the aforementioned term normalization vocabulary can be constructed based on the terms extracted from the historical medical examination report and then merged. The target reasoning graph is constructed by constructing the term information corresponding to the findings in the historical medical examination report and the term information corresponding to the diagnostic opinion, and the target term information is used as an input parameter to call the target reasoning graph to obtain the quality control result. It should be noted that the target reasoning graph can realize the quality control reasoning of four dimensions of "lack of diagnosis corresponding to findings", "lack of diagnosis corresponding to findings", "conflict between measurement value and diagnosis", and "lack of diagnosis corresponding to measurement value", and show the problems in the report to the doctor through the front-end interface.

[0031] In an optional embodiment, the quality control results in two different fields are as follows. For the convenience of description, the corresponding term information seen in the report and the term information corresponding to the conclusion are directly given below: (1) Cardiac ultrasound report Corresponding term information: The contraction amplitude of each segment of the left ventricular wall is reduced. The relationship and inner diameter of the great arteries are normal. The coronary sinus is widened, and the left superior vena cava is detected. No abnormality is found in the pericardial cavity. Doppler examination: moderate mitral regurgitation, small amount of aortic regurgitation. Pulmonary artery systolic pressure is 26mmHg; left superior vena cava blood flow signal is detected.

[0032] The corresponding term information of the conclusion: severe pulmonary hypertension, moderate mitral regurgitation, small amount of aortic regurgitation The quality control results include quality inspection problems and their causes. Quality inspection problem: "Measurement value conflicts with diagnosis", cause: pulmonary artery systolic pressure <30 mmHg-> mild pulmonary artery hypertension; "Lack of diagnosis corresponding to findings", reason: reduced contraction amplitude of each segment of the left ventricular wall -> segmental wall motion abnormality; (2) Chest X-ray report Corresponding term information: The texture of both lungs was normal, no consolidation was found; the ascending aorta was widened, the pulmonary artery segment was straight, the heart shadow was enlarged, and the left heart was significantly enlarged; the heart-to-chest ratio was 0.74; The term information corresponding to the conclusion: widening of the ascending aorta; Quality inspection issue: "Lack of corresponding diagnosis for findings", reason: left heart enlargement -> left heart enlargement.

[0033] It should be noted that the target reasoning graph is constructed by the term information corresponding to the findings in the historical medical examination report and the term information corresponding to the diagnosis opinion. The term information corresponding to the findings in the historical medical examination report and the term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination report by the target model. The target model can be the above-mentioned entity extraction model or the above-mentioned term generation model. It should be noted that the historical medical examination report can be the first historical medical examination report, the second historical medical examination report, or a historical medical examination report other than the first historical medical examination report and the second historical medical examination report.

[0034] It should be noted that the above-mentioned entity extraction model is trained by historical medical examination reports and the real entity relationship labels corresponding to the historical medical examination reports, and the above-mentioned terminology generation model is trained by historical medical examination reports and the target sequences corresponding to the historical medical examination reports. The target sequence is composed of the real terminology information corresponding to the findings in the historical medical examination reports and the real terminology information corresponding to the diagnostic opinions in the historical medical examination reports.

[0035] It should be noted that the entity extraction model can be a Global Pointer model, which is a model used to simulate the reference resolution task in natural language understanding. In natural language processing (NLP), the goal of the reference resolution task is to identify entities (such as people, places, things, etc.) mentioned in the text and link them to corresponding referents or descriptors. The term generation model can be a deep learning model, including a feature extraction layer, a fully connected layer, and an output layer.

[0036] It should be noted that in step S102, the entity extraction model can be used to first perform entity recognition relationship extraction on the target medical examination report to obtain information such as entities and relationships in the target medical examination report. After obtaining the entities and relationships in the target medical examination report, the meta-path terms, i.e., the above-mentioned target term information, are obtained from the entities and relationships through the preset term extraction template library. The target medical examination report can also be directly processed through the term generation model to generate the term content in the report and conclusion, i.e., the above-mentioned target term information. The meta-path terms in the target medical examination report can be quickly and accurately obtained through the term generation model and the entity extraction model.

[0037] To summarize, the terminology information in the medical examination report is extracted to obtain the target terminology information, and then the target medical examination report is quality inspected based on the target terminology information through the target reasoning graph to obtain the final quality inspection result. The quality inspection report obtained through the target reasoning graph is more interpretable, and the target reasoning graph is constructed based on the terminology information corresponding to the actual historical medical examination report and the terminology information corresponding to the diagnostic opinion, which can adapt to the characteristics of the medical report itself, thereby achieving the effect of improving the accuracy of quality inspection of medical examination reports.

[0038] Optionally, in the method for processing medical examination reports provided in the embodiment of the present application, before performing quality inspection on the medical examination report based on the target term information and the target reasoning graph to obtain the quality inspection result, the method also includes: obtaining multiple first historical medical examination reports; performing term extraction on the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnostic opinion in the first historical medical examination report; and constructing a target reasoning graph based on the historical term information.

[0039] In an optional embodiment, the following steps are used to construct the above-mentioned target reasoning graph: first, multiple existing historical medical examination reports (i.e., the above-mentioned first historical medical examination report) are obtained, and then the term information in the first historical medical examination report is extracted to obtain meta-path terms (i.e., the above-mentioned historical term information), and the above-mentioned target reasoning graph is constructed through the extracted meta-path terms.

[0040] Constructing the graph through true meta-path terms improves the accuracy of the target reasoning graph.

[0041] Optionally, in the medical examination report processing method provided in the embodiment of the present application, term extraction is performed on multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, including: obtaining multiple second historical medical examination reports, and constructing a target training set based on the multiple second historical medical examination reports; training an initial model based on the target training set to obtain a target model; and term extraction is performed on multiple first historical medical examination reports based on the target model to obtain historical term information.

[0042] In an optional embodiment, in order to improve the accuracy of term extraction, the following steps are used to implement term extraction from the first historical medical examination report: first, multiple existing historical medical examination reports (i.e., the second historical medical examination report mentioned above) are obtained. It should be noted that the second historical medical examination report and the first historical medical examination report may be the same or different. Then, a target training set is constructed through the second historical medical examination report, and the target model is obtained by training and iterating the initial model through the target training set constructed by the historical medical examination report, so that the term extraction from the first historical medical examination report can be implemented through the target model in the future, thereby obtaining the above-mentioned historical term information.

[0043] It should be noted that the initial model can be an entity extraction model framework or a term generation model framework. The target model is obtained by training and iterating the initial model with the target training set constructed from historical medical examination reports. The target model can be an entity extraction model or a term generation model. If the entity extraction model is used, the entities, different entities and relationships between entities in the historical medical examination report are first extracted through the entity extraction model, and then the historical term information corresponding to the historical medical examination report is obtained based on the term extraction library. If the term generation model is used, the historical medical examination report is input into the term generation model, and the historical term information is output through the term generation model.

[0044] Term extraction through the target model can effectively improve the efficiency of term extraction and avoid the problem of low accuracy in term extraction based on manual experience.

[0045] Optionally, in the medical examination report processing method provided in the embodiment of the present application, constructing a target training set based on multiple second historical medical examination reports includes: if the target model is an entity extraction model, then labeling the information in the multiple second historical medical examination reports according to a preset medical entity relationship database to obtain multiple third historical medical examination reports, and constructing a target training set based on the multiple third historical medical examination reports; if the target model is a terminology generation model, then labeling the terms in the multiple second historical medical examination reports to obtain a target sequence, and constructing a target training set based on the target sequence, wherein the target sequence is composed of terminology information corresponding to the findings in the second historical medical examination reports and terminology information corresponding to the diagnostic opinions in the second historical medical examination reports.

[0046] In an optional embodiment, since the target model can be an entity extraction model or a term generation model, the corresponding target training set constructed is also different, that is, if the target model is an entity extraction model, then first summarize and determine the schema system of entities and dependencies based on the historical patient report data, that is, the entities in the knowledge graph and the entity relationships between different entities, and obtaining the entity relationships between entities in the knowledge graph and different entities is a prior art. Then, according to these entities and the entity relationships between different entities, the information in multiple second historical medical examination reports is annotated to obtain multiple third historical medical examination reports, and then the target training set is constructed through multiple third historical medical examination reports with entities and entities between different entities. It should be noted that the third historical medical examination report includes multiple patient samples, and each patient sample includes information such as entities and entity relationships between different entities.

[0047] If the target model is a term generation model, the terms in the findings and conclusions of the second historical medical examination report are annotated to construct [found + conclusion -> term content], where there are multiple second historical medical examination reports, and each second historical medical examination report has multiple findings and conclusions, and the terms corresponding to multiple findings and conclusions + findings and conclusions constitute a sequence, so a sequence-to-sequence training set is constructed, that is, the terms in multiple second historical medical examination reports are annotated to obtain a target sequence, and then a target training set is constructed based on the target sequence.

[0048] Constructing different training sample sets according to different models can improve the performance of subsequent models in term extraction, thereby improving the accuracy of subsequent graph construction.

[0049] Optionally, in the medical examination report processing method provided in the embodiment of the present application, if the target model is an entity extraction model, term extraction is performed on multiple first historical medical examination reports based on the target model to obtain historical term information, including: extracting entities and relationships between entities in the first historical medical examination reports through the entity extraction model to obtain entity relationship data; obtaining a preset term extraction library, and performing term extraction on the entity relationship data based on the term extraction library to obtain historical term information.

[0050] In an optional embodiment, when the target model is an entity extraction model, the following steps are used to obtain the above-mentioned historical terminology information: the first historical medical examination report is input into the entity extraction model, and the entities and the relationships between entities in the first historical medical examination report are extracted through the entity extraction model to obtain entity relationship data.

[0051] Then, a preset term extraction library is obtained, and term extraction is performed on the entity relationship data based on the term extraction library to obtain historical term information. It should be noted that the term extraction library is constructed using real term information in historical medical examination reports.

[0052] In an optional embodiment, the following may be used: Figure 2 The schematic diagram shown implements the extraction of term information: obtaining historical medical examination reports, e.g. Figure 2 The ultrasound report stated, "The inner diameters of the atrioventricular cavities are generally normal, the thickness of the ventricular septum and each segment of the left ventricular wall is normal, the contraction amplitude is normal, the movement is coordinated, the contraction amplitude is normal, and there are no obvious changes in the morphology, structure, and opening and closing movements of the valves. The relationship and inner diameter of the great arteries are normal. The coronary sinus is widened, and the left superior vena cava is detected. No abnormalities were found in the pericardial cavity. Doppler examination showed moderate mitral valve regurgitation and a small amount of aortic valve regurgitation. The pulmonary artery systolic pressure is 26 mmHg; the left superior vena cava blood flow signal was detected."

[0053] Perform entity extraction and relationship extraction on the contents in the ultrasound report to obtain entity relationship data, for example, the entity "Location: each atrioventricular cavity; atrioventricular cavity; pulmonary artery; pathophysiological changes: generally normal; moderate reflux; pathophysiology: inner diameter; relationship; thickness; systolic pressure..; polarity: not seen...; measurement: 26mmHg...;" for example, the relationship between entities "{[aorta, location], [relationship, pathophysiology]}; {[relationship, pathophysiology], [normal, pathophysiological changes]}; {[pulmonary artery, location], [systolic pressure, pathophysiology]}; {[systolic pressure, pathophysiology], [130mmHg, measurement]}.

[0054] Finally, meta-path term extraction is performed on the obtained entities and relations through the term extraction library. It should be noted that the term extraction library includes multiple meta-paths. By calling the term extraction library (term template library) and extracting meta-path terms from entities and relations, meta-path terms can be obtained quickly and accurately. For example, meta-path 1: location-pathophysiology-pathophysiological changes, meta-path terms: normal relationship between the great arteries, normal inner diameter of the great arteries, coordinated ventricular septum movement, normal ventricular septum contraction amplitude; meta-path 2: location-pathophysiology-pathophysiological changes-polarity, meta-path terms: obvious changes in the opening and closing movement of the ventricular septum (not seen), obvious changes in the opening and closing movement of each segment of the left ventricular wall (not seen); meta-path 3: location-pathophysiological changes, meta-path terms: widening of the coronary sinus, moderate mitral regurgitation; meta-path 4: location-pathophysiological changes-polarity; meta-path terms: abnormal pericardial cavity (not seen).

[0055] The entity extraction model and term extraction library can be used to automatically extract terms in batches, thereby improving the efficiency of term acquisition.

[0056] In an optional embodiment, obtaining a preset term extraction library includes: obtaining multiple fourth historical medical examination reports, and obtaining term information corresponding to findings in the fourth historical medical examination reports and term information corresponding to diagnostic opinions in the fourth historical medical examination reports; preprocessing the term information corresponding to findings in the fourth historical medical examination reports and the term information corresponding to diagnostic opinions in multiple third historical medical examination reports to obtain processed term information; and obtaining a preset term extraction library based on the processed term information.

[0057] In an optional embodiment, the above-mentioned term extraction library is constructed by the following steps: a plurality of historical medical examination reports (i.e., the fourth historical medical examination report) is obtained, and then the term information corresponding to the findings in the fourth historical medical examination report and the term information corresponding to the diagnostic opinion in the fourth historical medical examination report can be determined based on prior knowledge, that is, based on prior knowledge, the meta-path between the entities in the fourth historical medical examination report and the entity relationships between different entities is determined, for example, meta-path 1: part-pathophysiology-pathophysiological changes. In order to improve the accuracy of the term information, the term information corresponding to the findings and the term information corresponding to the diagnostic opinion can be pre-processed, for example, filtering and quality inspection, and then the above-mentioned processed term information is obtained. Finally, the term extraction template library (i.e., the above-mentioned term extraction library) is constructed through the processed term information.

[0058] Extracting terms from entity relationship data through a term extraction library can effectively improve the accuracy of term extraction.

[0059] Optionally, in the medical examination report processing method provided in the embodiment of the present application, if the target model is a terminology generation model, terminology is extracted from multiple first historical medical examination reports based on the target model to obtain historical terminology information, including: preprocessing the multiple first historical medical examination reports to obtain processed first historical medical examination reports; performing feature extraction on the processed first historical medical examination reports through the target model to obtain target feature information; and obtaining historical terminology information based on the target feature information.

[0060] In an optional embodiment, if the target model is a terminology generation model, the following steps are used to obtain the above-mentioned historical terminology information: first, a plurality of first historical medical examination reports are preprocessed, for example, preprocessing such as trimming and removing invalid data, and then the processed first historical medical examination reports are input into the terminology generation model (i.e., the above-mentioned target model), and features of the first historical medical examination reports are extracted through the terminology generation model, and then the above-mentioned historical terminology information is obtained through the target feature information.

[0061] By using the term generation model to generate terms as a sequence-to-sequence deep model, the end-to-end generation method is more convenient and efficient.

[0062] Optionally, in the medical examination report processing method provided in the embodiment of the present application, constructing a target reasoning graph based on historical terminology information includes: obtaining a normalized vocabulary, and normalizing the historical terminology information based on the normalized vocabulary to obtain processed historical terminology information; determining, based on the processed historical terminology information, first terminology information corresponding to what is seen in the first historical medical examination report and second terminology information corresponding to the diagnosis opinion in the first historical medical examination report; determining a target reasoning link between the first terminology information corresponding to what is seen and the second terminology information corresponding to the diagnosis opinion; and constructing a target reasoning graph based on the target reasoning link.

[0063] In an optional embodiment, in order to improve the accuracy of constructing the target reasoning graph, a normalized vocabulary is constructed, and then the historical term information is normalized through the normalized vocabulary, for example, 'mitral regurgitation' and 'mitral regurgitation' are unified as 'mitral regurgitation'.

[0064] Then, based on the normalized historical term information (i.e., the processed historical term information mentioned above), distinguish the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion, and then determine the target reasoning link between the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion. For example, the reasoning link between the first term information and the second term information is determined based on expert experience. Finally, a target reasoning graph is constructed based on the target reasoning link. Based on the graph, the reasoning quality control service is developed, and the quality control output dimensions are as follows: a. "The diagnosis corresponding to the observation is missing"; b. "The diagnosis corresponding to the observation is missing"; c. "The measurement value conflicts with the diagnosis"; d. "The diagnosis corresponding to the measurement value is missing".

[0065] Through the above-mentioned target reasoning graph, medical examination reports can be accurately quality-checked based on the reasoning link, and error prompts can be provided, which is more explainable.

[0066] Optionally, in the medical examination report processing method provided in the embodiment of the present application, determining the target reasoning link between the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion includes: for each second term information, determining the total number of times each first term information and the second term information appear in the same first historical medical examination report; based on the total number of times and a preset threshold, determining the first term information matched by the second term information; and determining the target reasoning link based on the second term information and the first term information matched by the second term information.

[0067] In an optional embodiment, in order to improve the accuracy of the reasoning link between the first term information and the second term information, the co-occurrence relationship between the observed terms and the conclusion terms can be counted and arranged from high to low based on frequency, that is, the total number of times each first term information and second term information appears in the same first historical medical examination report is determined as mentioned above.

[0068] Then, according to the co-occurrence frequency and the set threshold, the reasonable logic is screened to obtain the screened reasoning link, that is, the first term information matched by the second term information is determined based on the total number of times and the preset threshold. Finally, the target reasoning link is determined by the first term information matched by the second term information.

[0069] By analyzing the co-occurrence relationship between terms after statistical normalization, doctors can save themselves the process of summarizing and generalizing. Based on frequency sorting, they can efficiently find the reasoning relationship contained in the data and improve the accuracy of subsequent construction of the target reasoning map.

[0070] In an optional embodiment, the Figure 3 The schematic diagram shown realizes the construction of the target reasoning graph: (1) Preliminary preparation: Solution 1: Based on historical patient report data, summarize and determine the schema system of entities and dependencies, that is, the entities in the knowledge graph and the entity relationships between different entities; based on prior knowledge, determine the term composition meta-path system and build a term extraction template library.

[0071] Solution 2: Label the terms in the report’s observations and conclusions, and construct a sequence-to-sequence training set of [observations + conclusions -> term content].

[0072] (2) Term extraction: Solution 1: (Reporting entities, relationship extraction, meta-path) According to the schema system obtained in the preliminary preparation plan 1, the entity relationships in the report and conclusion text are annotated, and the historical patient report data is processed to construct training data. The training data includes multiple samples, each of which includes entities and entity relationships between different entities; Based on the global pointer model (i.e. the entity extraction model mentioned above), the entity and relationship extraction models are trained using training data; entity and relationship extraction are performed on the findings and conclusion texts of historical patient reports to obtain the parsed content of the report; meta-path term extraction is performed based on the parsed content of the report and the term extraction template library to extract the term content in the findings and conclusions, and extract meta-path terms. The meta-path terms include report-connotation diagnostic terms, report-connotation findings terms, and report-connotation measurement value terms.

[0073] Solution 2: (Generative Term Extraction) Based on the sequence-to-sequence training data annotated in Scheme 2 in the preliminary preparation, the sequence-to-sequence generation model (i.e., the term generation model mentioned above) is trained; the historical patient report findings + conclusions are used as the input of the generation model, and the report-connoted diagnostic terms, report-connoted findings terms, and report-connoted measurement value terms are output.

[0074] (3) Normalized vocabulary construction: Merge the extracted terms to construct a normalized vocabulary, which is then imported into the preset graph database for maintenance.

[0075] (4) Mining the observation-conclusion reasoning link: According to the normalized vocabulary, the terms in the patient's historical report are normalized; for example, "mitral regurgitation" and "mitral regurgitation" are unified as "mitral regurgitation"; the co-occurrence relationship between the observed terms and the conclusion terms is counted and arranged from high to low based on frequency; according to the co-occurrence frequency and the set threshold, the reasonable logic is screened to obtain the screened reasoning links.

[0076] (5) Quality control map construction: Based on the screened reasoning links, they are imported into the preset graph database (neo4j) for maintenance to form a quality control map. Based on the map, reasoning quality control services are developed, and the quality control output dimensions are as follows: a. "Lack of diagnosis corresponding to observations"; b. "Lack of diagnosis corresponding to observations"; c. "Conflict between measurement values ​​and diagnosis"; d. "Lack of diagnosis corresponding to measurement values".

[0077] In an optional embodiment, the Figure 4 The flowchart shown implements the quality inspection of medical examination reports: Solution 1: Use the trained entity and relationship extraction model to perform entity recognition and relationship extraction on the doctor's real-time written report to obtain the content of the report after analysis; according to the term extraction template library obtained in Process 1 Solution 1, extract the terminology content in the observations and conclusions based on the content after report analysis.

[0078] Solution 2: Use the trained generative model to generate the terminology content in the report and conclusion; Based on the term normalization vocabulary, normalize the terms in the observations and conclusions to obtain the standardized terms in the observations and conclusions. Use the standardized terms in the observations and conclusions as input parameters, call the service of the quality control reasoning map, and obtain the quality control results.

[0079] The method for processing medical examination reports provided in the embodiment of the present application obtains a target medical examination report to be processed; performs term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnostic opinion in the target medical examination report; performs quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnostic opinion, and the term information corresponding to what is seen in the historical medical examination reports and term information corresponding to the diagnostic opinion are obtained by processing the historical medical examination reports by a target model, and the target model is one of the following: an entity extraction model and a term generation model, which solves the problem in the related art of quality inspection of medical examination reports through machine learning models, because the machine learning model is a prediction end-to-end black box, which leads to poor interpretability of quality inspection of medical examination reports. In this scheme, the terminology information in the medical examination report is extracted to obtain the target terminology information, and then the target medical examination report is quality-inspected based on the target terminology information through the target reasoning graph to obtain the final quality inspection result. The quality inspection report obtained through the target reasoning graph is interpretable, and the target reasoning graph is constructed based on the terminology information corresponding to the actual historical medical examination reports and the terminology information corresponding to the diagnostic opinions, which is equivalent to a higher reliability of the machine model target reasoning graph, thereby achieving the effect of improving the accuracy of quality inspection of medical examination reports.

[0080] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0081] The embodiment of the present application also provides a medical examination report processing device. It should be noted that the medical examination report processing device of the embodiment of the present application can be used to execute the medical examination report processing method provided in the embodiment of the present application. The medical examination report processing device provided in the embodiment of the present application is introduced below.

[0082] Figure 5 Schematic diagram of a device for processing medical examination reports according to an embodiment of the present application. Figure 5 As shown, the device includes: a first acquisition unit 501, an extraction unit 502 and a quality inspection unit 503.

[0083] A first acquisition unit 501 is used to acquire a target medical examination report to be processed; An extraction unit 502 is used to extract terms from a target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to findings in the target medical examination report and term information corresponding to diagnostic opinions in the target medical examination report; The quality inspection unit 503 is used to perform quality inspection on the target medical examination report based on the target terminology information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by the terminology information corresponding to the findings in the historical medical examination report and the terminology information corresponding to the diagnosis opinion, and the terminology information corresponding to the findings in the historical medical examination report and the terminology information corresponding to the diagnosis opinion are obtained by processing the historical medical examination report by the target model, and the target model is one of the following: an entity extraction model and a terminology generation model.

[0084] The medical examination report processing device provided in the embodiment of the present application obtains the target medical examination report to be processed through the first acquisition unit 501; the extraction unit 502 performs term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; the quality inspection unit 503 performs quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnosis opinion, and the term information corresponding to what is seen in the historical medical examination reports and term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination reports by a target model, and the target model is one of the following: an entity extraction model and a term generation model, which solves the problem of quality inspection of medical examination reports through machine learning models in related technologies, because the machine learning model is a prediction end-to-end black box, which leads to poor interpretability of the quality inspection results of the medical examination reports. In this scheme, the terminology information in the medical examination report is extracted to obtain the target terminology information, and then the target medical examination report is quality-inspected based on the target terminology information through the target reasoning graph to obtain the final quality inspection result. The quality inspection report obtained through the target reasoning graph is interpretable, and the target reasoning graph is constructed based on the terminology information corresponding to the actual historical medical examination reports and the terminology information corresponding to the diagnostic opinions. It can adapt to the characteristics of the medical report itself, thereby achieving the effect of improving the accuracy of quality inspection of medical examination reports.

[0085] Optionally, the medical examination report processing device provided in the embodiment of the present application also includes: a second acquisition unit, used to obtain multiple first historical medical examination reports before performing quality inspection on the medical examination reports based on target term information and a target reasoning graph to obtain quality inspection results; an extraction unit, used to extract terms from the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnostic opinion in the first historical medical examination report; a construction unit, used to construct a target reasoning graph based on the historical term information.

[0086] Optionally, in the medical examination report processing device provided in the embodiment of the present application, the extraction unit 502 includes: a first acquisition subunit, used to acquire multiple second historical medical examination reports, and construct a target training set based on the multiple second historical medical examination reports; a training subunit, used to train the initial model based on the target training set to obtain a target model; and an extraction subunit, used to extract terms from the multiple first historical medical examination reports based on the target model to obtain historical term information.

[0087] Optionally, the medical examination report processing device provided in the embodiment of the present application, the acquisition subunit includes: a first labeling module, which is used to label the information in multiple second historical medical examination reports according to a preset medical entity relationship database if the target model is an entity extraction model, to obtain multiple third historical medical examination reports, and to construct a target training set based on the multiple third historical medical examination reports; a second labeling module, which is used to label the terms in the multiple second historical medical examination reports if the target model is a term generation model, to obtain a target sequence, and to construct a target training set based on the target sequence, wherein the target sequence is composed of term information corresponding to the findings in the second historical medical examination reports and term information corresponding to the diagnostic opinions in the second historical medical examination reports.

[0088] Optionally, in the medical examination report processing device provided in the embodiment of the present application, if the target model is an entity extraction model, the extraction subunit includes: an extraction module, used to extract entities and relationships between entities in a first historical medical examination report through the entity extraction model to obtain entity relationship data; a first extraction module, used to obtain a preset term extraction library, and perform term extraction on the entity relationship data based on the term extraction library to obtain historical term information.

[0089] Optionally, in the medical examination report processing device provided in the embodiment of the present application, if the target model is a terminology generation model, the extraction subunit includes: a processing module, used to preprocess multiple first historical medical examination reports to obtain processed first historical medical examination reports; a second extraction module, used to extract features of the processed first historical medical examination reports through the target model to obtain target feature information; and a first determination module, used to obtain historical terminology information based on the target feature information.

[0090] Optionally, in the medical examination report processing device provided in the embodiment of the present application, the first extraction module includes: an acquisition submodule, used to acquire multiple fourth historical medical examination reports, and acquire terminology information corresponding to the findings in the fourth historical medical examination reports and terminology information corresponding to the diagnostic opinions in the fourth historical medical examination reports; a processing submodule, used to pre-process the terminology information corresponding to the findings in the fourth historical medical examination reports and the terminology information corresponding to the diagnostic opinions in the fourth historical medical examination reports, and acquire the processed terminology information; and a determination submodule, used to obtain a preset term extraction library based on the processed terminology information.

[0091] Optionally, in the medical examination report processing device provided in the embodiment of the present application, the construction unit includes: a second acquisition sub-unit, used to acquire a normalized vocabulary, and normalize the historical term information according to the normalized vocabulary to obtain the processed historical term information; a first determination sub-unit, used to determine the first term information corresponding to the observations in the first historical medical examination report and the second term information corresponding to the diagnostic opinion in the first historical medical examination report according to the processed historical term information; a second determination sub-unit, used to determine the target reasoning link between the first term information corresponding to the observations and the second term information corresponding to the diagnostic opinion; and a construction sub-unit, used to construct a target reasoning graph based on the target reasoning link.

[0092] Optionally, in the medical examination report processing device provided in the embodiment of the present application, the second determination subunit includes: a second determination module, used to determine, for each second term information, the total number of times that each first term information and the second term information appear in the same first historical medical examination report; a third determination module, used to determine the first term information matched by the second term information based on the total number of times and a preset threshold; and a fourth determination module, used to determine the target reasoning link based on the second term information and the first term information matched by the second term information.

[0093] The above-mentioned medical examination report processing device includes a processor and a memory. The above-mentioned first acquisition unit 501, extraction unit 502 and quality inspection unit 503 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0094] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the quality inspection of the medical examination report can be achieved by adjusting the kernel parameters.

[0095] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0096] An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, a method for processing a medical examination report is implemented.

[0097] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes a method for processing a medical examination report when it is run.

[0098] like Figure 6 As shown, an embodiment of the present invention provides an electronic device, the device includes a processor, a memory, and a program stored in the memory and executable on the processor, and the processor implements the following steps when executing the program: obtaining a target medical examination report to be processed; performing term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; performing quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnosis opinion, and the term information corresponding to what is seen in the historical medical examination reports and term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination reports by a target model, and the target model is one of the following: an entity extraction model and a term generation model.

[0099] Optionally, before performing quality inspection on the medical examination report based on the target term information and the target reasoning graph to obtain the quality inspection result, the method also includes: obtaining multiple first historical medical examination reports; performing term extraction on the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnostic opinion in the first historical medical examination report; and constructing a target reasoning graph based on the historical term information.

[0100] Optionally, performing term extraction on multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report includes: obtaining multiple second historical medical examination reports and constructing a target training set based on the multiple second historical medical examination reports; training an initial model based on the target training set to obtain a target model; and performing term extraction on multiple first historical medical examination reports based on the target model to obtain historical term information.

[0101] Optionally, constructing a target training set based on multiple second historical medical examination reports includes: if the target model is an entity extraction model, labeling the information in the multiple second historical medical examination reports according to a preset medical entity relationship database to obtain multiple third historical medical examination reports, and constructing a target training set based on the multiple third historical medical examination reports; if the target model is a terminology generation model, labeling the terms in the multiple second historical medical examination reports to obtain a target sequence, and constructing a target training set based on the target sequence, wherein the target sequence is composed of terminology information corresponding to the findings in the second historical medical examination reports and terminology information corresponding to the diagnostic opinions in the second historical medical examination reports.

[0102] Optionally, if the target model is an entity extraction model, term extraction is performed on multiple first historical medical examination reports based on the target model to obtain historical term information, including: extracting entities and relationships between entities in the first historical medical examination reports through the entity extraction model to obtain entity relationship data; obtaining a preset term extraction library, and performing term extraction on the entity relationship data based on the term extraction library to obtain historical term information.

[0103] Optionally, if the target model is a terminology generation model, terminology is extracted from multiple first historical medical examination reports based on the target model to obtain historical terminology information, including: preprocessing the multiple first historical medical examination reports to obtain processed first historical medical examination reports; performing feature extraction on the processed first historical medical examination reports through the target model to obtain target feature information; and obtaining historical terminology information based on the target feature information.

[0104] Optionally, obtaining a preset term extraction library includes: obtaining multiple fourth historical medical examination reports, and obtaining term information corresponding to the findings in the fourth historical medical examination reports and term information corresponding to the diagnostic opinions in the fourth historical medical examination reports; preprocessing the term information corresponding to the findings in the fourth historical medical examination reports and the term information corresponding to the diagnostic opinions in the fourth historical medical examination reports to obtain processed term information; and obtaining a preset term extraction library based on the processed term information.

[0105] Optionally, constructing a target reasoning graph based on historical terminology information includes: obtaining a normalized vocabulary, and normalizing the historical terminology information based on the normalized vocabulary to obtain processed historical terminology information; determining, based on the processed historical terminology information, first terminology information corresponding to what is seen in a first historical medical examination report and second terminology information corresponding to the diagnosis opinion in the first historical medical examination report; determining a target reasoning link between the first terminology information corresponding to what is seen and the second terminology information corresponding to the diagnosis opinion; and constructing a target reasoning graph based on the target reasoning link.

[0106] Optionally, determining the target reasoning link between the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion includes: for each second term information, determining the total number of times each first term information and the second term information appear in the same first historical medical examination report; based on the total number of times and a preset threshold, determining the first term information that matches the second term information; and determining the target reasoning link based on the second term information and the first term information that matches the second term information.

[0107] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0108] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes the following method steps: obtaining a target medical examination report to be processed; performing term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to what is seen in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; performing quality inspection on the target medical examination report based on the target term information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by using term information corresponding to what is seen in historical medical examination reports and term information corresponding to the diagnosis opinion, and the term information corresponding to what is seen in the historical medical examination reports and term information corresponding to the diagnosis opinion are obtained by processing the historical medical examination reports by a target model, and the target model is one of the following: an entity extraction model and a term generation model.

[0109] Optionally, before performing quality inspection on the medical examination report based on the target term information and the target reasoning graph to obtain the quality inspection result, the method also includes: obtaining multiple first historical medical examination reports; performing term extraction on the multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnostic opinion in the first historical medical examination report; and constructing a target reasoning graph based on the historical term information.

[0110] Optionally, performing term extraction on multiple first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report includes: obtaining multiple second historical medical examination reports and constructing a target training set based on the multiple second historical medical examination reports; training an initial model based on the target training set to obtain a target model; and performing term extraction on multiple first historical medical examination reports based on the target model to obtain historical term information.

[0111] Optionally, constructing a target training set based on multiple second historical medical examination reports includes: if the target model is an entity extraction model, labeling the information in the multiple second historical medical examination reports according to a preset medical entity relationship database to obtain multiple third historical medical examination reports, and constructing a target training set based on the multiple third historical medical examination reports; if the target model is a terminology generation model, labeling the terms in the multiple second historical medical examination reports to obtain a target sequence, and constructing a target training set based on the target sequence, wherein the target sequence is composed of terminology information corresponding to the findings in the second historical medical examination reports and terminology information corresponding to the diagnostic opinions in the second historical medical examination reports.

[0112] Optionally, if the target model is an entity extraction model, term extraction is performed on multiple first historical medical examination reports based on the target model to obtain historical term information, including: extracting entities and relationships between entities in the first historical medical examination reports through the entity extraction model to obtain entity relationship data; obtaining a preset term extraction library, and performing term extraction on the entity relationship data based on the term extraction library to obtain historical term information.

[0113] Optionally, if the target model is a terminology generation model, terminology is extracted from multiple first historical medical examination reports based on the target model to obtain historical terminology information, including: preprocessing the multiple first historical medical examination reports to obtain processed first historical medical examination reports; performing feature extraction on the processed first historical medical examination reports through the target model to obtain target feature information; and obtaining historical terminology information based on the target feature information.

[0114] Optionally, obtaining a preset term extraction library includes: obtaining multiple fourth historical medical examination reports, and obtaining term information corresponding to the findings in the fourth historical medical examination reports and term information corresponding to the diagnostic opinions in the fourth historical medical examination reports; preprocessing the term information corresponding to the findings in the fourth historical medical examination reports and the term information corresponding to the diagnostic opinions in the fourth historical medical examination reports to obtain processed term information; and obtaining a preset term extraction library based on the processed term information.

[0115] Optionally, constructing a target reasoning graph based on historical terminology information includes: obtaining a normalized vocabulary, and normalizing the historical terminology information based on the normalized vocabulary to obtain processed historical terminology information; determining, based on the processed historical terminology information, first terminology information corresponding to what is seen in a first historical medical examination report and second terminology information corresponding to the diagnosis opinion in the first historical medical examination report; determining a target reasoning link between the first terminology information corresponding to what is seen and the second terminology information corresponding to the diagnosis opinion; and constructing a target reasoning graph based on the target reasoning link.

[0116] Optionally, determining the target reasoning link between the first term information corresponding to the observation and the second term information corresponding to the diagnostic opinion includes: for each second term information, determining the total number of times each first term information and the second term information appear in the same first historical medical examination report; based on the total number of times and a preset threshold, determining the first term information that matches the second term information; and determining the target reasoning link based on the second term information and the first term information that matches the second term information.

[0117] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0118] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0121] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0122] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0123] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0124] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0125] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0126] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for processing a medical examination report, characterized in that: include: Obtain pending target medical examination reports; Performing term extraction on the target medical examination report to obtain target term information in the target medical examination report, wherein the target term information at least includes term information corresponding to the findings in the target medical examination report and term information corresponding to the diagnosis opinion in the target medical examination report; Based on the target terminology information and the target reasoning graph, the target medical examination report is quality inspected to obtain a quality inspection result, wherein the target reasoning graph is constructed by terminology information corresponding to what is found in historical medical examination reports and terminology information corresponding to diagnostic opinions, and the terminology information corresponding to what is found in historical medical examination reports and terminology information corresponding to diagnostic opinions are obtained by processing the historical medical examination reports with a target model, and the target model is one of the following: an entity extraction model and a terminology generation model.

2. The method according to claim 1, characterized in that Before performing quality inspection on the medical examination report based on the target term information and the target reasoning graph to obtain the quality inspection result, the method further includes: Get multiple first history medical exam reports; Performing term extraction on the plurality of first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, wherein the historical term information at least includes: term information corresponding to the findings in the first historical medical examination report and term information corresponding to the diagnosis opinion in the first historical medical examination report; The target reasoning graph is constructed based on the historical term information.

3. The method according to claim 2, characterized in that The term extraction is performed on the plurality of first historical medical examination reports to obtain historical term information corresponding to each first historical medical examination report, including: Acquire a plurality of second historical medical examination reports, and construct a target training set according to the plurality of second historical medical examination reports; Training the initial model according to the target training set to obtain a target model; The historical term information is obtained by performing term extraction on the plurality of first historical medical examination reports according to the target model.

4. The method according to claim 3, characterized in that Constructing a target training set according to the plurality of second historical medical examination reports comprises: If the target model is an entity extraction model, the information in the plurality of second historical medical examination reports is annotated according to a preset medical entity relationship database to obtain a plurality of third historical medical examination reports, and the target training set is constructed according to the plurality of third historical medical examination reports; If the target model is a terminology generation model, the terms in the multiple second historical medical examination reports are labeled to obtain a target sequence, and the target training set is constructed based on the target sequence, wherein the target sequence is composed of terminology information corresponding to the findings in the second historical medical examination reports and terminology information corresponding to the diagnostic opinions in the second historical medical examination reports.

5. The method according to claim 3, characterized in that: If the target model is an entity extraction model, term extraction is performed on the plurality of first historical medical examination reports according to the target model to obtain the historical term information including: Extracting entities and relationships between entities in the first historical medical examination report by using the entity extraction model to obtain entity relationship data; A preset term extraction library is obtained, and term extraction is performed on the entity relationship data according to the term extraction library to obtain the historical term information.

6. The method according to claim 3, characterized in that If the target model is a term generation model, term extraction is performed on the plurality of first historical medical examination reports according to the target model to obtain the historical term information including: Preprocessing the plurality of first historical medical examination reports to obtain processed first historical medical examination reports; Extracting features from the processed first historical medical examination report using the target model to obtain target feature information; The historical term information is obtained according to the target feature information.

7. The method according to claim 5, characterized in that Get the preset term extraction library including: Acquire a plurality of fourth historical medical examination reports, and acquire term information corresponding to findings in the fourth historical medical examination reports and term information corresponding to diagnostic opinions in the fourth historical medical examination reports; Preprocessing the terminology information corresponding to the findings in the fourth historical medical examination report and the terminology information corresponding to the diagnosis opinion in the fourth historical medical examination report to obtain processed terminology information; The preset term extraction library is obtained based on the processed term information.

8. The method according to claim 2, characterized in that: Constructing the target reasoning graph according to the historical term information includes: Acquire a normalized vocabulary, and perform normalization processing on the historical term information according to the normalized vocabulary to obtain processed historical term information; Determining, based on the processed historical term information, first term information corresponding to the findings in the first historical medical examination report and second term information corresponding to the diagnosis in the first historical medical examination report; Determining a target reasoning link between first term information corresponding to the observation and second term information corresponding to the diagnosis opinion; The target reasoning graph is constructed according to the target reasoning link.

9. The method according to claim 8, characterized in that Determining a target reasoning link between first term information corresponding to the observation and second term information corresponding to the diagnosis opinion includes: For each second term information, determining the total number of times each first term information and the second term information appear in the same first historical medical examination report; Determining the first term information matched by the second term information according to the total number of times and a preset threshold; The target inference link is determined according to the second term information and the first term information matched by the second term information.

10. A device for processing medical examination reports, characterized in that: include: A first acquisition unit, used for acquiring a target medical examination report to be processed; An extraction unit, configured to perform term extraction on the target medical examination report to obtain target term information in the medical examination report, wherein the target term information at least includes term information corresponding to findings in the target medical examination report and term information corresponding to diagnostic opinions in the target medical examination report; A quality inspection unit is used to perform quality inspection on the target medical examination report based on the target terminology information and the target reasoning graph to obtain a quality inspection result, wherein the target reasoning graph is constructed by terminology information corresponding to what is seen in historical medical examination reports and terminology information corresponding to diagnostic opinions, and the terminology information corresponding to what is seen in historical medical examination reports and terminology information corresponding to diagnostic opinions are obtained by processing the historical medical examination reports with a target model, and the target model is one of the following: an entity extraction model and a terminology generation model.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the storage medium is controlled to execute the method for processing a medical examination report according to any one of claims 1 to 9 on a device.

12. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for processing a medical examination report as described in any one of claims 1 to 9.

Images