A PET / CT report quality control method based on structured data
By using a structured data-based PET/CT report quality control method, the object attributes and relationships in the report are analyzed. Combined with rules and semantic quality control items, traceable quality control results are generated, which solves the problems of missing and inconsistent descriptions in PET/CT reports and improves the accuracy and stability of quality control.
Patent Information
- Application Number
- CN202611563543.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-10
- Publication Date
- 2026-11-17
AI Technical Summary
Existing PET/CT report quality control methods suffer from quality problems such as missing descriptions, non-standard terminology, and inconsistencies between what is seen and what is said. Current technologies are unable to achieve accurate location and stable, traceable quality control.
A quality control method based on structured data is adopted. By parsing the structured data in the report, the attributes, relationships and source text locations of the objects are read. Combined with the quality control rule base, the rule quality control items and semantic quality control items are judged to generate traceable quality control results.
It has achieved accuracy, stability and traceability in PET/CT report quality control, solved the problems of inability to make association judgments through full-text keyword matching and the unrestricted scope of semantic model judgment, and ensured the reliability of quality control results.
Smart Images

Figure FT_1 
Figure FT_2
Abstract
Description
Technical Field
[0001] This application relates to the field of medical information processing, and in particular to a quality control method for PET / CT reports based on structured data. Background Technology
[0002] PET / CT (Positron Emission Tomography / Computed Tomography) is a core tool for tumor staging, efficacy assessment, and recurrence monitoring, and the quality of its reports directly impacts the accuracy of clinical decisions. A well-structured and clearly described PET / CT report provides clinicians with crucial information such as lesion localization, metabolic activity assessment, and diagnostic opinions. However, due to differences in the experience and writing habits of report writers, PET / CT reports often suffer from quality issues such as missing descriptions, non-standard terminology, and inconsistencies between findings and opinions. Therefore, report quality control is a crucial aspect of ensuring medical quality and patient safety. With the rapid development of artificial intelligence technology, utilizing natural language processing and semantic models to assist in report quality control has become a research hotspot in the field of medical informatics. However, how to achieve accurate, stable, and traceable automated quality control remains an engineering problem that urgently needs to be solved.
[0003] A PET / CT report typically includes six parts: patient information, examination information, clinical diagnosis, brief medical history, examination findings, and diagnostic opinion. The examination findings section records abnormalities, dimensions, and SUVmax (maximum standardized uptake value) of various organs or sites in structured data format, while the diagnostic opinion section provides a conclusive judgment on the findings in structured data format.
[0004] There are two main approaches to report quality control. One approach involves matching keywords across the entire report to identify potential problems using a pre-defined keyword list. However, the lack of structural constraints in keyword matching means that matches may include statements irrelevant to the quality control objectives, resulting in a high false alarm rate. Furthermore, the location, size, metabolic activity, and related diagnostic opinions for the same finding are scattered throughout the report, making it difficult for existing full-text search methods to connect these related findings for joint judgment. This leads to many issues requiring a comprehensive assessment of examination findings and diagnostic opinions failing to be detected. The other approach involves directly inputting the entire report into a semantic model, which then reads the entire text and lists the problems. Since the semantic model must identify the judgment objects and their scope throughout the report, its judgment scope is unconstrained. The model may use any statement in the report as a basis for judgment, leading to unstable results—multiple quality control checks on the same report may yield different results. Additionally, the model's problem statements and cited content are self-organized and lack verifiability. Doctors cannot verify whether the model's cited evidence actually exists in the report, making its judgments difficult to retain as formal review evidence.
[0005] Therefore, there is an urgent need for a quality control method for PET / CT reports that can accurately locate the quality control objects and is stable and traceable. Summary of the Invention
[0006] The purpose of this application is to provide a PET / CT report quality control method based on structured data, which uses specific objects in the structured data as the judgment criteria, so that the quality control results can be traced back to the identified source object, thereby improving the accuracy, stability and traceability of quality control.
[0007] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a PET / CT report quality control method based on structured data, including: Obtain the report information and structured data of the PET / CT report to be quality controlled, and obtain the pre-set quality control rule base; The structured data is parsed to read the attributes, relationships, identifiers, source text, and source text location of each object in the structured data, and combined with the report information to obtain the parsing results; Load quality control items from the quality control rule base. Each quality control item includes a quality control method, quality control object screening conditions, and judgment conditions. The quality control items are divided into rule-based quality control items and semantic quality control items according to the quality control method. For the aforementioned rule-based quality control items, quality control objects are selected from the parsing results according to the quality control object screening conditions of the rule-based quality control items. The attributes and relationships of the quality control objects are taken and compared according to the judgment conditions of the rule-based quality control items to generate quality control results. For the semantic quality control items, quality control objects are selected from the parsing results according to the quality control object screening conditions of the semantic quality control items, and the judgment results and evidence of the quality control objects are obtained using the semantic model. The source object verification and reference content verification are performed on the evidence, and the quality control results are generated after passing the verification. The generated quality control results are summarized and saved in association with the PET / CT report to be controlled and the relevant quality control objects, according to the identifiers of the quality control objects involved.
[0008] Furthermore, the report information includes patient information, examination information, clinical diagnosis, brief medical history, examination findings, and diagnostic opinion; The structured data includes an examination findings section and a diagnostic opinion section. The examination findings section includes organ entries and found objects, and the diagnostic opinion section includes opinion conclusions.
[0009] Furthermore, the parsing of the structured data involves reading the attributes, relationships, identifiers, source text, and source text locations of each object within the structured data, and combining this information with the report information to obtain the parsing results, including: Patient information, examination information, clinical diagnosis and brief medical history are read from the report information to form context objects, and the original content of each context object in the report information is retained as the source text. From the inspection view portion of the structured data, organ entries and discovery objects are read by object type, and the attributes, identifiers, source text, and position of the source text in the inspection view are read for each organ entry and discovery object. The attributes of the organ entry include organ name and positive / negative status; the attributes of the discovery object include the organ to which it belongs, abnormal manifestations, size, maximum standardized uptake value, and degree of diagnostic certainty. From the diagnostic opinion section of the structured data, the opinion conclusions are read according to the object type, and the attributes, identifiers, source texts, and positions of the source texts in the diagnostic opinion are read. The attributes include the organ referred to, the abnormal manifestation, the positive or negative result, and the degree of diagnostic certainty. Read the correlation between the diagnostic opinion section of the structured data and the findings of the examination section; The read organ entries and their attributes, identifiers, source texts and locations, the discovered objects and their attributes, identifiers, source texts and locations, the opinion conclusions and their attributes, identifiers, source texts and locations, the context objects and source texts, and the relationships are summarized into the parsing results.
[0010] Furthermore, the quality control items also include quality control titles, problem levels, and prompt templates.
[0011] Further, the quality control objects are selected from the parsing results according to the filtering conditions of the rule-based quality control items. The attributes and relationships of the quality control objects are then retrieved and compared according to the judgment conditions of the rule-based quality control items to generate quality control results, including: From the analysis results, objects that meet the quality control object screening criteria are selected as quality control objects, and the attributes and relationships corresponding to the quality control objects are read. The attribute values and relationships of the read attributes are compared with the judgment conditions of the quality control items according to the rules. If the comparison result meets the judgment conditions, it is considered a hit. The quality control title, the identifier of the object involved, the problem level, the judgment basis, the source text of the object involved, and the prompt generated by filling in the attribute value or source text of the object involved according to the prompt template are recorded to obtain the quality control result. The judgment criteria for the rule quality control items include the presence or absence of the attribute values, their similarities or differences, and the presence or absence of association relationships. The criteria for determining the rule quality control items are the attribute values, relationships, and context object attribute values of the quality control objects involved.
[0012] Furthermore, the process of selecting quality control objects from the parsing results according to the filtering conditions of the rule-based quality control items, taking the attributes and relationships of the quality control objects, and comparing them according to the judgment conditions of the rule-based quality control items, also includes: If the rule quality control item specifies a control object, the control object is selected from the parsing results according to the control object screening conditions, and the corresponding attribute value and association relationship of the control object are obtained according to the type of the control object; The attribute values of the control object and the quality control object are compared according to the judgment conditions of the rule quality control items, and the correlation of the control object is compared according to the judgment conditions.
[0013] Furthermore, the process of selecting quality control objects from the parsing results according to the filtering conditions of the rule-based quality control items, taking the attributes and relationships of the quality control objects, and comparing them according to the judgment conditions of the rule-based quality control items, also includes: If the judgment condition of the rule quality control item involves the patient's gender, then the gender of the patient information in the context object is taken, and the gender of the patient information is compared with the judgment condition.
[0014] Further, the process involves selecting quality control objects from the parsing results based on the quality control object filtering criteria of the semantic quality control items, obtaining the judgment results and evidence of the quality control objects using a semantic model, verifying the source object and cited content of the evidence, and generating quality control results after passing the verification, including: From the parsing results, objects that meet the quality control object screening criteria are selected as quality control objects, and the corresponding attributes and source text of the quality control objects are read. The judgment criteria of the semantic quality control item, the attributes of the quality control object and the source text, and the context object and the source text are fed into the semantic model; When a control item specifies a reference object, the attributes and source text of the reference object are submitted. The semantic model determines whether the meaning expressed by the source text of the quality control object, the context object, and the comparison object conforms to the semantic quality control item based on the judgment conditions of the semantic quality control item. When the semantic quality control criteria are met, the semantic model returns several pieces of evidence that meet the semantic quality control criteria. The evidence includes cited content and source objects. The cited content includes a continuous segment of a source text of a quality control object, a control object, or a context object. The source object includes a quality control object, a control object, or a context object. The source object verification and cited content verification are performed on several pieces of evidence. Based on the several pieces of evidence that pass the source object verification and cited content verification, the quality control title, the identifiers of the quality control objects and control objects involved, the problem level, the prompts generated by filling in the attributes of the quality control objects and control objects involved or the source text according to the prompt template, and the evidence pieces used as the basis for judgment are recorded to obtain the quality control results.
[0015] Furthermore, the verification of the source object and the verification of the cited content for several pieces of evidence include: Verify whether the source of the evidence is the quality control object, control object, or context object currently submitted for judgment; Verify whether the cited content of the evidence is a continuous substring in the source text of the source object.
[0016] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the PET / CT report quality control method based on structured data as described above.
[0017] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a PET / CT report quality control method based on structured data. By parsing the report into specific objects with clear identifiers and relationships, it enables direct association between examination findings and diagnostic opinions found in the same location, regardless of their location scattered throughout the report, thus solving the problem that full-text keyword matching cannot link scattered content. By directly comparing object attribute values through rule-based quality control items, without matching keywords in the full text, it solves the problems of irrelevant statements and high false alarm rates caused by keyword matching. Semantic quality control items use the quality control objects determined by the structured data as the judgment targets, eliminating the need for the semantic model to identify judgment objects from the full text, thus solving the problems of unrestricted judgment scope and unstable results. Furthermore, by verifying the source object and the cited content, it ensures that evidence can be traced back to a definite source object, solving the problems of unverifiable citations and judgment results that cannot be retained as a basis for review. By associating and saving quality control results with report objects according to the identifiers of the involved objects, each quality control result can be accurately located to a specific object, thereby achieving a unified standard of accuracy, stability, and traceability in PET / CT report quality control. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 A flowchart illustrating a PET / CT report quality control method based on structured data, provided as an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0022] In one exemplary embodiment, such as Figure 1 As shown, a PET / CT report quality control method based on structured data is provided, which can be applied to a medical information processing platform. This method is executed by a computer device, specifically by a terminal or server alone, or by both. In this embodiment, the method includes the following steps: S1: Obtain the report information and structured data of the PET / CT report to be quality controlled, and obtain the pre-set quality control rule library.
[0023] As an optional implementation, the medical information processing platform receives report information and structured data from the PET / CT report to be quality controlled. The report information includes patient information, examination information, clinical diagnosis, brief medical history, examination findings, and diagnostic opinions. The structured data is divided into an examination findings section and a diagnostic opinion section. The examination findings section includes site entries, organ entries, and found objects. The site entries contain organ entries, and the organ entries contain found objects. The diagnostic opinion section includes opinion entries and opinion conclusions. The opinion entries contain the opinion conclusions.
[0024] The quality control rule base is a pre-configured set of rules, pre-set by medical experts and quality control managers according to clinical guidelines and quality control standards. It consists of multiple quality control items, each including the quality control method, title, selection criteria for the quality control object, judgment criteria, problem level, and prompt template. The quality control method indicates whether the judgment for this quality control item is made by a rule or a semantic model, where the semantic model uses a general large language model. The quality control title is the name of the quality control item, used in the quality control results to indicate which quality control item was hit, such as finding missing object dimensions. The selection criteria are the conditions used to select the quality control object from the current PET / CT report to be controlled, such as the discovery of abnormalities such as nodules or masses, and negative opinions. The judgment criteria are the conditions for determining whether each quality control object is hit, such as the size not being filled in with the actual value. The problem level is the severity recorded after the quality control item is hit, which is divided into error, warning and prompt. The prompt template is the sentence structure used to generate the prompt, which consists of fixed text and fields to be filled in. When generating the prompt, each field to be filled in is filled with the attribute value of the object involved or the original text involved.
[0025] S2: Parse the structured data, and read the attributes, relationships, identifiers, source texts and source text locations of each object in the structured data, and combine them with the report information to obtain the parsing results.
[0026] As an optional implementation, patient information, examination information, clinical diagnosis, and brief medical history are read from the report information of S1 to form a context object. Each piece of information retains the original content read by S1 as the source text; the gender in the patient information is used by S4 for determination.
[0027] The structured data received by S1 is parsed, and its attributes are read according to object type: organ name and positive / negative status are read from the organ entries in the examined section; organ, abnormality, size, maximum standardized uptake value (SUVmax), and certainty are read from the found objects in the examined section; and organ, abnormality, positive / negative status, and certainty are read from the opinion conclusion in the diagnostic opinion section. Simultaneously, the correlation between the opinion conclusion in the diagnostic opinion section and the found objects in the examined section is read.
[0028] Among these, "positive or negative" indicates whether the organ entry or the opinion / conclusion is abnormal or not; "belonging organ" refers to the organ where the discovery is located, and "referring to organ" refers to the organ addressed by the opinion / conclusion; "abnormal manifestation" refers to the abnormality described by the discovery or conclusion, such as nodules, masses, or increased metabolism; "certainty" refers to the degree of diagnostic certainty expressed by the discovery or conclusion; and "association" is the directional relationship between the opinion / conclusion and the discovery, indicating that the conclusion is made in response to this particular discovery. All of the above attributes and associations are provided by structured data and can be directly accessed.
[0029] Each object in the structured data has its own identifier, source text, and the position of the source text in the inspection findings or diagnostic opinions, which are read together; the source text is the original text fragment corresponding to that object in the inspection findings or diagnostic opinions.
[0030] The analysis results include the above objects and their attributes, the relationship between opinions and conclusions and the discovered objects, the identifiers and source texts of each object and their locations, as well as the context objects and the source texts of each piece of information.
[0031] S3: Load quality control items from the quality control rule base. The quality control items include quality control methods, quality control object screening conditions, and judgment conditions. The quality control items are divided into rule-based quality control items and semantic quality control items according to the quality control methods.
[0032] As an optional implementation, quality control items are defined in a quality control rule base. Each quality control item record is read, including its quality control method, title, object selection criteria, judgment criteria, problem level, and prompt template. These are categorized into two types based on the quality control method: records with a "rule" method are loaded as rule-based quality control items; records with a "semantic" method are loaded as semantic quality control items. Rule-based quality control items can be determined simply by comparing the attribute values of the object, resulting in a unique conclusion. Semantic quality control items determine the meaning of the free text, which is made by a semantic model, and the judgment result must be validated by the rules before it is adopted.
[0033] As an optional implementation, each quality control item pre-specifies whether a control object is set and the selection criteria for the control object. For quality control items that specify a control object, the selection criteria for the control object are also read. The control object is another object that must be compared with the quality control object during the judgment. The selection criteria for the control object are the pre-specified correlation or specified attribute relationship used to select control objects from this quality control PET / CT report, such as the opinion conclusions associated with the found object.
[0034] S4: For the rule-based quality control items, select quality control objects from the parsing results according to the quality control object screening conditions of the rule-based quality control items, take the attributes and relationships of the quality control objects, and compare them according to the judgment conditions of the rule-based quality control items to generate quality control results.
[0035] As an optional implementation method, the quality control items are executed one by one. For each quality control item, quality control objects are selected from the organ entries, discovery objects, or opinion conclusions parsed from S2 according to the quality control object screening conditions of a certain quality control item. The judgment is then carried out one by one in two steps: value acquisition and comparison. Specifically, this includes: Values: For quality control items that are organ entries, use the organ name and positive / negative sign; for quality control items that are discovery objects, use the organ, abnormal manifestation, size, SUVmax, and certainty; for quality control items that are opinion conclusions, use the organ referred to, abnormal manifestation, positive / negative sign, and certainty. If a quality control item specifies a control object, select a control object from this report according to the pre-defined control object screening criteria for that quality control item, and use the corresponding attributes mentioned above according to its object type; if the judgment criteria involve patient gender, use the gender from the patient information in the context object composed of S2. The values used are limited to the attributes of the quality control object itself, the attributes of the control object, and the gender in the patient information, and are not taken from the full report.
[0036] Comparison: The values are compared according to the judgment criteria. The comparison is limited to the presence, similarity, and difference of attribute values, as well as the presence or absence of association relationships. If the value is a placeholder (a symbol in structured data indicating that the value of that item has not yet been filled in, such as xxx), it is considered that the attribute has not been filled in with an actual value. If the judgment criteria specify a table that can be consulted, the comparison is made with the records in that table. A match is achieved if the comparison result meets the judgment criteria.
[0037] A successful hit generates a quality control result, recording the quality control item (the title of the hit quality control), the identifier of the involved object, the problem level, the judgment basis, the prompt, and the source text of the quality control object. The involved object refers to the quality control object and the control object used in this judgment. The prompt is generated according to the prompt template of the hit quality control item: according to each field to be filled in the template, the corresponding attribute value or source text of the involved object is filled in. The prompt is an explanation for doctors to read in the quality control result.
[0038] In one specific implementation, the following are examples of rule quality control items in the quality control rule base. Each example follows the two-step process of S4: value acquisition and comparison. In each item, the first part is the screening condition for quality control objects, and the second part is the judgment condition: objects with abnormal manifestations of nodules or masses, where the actual size value is not filled in; objects with abnormal manifestations of increased metabolism, where the actual SUVmax value is not filled in; objects with a correlation to the opinion conclusion, but whose certainty differs from the opinion conclusion; all objects, where the organ to which they belong is mutually exclusive with the patient's gender; organ entries with negative sex, where there are objects under them.
[0039] Whether the organ is mutually exclusive with the patient's gender is determined by referring to an organ dictionary table; this table is pre-established and records the names of each organ, as well as the gender of organs that belong only to men or only to women. All of the above are specified in the quality control items of the quality control rule base, and what is listed here is only a part of it.
[0040] S5: For the semantic quality control item, select the quality control object from the parsing results according to the quality control object screening conditions of the semantic quality control item, and use the semantic model to obtain the judgment result and evidence of the quality control object. Perform source object verification and reference content verification on the evidence, and generate the quality control result after passing the verification.
[0041] As an optional implementation method, semantic quality control items are executed one by one. For each semantic quality control item, quality control objects are selected from the organ entries, discovery objects or opinion conclusions parsed by S2 according to their quality control object screening conditions, and the judgment is made one by one in three steps: submission, response and verification.
[0042] Input: Input the judgment criteria of the semantic quality control item, the attributes of the quality control object and the source text into the semantic model; if the semantic quality control item specifies a reference object, select the reference object in the same way as S4, and input the attributes and source text of the reference object together; input the context object together, but do not input the full text of the report.
[0043] Answer: The semantic model compares the meanings of the source texts of the quality control object, context object, and control object based on the judgment criteria to determine whether the quality control item is met, and returns whether it is a hit and several pieces of evidence. Each piece of evidence consists of cited content and source object. The cited content must come from a continuous segment of source text of the quality control object, control object, or context object, and must not be rewritten, spliced, or summarized; the source object includes the quality control object, control object, or context object.
[0044] Verification: The following two verifications are performed on the response returned by the model in sequence: (1) Source object verification: The source object of each piece of evidence must be the object sent in this judgment, i.e., the quality control object, the control object, or the context object; (2) Citation content verification: The citation content of each piece of evidence must be a continuous substring in the source text of its source object, which is determined by retrieval.
[0045] If the answer is correct and passes both of the above checks, it is considered a successful test, generating a quality control result that records the quality control item, the identifier of the involved object, the problem level, the prompt, and the evidence used as the basis for the judgment. The involved object refers to the quality control object for this judgment, as well as the control object used as the source of evidence; the prompt is generated in the same way as in S4.
[0046] In one specific implementation, the following are examples of semantic quality control items in the quality control rule base. Each example is executed according to the three steps of input, response, and verification in S5. Each item consists of the quality control object screening conditions, judgment conditions, and control object screening conditions in sequence: (1) All discovered objects, determine whether the diagnostic opinion has omitted the discovery, and the control object is the opinion conclusions that are the same as the organ to which the object belongs; (2) All discovered objects, determine whether the organ described is the organ that has been removed as recorded in the brief medical history; (3) All opinion conclusions, determine whether there are other conclusions in the diagnostic opinion that contradict it, and the control object is other opinion conclusions that are the same as the organ to which the object belongs; (4) Discovered objects that are related to the opinion conclusion, determine whether they contradict the opinion conclusion, and the control object is the opinion conclusions that are related to them.
[0047] Taking the semantic quality control item that determines whether the discovered object and the associated opinion conclusion contradict each other as an example, when executing this semantic quality control item, the discovered objects that are related to the opinion conclusion are selected from all the discovered objects parsed by S2, and judged one by one. Taking one of the discovered objects as an example, the organ to which the discovered object belongs is "right lower lobe", the abnormal manifestation is "nodule", and the SUVmax is "1.8". The source text is "A nodule is visible in the right lower lobe, with slightly increased radioactive uptake, SUVmax 1.8". Based on the association relationship, the opinion conclusion associated with this discovered object is obtained as a control object. The organ referred to in this opinion conclusion is "both lungs", the positive / negative status is "negative", and the source text is "No abnormal metabolic increase foci were seen in both lungs". The semantic model compares the attributes of the two objects with the source text according to the judgment conditions, judges that the two contradict each other, and returns the hit and corresponding evidence. After the evidence passes the source object verification and the cited content verification, the quality control result is generated.
[0048] S6: Summarize the generated quality control results, and save the summarized quality control results along with the PET / CT report to be controlled and the relevant quality control objects according to the identifiers of the quality control objects involved.
[0049] As an optional implementation, the quality control results generated by S4 and S5 are summarized. The quality control results of the two types of quality control items record the same content: the quality control item that was hit, the identifier of the object involved, the problem level, the prompt, the judgment basis, and the original text involved. The original text involved is the source text of the object involved in the rule quality control item, or the reference content of each piece of evidence that passed the verification of the semantic quality control item.
[0050] Understandably, each hit on a quality control item records the basis for judgment, and this basis can be traced back to a definite source: the basis for judgment of rule-based quality control items is the attributes and relationships of the objects involved, tracing back to the identifier of the object; the basis for judgment of semantic quality control items is the evidence cited by the semantic model, the content cited by the evidence is taken from the input content, tracing back to the source object of the evidence.
[0051] Each quality control result is associated with and saved in relation to the PET / CT report to be quality controlled and its associated objects, based on the identifiers of the objects involved. During the review, the specific object is located by its identifier, and the content is verified against the original text.
[0052] Implementing S1 to S6, S2 parses the structured data, decomposing the scattered examination findings and diagnostic opinions in the report into specific objects with clear identifiers and attribute values—organ entries, discovery objects, and opinion conclusions. It also reads the correlation between opinion conclusions and discovery objects, enabling examination findings and diagnostic opinions found in the same location to be directly linked through these correlations, regardless of their location within the full report text. This solves the problem that full-text keyword matching cannot correlate scattered content. Furthermore, this application categorizes quality control items into rule-based and semantic quality control items for separate processing. Rule-based quality control items are judged by directly comparing the attribute values of the object in S4, without matching keywords in the report text. This solves the problem of irrelevant statements and high false alarm rates in full-text keyword matching. Semantic quality control items, in S5, input the object's attributes and source text. The object and its scope are determined by the structured data, and the semantic model does not need to identify the judgment object from the full report text. It makes a judgment on only one quality control object at a time, and the input content is shorter than the full report text, reducing the impact of irrelevant content on the judgment. This application addresses the issues of unrestricted judgment scope and unstable judgment results in semantic models. Furthermore, in S5, it verifies the source object and reference content of the evidence returned by the semantic model. Source object verification ensures that the source object of each piece of evidence is the object submitted for judgment, while reference content verification ensures that the reference content of each piece of evidence is a continuous substring in the source text of its source object. This ensures that the reference content of the adopted evidence comes from the submitted content and is traceable to a definite source object, solving the problems of unverifiable semantic model references and difficulty in retaining judgment results as a basis for review in existing technologies. Finally, in S6, this application associates and saves the quality control results with the report and its object according to the identifier of the object involved, making each quality control result traceable to a definite source. During review, doctors can locate the specific object by the identifier and check the content against the original text. This ensures that each quality control finding corresponds to a specific object in the structured data, the quality control basis is traceable to a definite source, and the location of the quality control result is accurate to the object level, effectively improving the accuracy, stability, and traceability of PET / CT report quality control.
[0053] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 2As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores PET / CT report data to be quality controlled. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a PET / CT report quality control method based on structured data.
[0054] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0055] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0056] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0057] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0058] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0059] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0060] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0062] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A PET / CT report quality control method based on structured data, characterized in that, The PET / CT report quality control methods include: Obtain the report information and structured data of the PET / CT report to be quality controlled, and obtain the pre-set quality control rule base; The structured data is parsed to read the attributes, relationships, identifiers, source text, and source text location of each object in the structured data, and combined with the report information to obtain the parsing results; Load quality control items from the quality control rule base. Each quality control item includes a quality control method, quality control object screening conditions, and judgment conditions. The quality control items are divided into rule-based quality control items and semantic quality control items according to the quality control method. For the aforementioned rule-based quality control items, quality control objects are selected from the parsing results according to the quality control object screening conditions of the rule-based quality control items. The attributes and relationships of the quality control objects are taken and compared according to the judgment conditions of the rule-based quality control items to generate quality control results. For the semantic quality control items, quality control objects are selected from the parsing results according to the quality control object screening conditions of the semantic quality control items, and the judgment results and evidence of the quality control objects are obtained using the semantic model. The source object verification and reference content verification are performed on the evidence, and the quality control results are generated after passing the verification. The generated quality control results are summarized and saved in association with the PET / CT report to be controlled and the relevant quality control objects, according to the identifiers of the quality control objects involved.
2. The PET / CT report quality control method based on structured data according to claim 1, characterized in that, The report information includes patient information, examination information, clinical diagnosis, brief medical history, examination findings, and diagnostic opinion; The structured data includes an examination findings section and a diagnostic opinion section. The examination findings section includes organ entries and found objects, and the diagnostic opinion section includes opinion conclusions.
3. The PET / CT report quality control method based on structured data according to claim 2, characterized in that, The process involves parsing the structured data, reading the attributes, relationships, identifiers, source text, and source text location of each object within the structured data, and combining this information with the report information to obtain the parsing results, including: Patient information, examination information, clinical diagnosis and brief medical history are read from the report information to form context objects, and the original content of each context object in the report information is retained as the source text. From the inspection view portion of the structured data, organ entries and discovery objects are read by object type, and the attributes, identifiers, source text, and position of the source text in the inspection view are read for each organ entry and discovery object. The attributes of the organ entry include organ name and positive / negative status; the attributes of the discovery object include the organ to which it belongs, abnormal manifestations, size, maximum standardized uptake value, and degree of diagnostic certainty. From the diagnostic opinion section of the structured data, the opinion conclusions are read according to the object type, and the attributes, identifiers, source texts, and positions of the source texts in the diagnostic opinion are read. The attributes include the organ referred to, the abnormal manifestation, the positive or negative result, and the degree of diagnostic certainty. Read the correlation between the diagnostic opinion section of the structured data and the findings of the examination section; The read organ entries and their attributes, identifiers, source texts and locations, the discovered objects and their attributes, identifiers, source texts and locations, the opinion conclusions and their attributes, identifiers, source texts and locations, the context objects and source texts, and the relationships are summarized into the parsing results.
4. The PET / CT report quality control method based on structured data according to claim 3, characterized in that, The quality control items also include quality control titles, problem levels, and prompt templates.
5. The PET / CT report quality control method based on structured data according to claim 4, characterized in that, The quality control object selection criteria for the rule-based quality control items select quality control objects from the parsing results, retrieve the attributes and relationships of the quality control objects, and compare them according to the judgment criteria of the rule-based quality control items to generate quality control results, including: From the analysis results, objects that meet the quality control object screening criteria are selected as quality control objects, and the attributes and relationships corresponding to the quality control objects are read. The attribute values and relationships of the read attributes are compared with the judgment conditions of the quality control items according to the rules. If the comparison result meets the judgment conditions, it is considered a hit. The quality control title, the identifier of the object involved, the problem level, the judgment basis, the source text of the object involved, and the prompt generated by filling in the attribute value or source text of the object involved according to the prompt template are recorded to obtain the quality control result. The judgment criteria for the rule quality control items include the presence or absence of the attribute values, their similarities or differences, and the presence or absence of association relationships. The criteria for determining the rule quality control items are the attribute values, relationships, and context object attribute values of the quality control objects involved.
6. The PET / CT report quality control method based on structured data according to claim 5, characterized in that, The process of selecting quality control objects from the parsing results based on the filtering criteria of the rule-based quality control items, taking the attributes and relationships of the quality control objects, and comparing them according to the judgment criteria of the rule-based quality control items, also includes: If the rule quality control item specifies a control object, the control object is selected from the parsing results according to the control object screening conditions, and the corresponding attribute value and association relationship of the control object are obtained according to the type of the control object; The attribute values of the control object and the quality control object are compared according to the judgment conditions of the rule quality control items, and the correlation of the control object is compared according to the judgment conditions.
7. The PET / CT report quality control method based on structured data according to claim 5, characterized in that, The process of selecting quality control objects from the parsing results based on the filtering criteria of the rule-based quality control items, taking the attributes and relationships of the quality control objects, and comparing them according to the judgment criteria of the rule-based quality control items, also includes: If the judgment condition of the rule quality control item involves the patient's gender, then the gender of the patient information in the context object is taken, and the gender of the patient information is compared with the judgment condition.
8. The PET / CT report quality control method based on structured data according to claim 4, characterized in that, The process involves selecting quality control objects from the parsing results based on the quality control object screening criteria of semantic quality control items, obtaining the judgment results and evidence of the quality control objects using a semantic model, verifying the source object and cited content of the evidence, and generating quality control results after passing the verification, including: From the parsing results, objects that meet the quality control object screening criteria are selected as quality control objects, and the corresponding attributes and source text of the quality control objects are read. The judgment conditions of the semantic quality control items, the attributes of the quality control objects and the source text, and the context objects and the source text are fed into the semantic model; When a control item specifies a reference object, the attributes and source text of the reference object are submitted. The semantic model determines whether the meaning expressed by the source text of the quality control object, the context object, and the comparison object conforms to the semantic quality control item based on the judgment conditions of the semantic quality control item. When the semantic quality control criteria are met, the semantic model returns several pieces of evidence that meet the semantic quality control criteria. The evidence includes cited content and source objects. The cited content includes a continuous segment of a source text of a quality control object, a control object, or a context object. The source object includes a quality control object, a control object, or a context object. The source object verification and cited content verification are performed on several pieces of evidence. Based on the several pieces of evidence that pass the source object verification and cited content verification, the quality control title, the identifiers of the quality control objects and control objects involved, the problem level, the prompts generated by filling in the attributes of the quality control objects and control objects involved or the source text according to the prompt template, and the evidence pieces used as the basis for judgment are recorded to obtain the quality control results.
9. The PET / CT report quality control method based on structured data according to claim 8, characterized in that, The process of verifying the source object and the cited content of several pieces of evidence includes: Verify whether the source of the evidence is the quality control object, control object, or context object currently submitted for judgment; Verify whether the cited content of the evidence is a continuous substring in the source text of the source object.
10. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the PET / CT report quality control method based on structured data as described in any one of claims 1-9.
