Artificial intelligence-based electronic medical record intelligent medical record quality control method, system, device and storage medium

Abstract

This invention proposes an intelligent medical record quality control method, system, device, and storage medium based on artificial intelligence. The method includes: obtaining the screen state of the electronic medical record based on the visual presentation content of the electronic medical record window and extracting semantic elements to obtain multiple medical record element categories; calculating the relative evidence strength of each medical record element category based on the number of text hits; constructing a co-occurrence association matrix among each medical record element category; solving a preset operational state parameter vector based on all relative evidence strengths and the co-occurrence association matrix to obtain a medical record operational state parameter set; obtaining a set of medical record quality control constraint rules corresponding to the medical record operational state parameter set; calculating the deviation degree of each medical record quality control constraint rule based on the medical record operational state parameter set, target element category, and rule threshold; selecting medical record quality control constraint rules with deviation degrees greater than a preset deviation threshold into a violation item set; and generating medical record quality control results based on the violation item set, thus achieving accurate medical record quality control.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, and in particular relates to an intelligent medical record quality control method, system, device and storage medium based on artificial intelligence. Background Technology

[0002] With the continuous advancement of medical informatization, electronic medical records (EMRs) have become a crucial foundation for clinical diagnosis and treatment, medical record management, and medical quality control. For a long time, quality control of EMRs has primarily relied on post-hoc quality control methods. This involves extracting, comparing, and statistically analyzing the content of completed medical records through a backend system to identify non-standard writing or omissions. However, this method struggles to reflect problems during the record creation process in a timely manner, often requiring rectification only after submission. This approach is costly and has limited guidance for clinical practice. In contrast, in-process quality control can identify potential risks in real time during the record writing process and is considered more beneficial for improving record quality and clinical standardization. However, its practical implementation faces numerous limitations. Firstly, many medical institutions still use older EMR systems. These systems have closed architectures and insufficient interface capabilities, making it difficult to provide stable and reusable data access methods to external quality control systems. This often necessitates custom development or system modification by vendors for existing in-process quality control solutions, resulting in long implementation cycles, high adaptation costs, and difficulties in large-scale deployment. On the other hand, existing in-process quality control technologies mostly rely on fixed fields or static rules as the core judgment criteria, lacking a holistic portrayal of the medical record formation process. They cannot distinguish between reasonable and unreasonable behaviors at different writing stages, easily leading to mismatches between prompts and actual writing progress. This affects both the smoothness of doctors' operations and reduces the interpretability of quality control results. In a reality where interfaces are closed and intrusion is prohibited, how to accurately perceive the medical record formation status without relying on the underlying data structure and conduct effective quality control based on the formation stage has become an urgent technical problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to design an intelligent medical record quality control method, system, device, and storage medium based on artificial intelligence, which can accurately perceive the formation status of medical records without relying on the underlying data structure, and perform effective quality control based on the formation stage.

[0004] To achieve the above objectives, a first aspect of the present invention provides an intelligent medical record quality control method based on artificial intelligence, the method comprising: Obtain the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the electronic medical record screen state. Semantic elements are extracted based on the screen state of the electronic medical record to obtain multiple medical record element categories and the number of text hits for each medical record element category. The relative evidence strength of each medical record element category is calculated based on the number of text hits, the preset regional hierarchy weight, and the preset expansion state weight. All the relative evidence strengths constitute a stage evidence vector. Based on the electronic medical record screen state, a co-occurrence association matrix is ​​constructed between each of the medical record element categories. Based on the stage evidence vector and the co-occurrence association matrix, the preset running state parameter vector is solved to obtain the medical record running state parameter set. The medical record operation status parameter set is filtered in a preset stage rule mapping table to obtain the corresponding medical record quality control constraint rule set; wherein, the medical record quality control constraint rule includes target element category and rule threshold; The deviation degree of each medical record quality control constraint rule is calculated based on the set of medical record operational parameters, the target element category, and the rule threshold. Medical record quality control constraint rules with deviation degrees greater than a preset deviation threshold are selected into the violation item set, and medical record quality control results are generated based on the violation item set.

[0005] Furthermore, the step of generating medical record quality control results based on the set of violations includes: Merge the medical record quality control constraint rules with the same target element category in the set of violations into the same violation cluster; A comprehensive score for each violation cluster is calculated based on the set of medical record operation parameters, the deviation degree, and the number of violation items in the violation cluster. The violation clusters are sorted from high to low according to the comprehensive score to obtain the sorting result. The medical record quality control result is generated based on the violation clusters and the sorting result.

[0006] Furthermore, after generating the medical record quality control results based on the set of violations, the method further includes: Obtain the target value corresponding to each violation cluster; wherein, the minimum value between the comprehensive score and the preset cutoff value of each violation cluster is used as the target value; The global strength index is obtained by summing all the target values.

[0007] Furthermore, the step of calculating the comprehensive score for each violation cluster based on the set of medical record operational parameters, the deviation degree, and the number of violation items in the violation cluster includes: The largest deviation in the violation cluster is defined as the dominant violation item deviation. Multiply the preset value by the number of violations and perform a logarithmic operation, then multiply by the preset adjustment coefficient to obtain the first data; The first data is added to the deviation of the dominant violation, and then multiplied by a preset stage stability coefficient to obtain the comprehensive score.

[0008] Furthermore, the co-occurrence association matrix includes multiple co-occurrence association elements. The step of solving the preset operational parameter vector based on the stage evidence vector and the co-occurrence association matrix to obtain the case operational parameter set includes: Calculate the L2 norm of the preset running state parameter vector to be solved and the stage evidence vector, and obtain the running state parameter vector to be solved when the L2 norm is minimized; Subtract the first element value from the second element value, square the result, and then multiply by the co-occurring associated element to obtain the second data. Add all the second data together and multiply by a preset balance coefficient, then add the vector of running parameters to be solved to obtain the set of running parameters of the medical record.

[0009] Further, the step of calculating the deviation of each medical record quality control constraint rule based on the medical record operational parameter set, the target element category, and the rule threshold includes: Perform on-screen consistency convergence on the set of medical record runtime parameters to obtain the converged set of medical record runtime parameters. Based on the target element category, extract the corresponding dimension value from the converged medical record running state parameter set; Subtract the rule threshold from the dimension value to obtain the third data, and select the largest value between zero and the third data as the target value; The deviation is obtained by multiplying the target value by a preset stage stability coefficient.

[0010] Further, the step of calculating the relative evidence strength of each of the case element categories based on the text hit count, preset regional hierarchy weight, and preset expansion state weight includes: The fourth data is obtained by weighting the text hit count, the region hierarchy weight, and the expansion state weight; The fourth data of all the case element categories are added together to obtain the fifth data. The fourth data is then divided by the fifth data to obtain the relative strength of evidence.

[0011] In a second aspect, the present invention provides an intelligent medical record quality control system based on artificial intelligence, the system comprising: The acquisition unit is used to acquire the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the screen state of the electronic medical record. The extraction unit is used to extract semantic elements based on the screen state of the electronic medical record, obtain multiple medical record element categories and the number of text hits for each medical record element category, and calculate the relative evidence strength for each medical record element category based on the number of text hits, preset regional hierarchy weights and preset expansion state weights; wherein, all the relative evidence strengths constitute a stage evidence vector. The construction unit is used to construct a co-occurrence association matrix between each of the medical record element categories based on the electronic medical record screen state, and to solve the preset running state parameter vector based on the stage evidence vector and the co-occurrence association matrix to obtain the medical record running state parameter set. The filtering unit is used to filter the medical record operation status parameter set in a preset stage rule mapping table to obtain the corresponding medical record quality control constraint rule set; wherein, the medical record quality control constraint rule includes target element category and rule threshold; The generation unit is used to calculate the deviation degree of each of the medical record quality control constraint rules based on the medical record running state parameter set, the target element category and the rule threshold, select the medical record quality control constraint rules with deviation degrees greater than the preset deviation threshold into the violation item set, and generate the medical record quality control result based on the violation item set.

[0012] In a third aspect of the invention, an electronic device is provided, the electronic device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the method described in the first aspect above.

[0013] In a fourth aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0014] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides an AI-based intelligent medical record quality control method, system, device, and storage medium. Its core lies in shifting the focus of quality control from static medical record content to the operational state of the medical record during the doctor's writing process. This allows for continuous perception and phased analysis of this operational state without altering the existing electronic medical record system's operation. This invention synchronously acquires the doctor's currently visible medical record presentation state, constructs a set of medical record operational parameters reflecting the medical record formation process, and introduces a quality control constraint mechanism matching the formation stage. This enables quality control judgments to be combined with writing progress and contextual relationships, thus avoiding misjudgments and interference caused by the lack of stage semantics in traditional technologies. Simultaneously, this invention unifies the formation stage with violations, generating medical record quality control results highly consistent with the current writing stage. This ensures that the quality control output reflects overall quality risk while pinpointing specific problems, balancing real-time performance, interpretability, and engineering feasibility. Through this technical approach, this invention provides a solution for in-process medical record quality control in interface-free environments, decoupled from existing systems and adapted to actual clinical writing behaviors. Attached Figure Description

[0015] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0016] Figure 1 This is a flowchart of an intelligent medical record quality control method based on artificial intelligence provided in an embodiment of this application.

[0017] Figure 2 This is a schematic diagram of the structure of the AI-based intelligent medical record quality control system provided in this application embodiment. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] Please refer to Figure 1 , Figure 1 This is a flowchart of an intelligent medical record quality control method based on artificial intelligence provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0020] Step S101: Obtain the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the screen state of the electronic medical record. Step S102: Semantic elements are extracted based on the screen state of the electronic medical record to obtain multiple medical record element categories and the number of text hits for each medical record element category. The relative evidence strength of each medical record element category is calculated based on the number of text hits, the preset regional level weight, and the preset expansion state weight. Among them, all relative evidence strengths constitute the stage evidence vector. Step S103: Construct a co-occurrence association matrix between each medical record element category based on the electronic medical record screen state; solve the preset operational state parameter vector based on the stage evidence vector and the co-occurrence association matrix to obtain the medical record operational state parameter set. Step S104: Based on the set of medical record operational parameters, filter the data in the preset stage rule mapping table to obtain the corresponding set of medical record quality control constraint rules; wherein, the medical record quality control constraint rules include target element categories and rule thresholds; Step S105: Calculate the deviation of each medical record quality control constraint rule based on the medical record running state parameter set, target element category, and rule threshold. Select medical record quality control constraint rules with deviation greater than the preset deviation threshold into the violation item set. Generate medical record quality control results based on the violation item set.

[0021] In step S101 of some embodiments, an electronic medical record screen state that reflects the current presentation status of the medical record is constructed based on the actual operation scenario of the doctor writing or viewing the electronic medical record. This serves as the foundational input for subsequent medical record quality control analysis. To this end, a desktop floating AI agent is deployed on the doctor's workstation. This agent runs as a resident process and triggers a data collection operation by clicking the floating AI agent when the doctor needs to perform in-process medical record quality control. This triggering operation initiates the local data collection process, enabling the system to simultaneously acquire the current presentation status of the medical record interface while the doctor is using the electronic medical record system normally.

[0022] During the data acquisition process, the suspended agent invokes the local RPA execution component to first locate the electronic medical record window currently displayed in the foreground. This window corresponds to the specific medical record page the doctor is currently working on, such as the admission record page, progress note page, or discharge summary page. Leveraging the window management capabilities provided by the operating system, the RPA obtains the display area corresponding to this window and determines the spatial boundaries of this acquisition, ensuring that the acquired content remains consistent with the doctor's current visible medical record interface. The acquisition method is matched accordingly to different operating modes of the electronic medical record system: when the electronic medical record is run as a client, image acquisition is performed on the window display area to obtain visually presented content including text, tables, and their layout relationships; when the electronic medical record is run as a browser, the browser kernel capabilities are used to obtain a description of the page structure within the currently visible area and convert it into a structured representation consistent with the visual layout. In both cases, the acquisition revolves around the content currently displayed on the screen, remaining synchronized with the doctor's real-time writing and viewing behavior.

[0023] After data collection, the obtained visual content is processed for medical record quality control purposes. This process combines typical behavioral characteristics of electronic medical record writing, distinguishing different display areas on the interface and extracting state features that reflect the current writing context. For example, on the admission record page, doctors typically prioritize filling in the chief complaint and present illness, while physical examination and auxiliary examination information are located in other areas of the page; on the progress note page, doctors often focus on editing the current progress note, while historical progress notes are collapsed. By identifying these interface structures and display states, the actual medical record paragraphs, display levels, and expanded or collapsed states presented on the current interface are transformed into state features, ensuring that the constructed results reflect the doctor's current actual writing environment.

[0024] Based on the above processing, the visual content of the current electronic medical record window is mapped to a unified state description format, and these state features are combined to form the electronic medical record screen state. The construction process can be formally represented as: ; in, This indicates the visual content of the electronic medical record window obtained by RPA within this trigger cycle, which originates from the display area of ​​the operating system window or the visible area of ​​the browser. This describes the screen state construction process, which includes three consecutive steps: interface area identification, visible content filtering, and state feature mapping. This process transforms the visually presented content into a screen state representation that describes the current medical record writing state. The electronic medical record screen state formed through this method... This describes the state of the medical record interface that the doctor sees and interacts with at any given moment.

[0025] In step S102 of some embodiments, the electronic medical record screen state This includes the visible area division of the current medical record interface, the spatial distribution of text blocks, the hierarchical structure of the regions, and the expanded or collapsed state of each region. Based on this information, we first start from... The system analyzes structural clues at the interface level to form a candidate set of elements related to the stage of medical record formation. Structural clues are divided into two categories: page-level and region-level. Page-level clues originate from the current page's title text, the names of fixed columns, and the overall page layout features, used to identify the current record type. Region-level clues originate from the arrangement and adjacency relationships of visible text blocks in the body text area, form area, and list area, used to depict how the medical record content is presented on the interface. This is achieved through a unified approach. The existing "text block-region-hierarchy" representation format in the system ensures that this step maintains a consistent processing flow across different electronic medical record systems.

[0026] Based on the interface structure analysis, semantic elements are extracted from the visible text content in the electronic medical record screen and mapped to medical record element categories closely related to the medical record formation stage. The same medical record element category can be composed of multiple text contents, and a single text content may also belong to different medical record element categories at different formation stages. Ultimately, the medical record element category serves as the basic object for quality control and stage judgment. Medical record element categories are set according to the actual needs of medical record quality control, such as record type prompts, diagnostic descriptions, medical orders or treatments, examination and test results, and disease progression descriptions. Each type of element is extracted through two parallel paths: one path is based on rule matching between interface labels and field names, such as if the text block is located under an identified column title or if the text block itself contains highly stable field prompts; the other path is based on the semantic annotation results of the sequence within the text block. The semantic annotation model is deployed in a lightweight architecture on the terminal side. Its input consists of word-segmented text sequence embeddings and position embeddings. The encoding layer uses a two-layer stack of bidirectional gated recurrent networks to capture contextual information. The output layer uses linear mapping followed by conditional random field decoding to determine the boundaries of text fragments corresponding to the aforementioned feature categories. Model inference is performed on a per-text-block basis, preserving the regional location and hierarchical information of each feature fragment in the interface for subsequent fusion with interface structural cues.

[0027] Subsequently, evidence vectors were constructed during the construction phase. Used to quantify the current screen status of electronic medical records The relative strength of evidence for each category of medical record elements. Each dimension corresponds to a category of medical record elements, and its value is determined by the electronic medical record screen state. The hit rate of the medical record element category is comprehensively calculated, and the hit strength considers the number of text hits, the regional hierarchy weight of the region, and the unfolded state weight. The number of text hits is counted based on the "candidate element fragments / field units" obtained after parsing the electronic medical record screen state: fragments from different regions or different field tags within the same element category are counted separately; repeated displays within the same region or repeated paragraphs in the template are counted once after fragment merging. Parsing the electronic medical record screen state yields the medical record element category and its set of candidate element fragments in the interface. The medical record element category is a predefined quality control object type used for subsequent stage judgment and rule verification; candidate element fragments are specific text fragments or field units identified from the screen state that may belong to a certain element category, used to provide evidence supporting that element category. The regional hierarchy weight and unfolded state weight are configurable engineering parameters derived from the hospital's common EMR page structure experience and quality control requirements. Fixed values ​​are given according to the regional type and display state, and are maintained synchronously with system version / page template updates; at runtime, values ​​are directly matched based on the regional hierarchy and unfolded / collapsed state identified in the screen state. Hit contributions from text elements located in the currently visible text area and in an expanded state have higher weight; hit contributions from text elements located in the title bar or fixed tooltip area have stable weight; and hit contributions from text elements located in collapsed areas have lower weight. To ensure the comparability of evidence under different page densities and expansion levels, the hit strength of each case element category is proportionalized to obtain relative evidence strength. A weighted count is performed based on the number of text hits, the weight of the region level, and the weight of the expanded state to obtain the fourth data point; the fourth data points for all case element categories are summed to obtain the fifth data point; the relative evidence strength is obtained by dividing the fourth data point by the fifth data point. The formula is as follows: ; in, Indicates the first The relative strength of evidence for each category of medical record element in the current electronic medical record screen state, and the relative strength of evidence for all categories of medical record element constitute the stage evidence vector. . This is the screen view for electronic medical records. This represents a weighted count formed by rule matching hits and semantic annotation hits, superimposed with region hierarchy and expanded state weights. Specifically, From electronic medical record screen status The middle belongs to the first The weighted summation of "hit fragments" from the categories of medical record elements is obtained first, that is, in Candidate segments are located within the text content using rule matching of column titles / field names. Simultaneously, a sequence labeling model is used to mark semantic segments corresponding to the medical record element category within the same text content. Segments obtained from the two paths are aligned and merged according to their region and level (segments repeatedly displayed within the same region are counted once). Then, each retained segment is multiplied by its corresponding region / level weight and expanded / collapsed state weight, and the results are summed to obtain a weighted count of the medical record element category. The number of text hits only reflects quantity. It reflects the intensity resulting from the combined effect of quantity and presentation location / visibility. This represents an index of all case element categories. This scaling allows staged evidence to reflect the element composition structure, unaffected by page length or total text volume.

[0028] In step S103 of some embodiments, after obtaining the stage evidence vector, further information is obtained from the electronic medical record screen state. Construct a co-occurrence association matrix among the categories of medical record elements. It is used to depict the spatial and structural relationships of different case element categories in the current interface. The construction is based on regional adjacency and co-occurrence relationships: when two types of medical record elements appear in the same visible area, or are located in adjacent areas and are simultaneously expanded, the corresponding association value is high; when two types of elements are on the same page but far apart, or one of them is collapsed, the corresponding association value is low. The process follows "fragment attribution area → regional relationship → element pair accumulation": first, the identified types of medical record elements are classified according to their location within the visible area... The area identifiers are categorized; then each pair of medical record elements is categorized ( , If they appear in the same visible area, then... Accumulate the contribution within the same region; if it appears in adjacent regions, then... Accumulate adjacent contributions once; during accumulation, weights are applied based on the region hierarchy and the expanded / collapsed state (expanded weights are greater than collapsed weights), and finally... Symmetry and normalization are performed to make correlation values ​​comparable under different page densities. This design reflects common patterns in clinical writing. For example, the co-occurrence of course descriptions and examination results in the same area usually indicates the supplementary stage, while the co-occurrence of diagnostic descriptions and treatment elements in adjacent areas usually indicates the implementation stage. It should be noted that the same visible area and adjacent areas are distinguished based on the interface layout parsing results in the screen state. Specifically, the area units and their boundaries in the page (e.g., title area, body text area, form area, list area and its sub-areas) are first identified. If two types of medical record elements fall within the same area unit, they are judged as the same visible area; if they fall within different area units, but the boundaries of the two areas are adjacent or the distance is within a preset threshold range, they are judged as adjacent areas. Information such as belonging to the same visible area, adjacent areas, and collapsed or expanded status are all provided by the electronic medical record screen state: the screen state already includes area division and hierarchical relationship, as well as the visibility and expanded / collapsed markers of each area during construction. Therefore, subsequent co-occurrence correlation and constraint verification directly read these markers for judgment and weighting.

[0029] Based on stage evidence vectors Co-occurrence correlation matrix Solve the set of runtime parameters for medical records. This ensures that the system maintains a response to the evidence at the current screen state stage while also reflecting the stage consistency brought about by the co-occurrence relationships of elements. This process is achieved through a quadratic objective with a co-occurrence smoothing term. The L2 norms of the preset operational state parameter vector and the stage evidence vector are calculated, and the operational state parameter vector corresponding to the minimum L2 norm is obtained. The first element value is subtracted from the second element value and squared, then multiplied by the co-occurrence association element to obtain the second data. All second data are added together and multiplied by a preset balance coefficient, then added to the operational state parameter vector to obtain the case operational state parameter set. As shown in the following formula: ; in, This is a set of runtime parameters for medical records. Let be the running-state parameter vector to be solved. This is the stage evidence vector formed by the screen state of the electronic medical record. For the co-occurrence correlation matrix, This is the balance coefficient, used for evidence vector fitting and co-occurrence consistency constraints during the balance phase. Indicates the category of medical record elements and The correlation strength of co-occurrence in the same or adjacent areas of the current screen. Represents the runtime parameter vector In the categories of medical record elements The parameter values ​​in the corresponding dimension. Represents the runtime parameter vector In the categories of medical record elements The parameter values ​​correspond to the dimensions. The first term ensures that the runtime parameter vector closely follows the current stage evidence on the interface, while the second term ensures that case element categories that co-occur on the same screen or adjacent elements maintain coordinated changes in runtime parameters, thereby mitigating fluctuations in stage evidence caused by local occlusion, folding, or template differences. This objective is a convex quadratic form and can be obtained on the terminal side through linear equation solving. The final set of runtime parameters for the medical records. The output is in a structured format, covering the current recording stage, the coverage of key medical record elements, and contextual consistency features smoothed by co-occurrence relationships. For example, when a doctor opens the admission record page and only fills in the chief complaint and present illness history, Narrative elements are prominent, and co-occurrence constraints ensure consistency with record type cues, thus... The stable response is observed during the initial recording phase; when the doctor switches to the discharge summary page and expands the diagnosis and medication summary section... The increased proportion of diagnostic descriptions and treatment elements, coupled with co-occurrence constraints strengthening the relationship between the two, makes... The process naturally transitions to the summary recording stage. In this way, the electronic medical record is presented in a screen-based manner. Transformed into a set of medical record runtime parameters that can be directly used for subsequent medical record quality control constraint verification. .

[0030] In step S104 of some embodiments, the medical record running state parameter set This includes three types of information: first, stage-related parameters, used to pinpoint the current stage interval in case formation; second, element coverage parameters, used to measure the degree of presentation of key elements in the current stage; and third, contextual consistency parameters, used to reflect the stage consistency supported by the combination of elements on the same screen. First, based on... The system determines the stage interval to which the current medical record's operational parameter set belongs based on relevant parameters, and loads the corresponding set of medical record quality control constraint rules within that stage interval. A pre-defined mapping table is used, with each stage interval having a corresponding set of medical record quality control constraint rules. The mapping table is used to find the set of medical record quality control constraint rules corresponding to the stage interval. The set of medical record quality control constraint rules is stored in structured rule format. Each medical record quality control constraint rule consists of "target element category," "constraint direction," "rule threshold," and "trigger window": the target element category corresponds to... The rules include one-dimensional or multi-dimensional feature coverage parameters; constraint direction describing the existence or strength requirements of the target feature category at this stage; rule threshold used to quantify judgment conditions; and trigger window used to characterize the operational stage range to which the rule applies. The target feature category is the category of medical record elements constrained by the rule. Medical record quality control constraint rules are compiled from medical record management specifications and hospital quality control systems before deployment, and use a unified parameter naming and indexing method to ensure compatibility with... The element dimensions are directly aligned.

[0031] In step S105 of some embodiments, before performing rule verification, a "screen-to-screen consistency convergence" is performed on the medical record runtime parameter set to ensure that constraint judgments are stable in the face of page folding, template differences, and local visibility changes. This convergence process utilizes... The internally included contextual consistency information coordinates mutually supporting element dimensions, enabling the operational parameters to exhibit more consistent structural characteristics within the same stage. Operational parameters refer to the various parameter components in the medical record operational parameter set (e.g., parameter dimensions corresponding to different medical record element categories), used to characterize the operational strength and consistency features of the current stage. This process is achieved through a target with a sparse offset term: the offset term allows a small number of elements to have structural omissions or delayed presentation on certain pages, adapting to common operational paths in clinical writing such as "writing the description first and then adding examinations" and "entering medical orders first and then adding records." The converged operational parameter vector is denoted as... The solution is as follows: ; in, This is a set of parameters for the running state of medical records. This is the vector of running parameters to be solved. Represents the set of runtime parameters based on medical records. The feature association weight, calculated from the context consistency parameter, is used to describe the feature association weight. With elements This section describes the support relationships within the current runtime structure. First, the various medical record element categories identified in the electronic medical record screen are categorized according to their respective interface regions. Then, for any two categories of medical record elements, the cases where they appear simultaneously in the same visible area and in adjacent areas are statistically analyzed. Different weights are assigned to these occurrences based on the region hierarchy and whether the elements are expanded or collapsed, and these weights are accumulated to obtain the association strength between the two categories of medical record elements within the current interface structure. Finally, the association strength is normalized to obtain the element association weight. This allows for direct comparison of results under different page densities. Used to balance "fitting the current operating state" and "element consistency convergence". Used to control the effect of sparse offset terms, allowing a small number of dimensions to be offset to accommodate clinical page differences. This represents the parameter value of the running state parameter vector p in the dimension corresponding to the feature category i. This represents the parameter value of the running state parameter vector p in the dimension corresponding to the feature category j. Represents the set of runtime parameters for medical records. The parameter values ​​for the corresponding dimension of the medical record element category are the baseline parameter values ​​obtained in the previous stage, used to characterize the intensity of the medical record element category in the current operating state. This objective is convex and can be quickly solved on the terminal side by combining iterative thresholding and linear equations. .

[0032] get Then, each medical record quality control constraint rule is validated and a violation judgment is generated. Each rule is indexed by its target element category from... The corresponding dimension values ​​are extracted, and then the degree of fulfillment of the rule in the current operational state (medical record operational state parameter set) is calculated by combining the rule threshold and constraint direction. The dimension values ​​are obtained by retrieving vector components based on the target element category index in the rule: each rule in the rule base stores an element category index, and the system has already fixedly mapped each medical record element category to a certain dimension of the parameter vector when generating the operational state parameter set. Therefore, during verification, the corresponding component is directly retrieved from the converged parameter vector using this index as the corresponding dimension value. To make the violations more in line with the actual needs of quality control rectification, the degree of violation of each rule is expressed as a continuous "deviation degree." The deviation degree considers both the degree of insufficiency of the target element dimension and the stability coefficient of the operational state stage, so that the severity of the same element's absence varies at different stages. This deviation is used to determine whether a rule constitutes a violation and the priority of the violation. It is calculated by extracting the corresponding dimension value from the converged set of case operation parameters based on the target element category; subtracting the dimension value from the rule threshold to obtain the third data; selecting the largest value between zero and the third data as the target value; and multiplying the target value by a preset stage stability coefficient to obtain the deviation. The formula is shown below: ; in, This indicates the first rule in the set of medical record quality control constraint rules. The deviation of the rule in the current running state. This represents the rule threshold. Representation rules Corresponding target feature dimension index, These are the dimension values ​​extracted from the converged set of operational parameters of medical records based on the target element category. This represents the stage stability coefficient calculated from stage-related parameters in the medical record's operational parameter set. It reflects the stability of the current operational state within the applicable stage interval of the rule. The stage stability coefficient first determines the most likely stage interval based on the stage-related parameters, and then calculates the dominance range of that stage interval. For example, it takes the difference between the parameter strength corresponding to the current stage and the parameter strength of the second-highest stage, and then maps this difference to a fixed interval to form the stage stability coefficient. The larger the dominance range, the more stable the stage determination, and the higher the resulting stage stability coefficient. A positive deviation indicates that the rule has triggered a violation; a deviation of zero indicates that the rule satisfies the constraint. By introducing the stage stability coefficient, deviations in the same dimension yield higher deviations when the stage determination is more stable, thus more closely aligning with the execution logic in clinical quality control that "the more clearly defined the stage, the stricter the constraint should be." It should be noted that the trigger window is used to limit the applicable stage interval of the rule: first, the current stage interval is determined using the medical record's operational parameter set, and then only rules covering that stage interval are loaded for validation and deviation calculation.

[0033] Based on the above deviation calculation, all rule instances that meet the triggering conditions are summarized into a set of violations. The preset deviation threshold is set to 0; a deviation greater than 0 (positive deviation) indicates that the trigger condition is met. Each violation (i.e., a rule that meets the trigger condition) is saved in a structured record format, containing a rule identifier, target element category index, deviation degree, and relevant marker information for the current stage, enabling subsequent steps to directly address the issue. It can generate organizational output without repeatedly executing rule matching. For example, in a clinical scenario, when a doctor is on the admission record page and has completed the chief complaint and present illness history, but the physical examination section is not yet reflected in the current running state, the relevant "key element coverage in the initial record stage" rule will be executed. The deviation is reflected when the corresponding dimension is below the threshold. If the value is positive, a violation is generated. When the doctor is on the discharge summary page and the diagnosis and treatment elements are displayed, but the examination and test summary elements are missing, the rule is also triggered in the form of deviation degree, and a higher priority is formed due to the higher stage stability coefficient. In this way, this step sets the medical record running parameters. Convert into a set of violations that can be directly used for subsequent quality control result organization. This completes the key implementation step from "forming a state description" to "constraining the location of violations".

[0034] Furthermore, it receives the set of operational parameters for medical records. With the set of violations Generate medical record quality control results This step involves organizing a quality control process for the current writing stage. Previous steps transformed the electronic medical record screen state into a set of runtime parameters describing the medical record's formation process, mapping this set of runtime parameters to a set of violations corresponding to stage constraints. This step further unifies "stage position" and "violation facts" into a single result object, ensuring that the quality control output reflects both the current stage's context and a list of specific, executable violations. In real-world clinical scenarios, doctors often switch between admission records, progress notes, discharge summaries, and other pages rapidly. The quality control results need to consistently express "which formation stage is currently in," while simultaneously merging and sorting multiple violations within the same stage to form a complete quality control output.

[0035] Specifically, starting with the set of operational parameters of medical records The process extracts stage-related parameters and context consistency parameters to generate stage intervals and stage stability coefficients for the current running state, placing the set of violations within a specific stage context. The generation of stage intervals uses record-type-related dimensions in P as the primary factor, corrected by structural features covering key elements, ensuring stage labels reflect the "formation process" rather than just page names. The stage stability coefficient characterizes the consistency of P at the current stage, allowing subsequent organization of violations to automatically adjust with stage stability. Subsequently, stage consistency merging is performed on V, merging violations with the same target element category or strong correlation into the same violation cluster to avoid repeated representation of the same type of problem under interface collapse or template differences. This merging process uses... The context consistency parameter is used as the basis for association: when the target element categories corresponding to two violations are in When the reflected operational structure shows strong correlations, they are grouped into the same cluster, and the violation with the largest deviation is retained as the dominant violation. The deviation of the dominant violation is the dominant violation deviation, and the rest are considered supplementary violations. The strong correlation determination is directly performed using the element correlation weight matrix derived from the medical record operational parameter set: first, the target element category index corresponding to each violation is extracted; then, the weight value at the intersection of these two indices is read from the element correlation weight matrix, and this weight value is used as the correlation strength between the two violations; next, this correlation strength is compared with a preset threshold. If the threshold is reached or exceeded, the two violations are considered to have a strong correlation and are grouped into the same violation cluster; otherwise, they remain in separate clusters. The threshold is a configurable parameter set according to the differences in hospital templates. By default, the high-segment boundary value can be selected as the threshold based on the historical statistical distribution of correlation strength.

[0036] After merging, a comprehensive score is calculated for each violation cluster for ranking and intensity expression. This ensures that the quality control results highlight key issues requiring priority while maintaining a consistent output scale within the same phase. The comprehensive score considers the deviation of the dominant violation within the cluster, the phase stability coefficient, and "intra-cluster consistency." Intra-cluster consistency reflects the concentration of similar violations within the current operational structure: when multiple related elements deviate simultaneously, it usually indicates a more significant phase gap. The comprehensive score is calculated by multiplying a preset value by the number of violations and performing a logarithmic operation, then multiplying by a preset adjustment coefficient to obtain the first data point (which can be set to 1). The first data point is then added to the deviation of the dominant violation and multiplied by the preset phase stability coefficient to obtain the comprehensive score. The formula is shown below: ; in, Indicates the first The overall score of each non-compliant cluster; This represents the set of runtime parameters of the medical record. The obtained stage stability coefficient; For the first The deviation of the dominant violation item in the i-th violation cluster represents the deviation of the i-th violation cluster. The deviation of the dominant violation item with the largest deviation within a violation cluster; Indicates the first The number of violations within a single violation cluster after merging; This is an adjustment coefficient used to adjust the impact of intra-cluster consistency. This item is used to reflect the scenario of "simultaneous occurrence of similar violations" without introducing large numerical fluctuations, so that the common clinical writing phenomenon of "multiple elements being missing at the same time in a certain stage" can be reflected in the result ranking.

[0037] The violation clusters were then sorted from highest to lowest based on their comprehensive scores, and the results were combined with the current stage of the structured medical record quality control results organized by labels. Specifically, The generation process uses stage labels as the top-level field and sorted violation clusters as the detailed list field. At the same time, each violation cluster retains the target element category of the dominant violation, the corresponding rule identifier, and the deviation information, and includes a brief index of supplementary items within the cluster to support subsequent positioning.

[0038] In order to form an overall expression of quality control intensity, the quality control results of medical records are generated. During the process, the global intensity index is further calculated. and use this indicator as The constituent fields are used to characterize the overall quality risk level of medical records at the current stage. The global intensity index is obtained by truncating and summing the comprehensive scores of violation clusters, so that multiple high-risk clusters have a cumulative effect on the overall intensity, while avoiding the distortion of the overall intensity caused by a large number of low-risk clusters. Its calculation method is to obtain the target value corresponding to each violation cluster; where the minimum value between the comprehensive score of each violation cluster and the preset truncation value is used as the target value; all target values ​​are added together to obtain the global intensity index. As shown in the following formula: ; in, This represents the overall strength index of the quality control results of this medical record; Indicates the first The overall score of each non-compliant cluster; The preset cutoff value represents the upper limit of the score for a single cluster, used to limit the dominant effect of a single violating cluster on the overall strength, so that the overall strength better reflects the true risk structure of "multiple problems overlapping". This global strength index, together with the details of violating clusters, constitutes... The core fields enable the quality control results to simultaneously support overall control and specific rectification positioning.

[0039] Using a clinical writing scenario as an example, when a doctor is in the process of creating an admission record and has completed the narrative elements, but key physical signs and auxiliary examination elements have not yet been presented, This will reflect the stability of this stage. Deviations from the corresponding rules form one or more violation clusters; if the physical signs and auxiliary inspections are strongly correlated in the operational structure, they are merged to form highly consistent violation clusters. The item allows the cluster to obtain a higher ranking, thus in Prioritize presentation of information in the middle section. For example, when a doctor is in the discharge summary stage, and both the diagnostic description and the treatment summary are insufficient, multiple related violations are merged and form a higher overall score. The stage stability coefficient further amplifies its weight, making... global strength index Together with the detailed sorting, it reflects the quality risks at this stage.

[0040] Steps S101 to S105 of this embodiment involve acquiring the visual content of the electronic medical record window, constructing a screen state from the visual content, and obtaining the electronic medical record screen state. Semantic element extraction is performed based on the electronic medical record screen state to obtain multiple medical record element categories and the text hit count for each category. The relative evidence strength of each medical record element category is calculated based on the text hit count, a preset regional hierarchy weight, and a preset expansion state weight. All relative evidence strengths constitute a stage evidence vector. A co-occurrence association matrix is ​​constructed between each medical record element category based on the electronic medical record screen state. A preset operational state parameter vector is solved based on the stage evidence vector and the co-occurrence association matrix to obtain a medical record operational state parameter set. The medical record operational state parameter set is then filtered in a preset stage rule mapping table to obtain a corresponding medical record quality control constraint rule set. The medical record quality control constraint rules include target element categories and rule thresholds. The deviation of each medical record quality control constraint rule is calculated based on the set of medical record operation parameters, target element categories, and rule thresholds. Medical record quality control constraint rules with deviations greater than the preset deviation threshold are selected into the violation item set. Medical record quality control results are generated based on the violation item set. This enables accurate perception of the medical record formation status and effective quality control based on the formation stage without relying on the underlying data structure.

[0041] Please see Figure 2 This application also provides an AI-based intelligent medical record quality control system, which can implement the above-mentioned AI-based intelligent medical record quality control method. The system includes: The acquisition unit 201 is used to acquire the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the screen state of the electronic medical record. Extraction unit 202 is used to extract semantic elements based on the screen state of electronic medical records, obtain multiple medical record element categories and the number of text hits for each medical record element category, and calculate the relative evidence strength for each medical record element category based on the number of text hits, preset regional level weights and preset expansion state weights; wherein, all relative evidence strengths constitute a stage evidence vector. The construction unit 203 is used to construct a co-occurrence association matrix between each medical record element category based on the screen state of the electronic medical record, and to solve the preset operational state parameter vector based on the stage evidence vector and the co-occurrence association matrix to obtain the medical record operational state parameter set. The filtering unit 204 is used to filter the medical record operation parameters in a preset stage rule mapping table to obtain the corresponding medical record quality control constraint rule set; wherein, the medical record quality control constraint rule includes target element category and rule threshold; The generation unit 205 is used to calculate the deviation of each medical record quality control constraint rule based on the medical record running state parameter set, target element category and rule threshold, select medical record quality control constraint rules with deviation greater than the preset deviation threshold into the violation item set, and generate medical record quality control results based on the violation item set.

[0042] The specific implementation of the AI-based intelligent medical record quality control system is basically the same as the specific implementation of the AI-based intelligent medical record quality control method described above, and will not be repeated here.

[0043] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An intelligent medical record quality control method based on artificial intelligence, characterized in that, The method includes: Obtain the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the electronic medical record screen state. Semantic elements are extracted based on the screen state of the electronic medical record to obtain multiple medical record element categories and the number of text hits for each medical record element category. The relative evidence strength of each medical record element category is calculated based on the number of text hits, the preset regional hierarchy weight, and the preset expansion state weight. All the relative evidence strengths constitute a stage evidence vector. Based on the electronic medical record screen state, a co-occurrence association matrix is ​​constructed between each of the medical record element categories. Based on the stage evidence vector and the co-occurrence association matrix, the preset running state parameter vector is solved to obtain the medical record running state parameter set. The medical record operation status parameter set is filtered in a preset stage rule mapping table to obtain the corresponding medical record quality control constraint rule set; wherein, the medical record quality control constraint rule includes target element category and rule threshold; The deviation degree of each medical record quality control constraint rule is calculated based on the set of medical record operational parameters, the target element category, and the rule threshold. Medical record quality control constraint rules with deviation degrees greater than a preset deviation threshold are selected into the violation item set, and medical record quality control results are generated based on the violation item set.

2. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 1, characterized in that, The process of generating medical record quality control results based on the set of violations includes: Merge the medical record quality control constraint rules with the same target element category in the set of violations into the same violation cluster; A comprehensive score for each violation cluster is calculated based on the set of medical record operation parameters, the deviation degree, and the number of violation items in the violation cluster. The violation clusters are sorted from high to low according to the comprehensive score to obtain the sorting result. The medical record quality control result is generated based on the violation clusters and the sorting result.

3. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 2, characterized in that, After generating the medical record quality control results based on the set of violations, the method further includes: Obtain the target value corresponding to each violation cluster; wherein, the minimum value between the comprehensive score and the preset cutoff value of each violation cluster is used as the target value; The global strength index is obtained by summing all the target values.

4. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 2, characterized in that, The step of calculating a comprehensive score for each violation cluster based on the set of operational parameters of the medical record, the deviation degree, and the number of violations in the violation cluster includes: The largest deviation in the violation cluster is defined as the dominant violation item deviation. Multiply the preset value by the number of violations and perform a logarithmic operation, then multiply by the preset adjustment coefficient to obtain the first data; The first data is added to the deviation of the dominant violation item, and then multiplied by a preset stage stability coefficient to obtain the comprehensive score.

5. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 1, characterized in that, The co-occurrence association matrix includes multiple co-occurrence association elements. The process of solving a preset operational parameter vector based on the stage evidence vector and the co-occurrence association matrix yields a set of medical record operational parameters, including: Calculate the L2 norm of the preset running state parameter vector to be solved and the stage evidence vector, and obtain the running state parameter vector to be solved when the L2 norm is minimized; Subtract the first element value from the second element value, square the result, and then multiply by the co-occurring associated element to obtain the second data. Add all the second data together and multiply by a preset balance coefficient, then add the vector of running parameters to be solved to obtain the set of running parameters of the medical record.

6. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 1, characterized in that, The step of calculating the deviation of each medical record quality control constraint rule based on the medical record operational parameter set, the target element category, and the rule threshold includes: Perform on-screen consistency convergence on the set of medical record runtime parameters to obtain the converged set of medical record runtime parameters. Based on the target element category, extract the corresponding dimension value from the converged medical record running state parameter set; Subtract the rule threshold from the dimension value to obtain the third data, and select the largest value between zero and the third data as the target value; The deviation is obtained by multiplying the target value by a preset stage stability coefficient.

7. The intelligent medical record quality control method based on artificial intelligence for electronic medical records according to claim 1, characterized in that, The calculation of the relative evidence strength for each of the medical record element categories based on the text hit count, preset regional hierarchy weights, and preset expansion state weights includes: The fourth data is obtained by weighting the text hit count, the region hierarchy weight, and the expansion state weight; The fourth data of all the case element categories are added together to obtain the fifth data. The fourth data is then divided by the fifth data to obtain the relative strength of evidence.

8. An intelligent medical record quality control system based on artificial intelligence, characterized in that: The system includes: The acquisition unit is used to acquire the visual content of the electronic medical record window, construct the screen state of the visual content, and obtain the screen state of the electronic medical record. The extraction unit is used to extract semantic elements based on the screen state of the electronic medical record, obtain multiple medical record element categories and the number of text hits for each medical record element category, and calculate the relative evidence strength for each medical record element category based on the number of text hits, preset regional hierarchy weights and preset expansion state weights; wherein, all the relative evidence strengths constitute a stage evidence vector. The construction unit is used to construct a co-occurrence association matrix between each of the medical record element categories based on the electronic medical record screen state, and to solve the preset running state parameter vector based on the stage evidence vector and the co-occurrence association matrix to obtain the medical record running state parameter set. The filtering unit is used to filter the medical record operation status parameter set in a preset stage rule mapping table to obtain the corresponding medical record quality control constraint rule set; wherein, the medical record quality control constraint rule includes target element category and rule threshold; The generation unit is used to calculate the deviation degree of each of the medical record quality control constraint rules based on the medical record running state parameter set, the target element category and the rule threshold, select the medical record quality control constraint rules with deviation degrees greater than the preset deviation threshold into the violation item set, and generate the medical record quality control result based on the violation item set.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the intelligent medical record quality control method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent medical record quality control method for electronic medical records based on artificial intelligence as described in any one of claims 1 to 7.

Images