CN-122000025-A - Physical examination abnormal data identification method and system for artificial intelligent main examination

Abstract

The embodiment of the application provides a physical examination abnormal data identification method and system of an artificial intelligent main sample, which belong to the technical field of intelligent medical treatment, wherein a physical examination data scene association set is generated by firstly associating a full-dimensional physical examination data set and a physical examination scene information set of physical examination objects, then a scene physical examination characteristic association set is obtained by carrying out characteristic scene association processing on the physical examination data scene association set, then a multi-round characteristic interaction analysis is carried out on the scene physical examination characteristic association set by calling an artificial intelligent main sample model with a dynamic interaction mechanism to generate a characteristic dynamic interaction result, an abnormal characteristic propagation chain is constructed based on the result, feature nodes meeting clinical abnormal association rules are screened to form a physical examination abnormal characteristic aggregation set, and finally the physical examination abnormal data aggregation set is matched with a clinical diagnosis path template to generate a physical examination abnormal data identification report, so that physical examination abnormal data is comprehensively and accurately identified.

Inventors

• ZHANG XIAOYONG
• ZHAO ZHIJIAN
• LIU YUXIN
• Zeng Junjin
• KONG ZHENFENG

Assignees

• 广东康软科技股份有限公司

Dates

Publication Date

20260508

Application Date

20260121

Claims (10)

1. 1. A method for identifying physical examination abnormal data of an artificial intelligent master sample, the method comprising: Associating a full-dimensional physical examination data set of physical examination objects with a physical examination scene information set to generate a physical examination data scene association set, wherein the full-dimensional physical examination data set comprises current sub-item detection data, historical continuous detection data and basic health record data, and the physical examination scene information set comprises detection equipment running states, detection environment parameters and detection operation specification records; Performing feature scene association processing on the physical examination data scene association set to obtain a scene physical examination feature association group, wherein the scene physical examination feature association group comprises coupling features of detection data and scene information, time sequence evolution features of the same detection item and interaction features of cross-item data; invoking an artificial intelligent main inspection model with a dynamic interaction mechanism, performing multi-round feature interaction analysis on the scene physical examination feature association group, and generating a feature dynamic interaction result, wherein the artificial intelligent main inspection model comprises a scene feature analysis module, a feature interaction engine and a decision feedback unit; Constructing an abnormal feature propagation chain based on the feature dynamic interaction result, screening feature nodes meeting clinical abnormal association rules in the abnormal feature propagation chain, and forming a physical examination abnormal feature aggregation set; Matching the physical examination abnormal characteristic aggregation set with a corresponding clinical diagnosis and treatment path template to generate a physical examination abnormal data identification report containing abnormal association logic, clinical reference basis and follow-up examination suggestion.
2. 2. The method for identifying abnormal physical examination data of an artificial intelligent main examination object according to claim 1, wherein the step of associating the full-dimensional physical examination data set and the physical examination scene information set of the physical examination object to generate the physical examination data scene association set comprises the steps of: Extracting detection item identifiers, detection values and detection completion time of the sub-item detection data in the full-dimension physical examination data set, detecting time sequences and value change records of historical continuous detection data, classifying and integrating three types of data according to the detection item identifiers to form an item data unit group, wherein the age distribution information, the past medical history description and the genetic medical history labeling of basic health record data are obtained; Extracting the running state data of the detection equipment corresponding to each detection item in the physical examination scene information set, wherein the running state data comprises equipment starting time, calibration records and real-time running parameters, the detection environment parameters comprise detection space temperature, humidity and electromagnetic interference records, the detection operation specification records comprise operator qualification information and operation flow execution node records, and a scene information sequence is formed by sequencing according to detection completion time; Matching the item data unit group with an item with the same time stamp in the scene information sequence by taking the detection item identification and the detection completion time as double-association keys to obtain an item scene matching pair, wherein the item scene matching pair comprises full-dimension data and scene information of the same detection activity; Performing association verification on project scene matching pairs, comparing a calibration record in the running state of the detection equipment with stability characteristics of detection values, reserving the matching pairs with complete calibration records and satisfactory value stability characteristics, and removing the matching pairs with abnormal calibration or no scene interpretation of value fluctuation; Adding a unique identification of a physical examination object for the reserved project scene matching pair, grouping according to the system attribution of the detection project, sequencing in each group according to the detection completion time, and generating a physical examination data scene association set containing the identification, grouping, time sequence and complete data; And extracting common scene characteristics of each group in the physical examination data scene association set, adding the common scene characteristics as group labels to the corresponding groups, realizing hierarchical association of scene information, and perfecting the structure of the physical examination data scene association set.
3. 3. The method for identifying abnormal physical examination data of an artificial intelligent main examination object according to claim 1, wherein the step of performing feature scene association processing on the physical examination data scene association set to obtain a scene physical examination feature association group comprises the steps of: Aiming at each item scene matching pair in the physical examination data scene association set, extracting association features of detection values and equipment operation parameters, and generating equipment data coupling features by analyzing the synchronism of equipment real-time operation parameter changes and the detection values, wherein the equipment data coupling features comprise association directions of parameter changes and value changes and synchronous degree descriptions; Extracting the association characteristics of the detection values and the environment parameters, comparing the value differences of the same detection item under different environment parameters, and generating environment data coupling characteristics by combining clinically known environment influence rules, wherein the environment data coupling characteristics comprise influence attributes and influence range descriptions of the environment parameters on the detection values; Performing time sequence analysis on historical continuous detection data and current detection values of the same detection item, extracting an evolution track of the values in a time dimension, and generating time sequence evolution features by combining flow differences in detection operation specification records, wherein the time sequence evolution features comprise a value long-term change trend, a short-term fluctuation mode and change attributes caused by the flow differences; Performing interaction analysis on cross-project data in the same system attribution group, and referring to medical history information in basic health record data, identifying the interaction relation among different detection project values, and generating cross-project interaction characteristics, wherein the cross-project interaction characteristics comprise a cooperative change mode of the inter-project values and influence logic under medical history association; Binding the device data coupling features and the environment data coupling features of the same item scene matching pair to form scene coupling feature groups, wherein each scene coupling feature group corresponds to a unique detection activity; Integrating the scene coupling feature group, the time sequence evolution feature and the cross-item interaction feature according to the detection item identification and the system attribution grouping, and establishing association indexes among different features through the system attribution grouping label to generate a scene physical examination feature association group; And removing repeated feature descriptions and isolated features without associated indexes in the scene physical examination feature association group, so that each feature can trace back to corresponding detection items, scene information and health record data through the indexes to form the scene physical examination feature association group.
4. 4. The method for identifying abnormal data of physical examination of an artificial intelligent main examination according to claim 1, wherein the step of calling an artificial intelligent main examination model with a dynamic interaction mechanism, performing multi-round feature interaction analysis on a scene physical examination feature association group, and generating a feature dynamic interaction result comprises the steps of: inputting the scene physical examination feature association group into a scene feature analysis module of the artificial intelligent main examination model, extracting scene information elements in each feature, and carrying out scene suitability analysis on the scene coupling feature group by combining a clinical scene influence knowledge base to generate scene adaptation coefficients and scene influence description; Integrating the scene adaptation coefficient and the scene influence description with the corresponding time sequence evolution feature and cross-project interaction feature to generate initial feature interaction packages, wherein each initial feature interaction package comprises complete features of a single system attribution group and scene analysis results; Inputting the initial feature interaction package into a feature interaction engine of an artificial intelligent master detection model, triggering first-round interaction of features in the same system attribution group based on a preset feature interaction rule, and generating a first-round interaction correlation matrix by calculating correlation strength of cross-project interaction features and time sequence evolution features; the feature interaction engine calls a decision feedback unit to perform preliminary evaluation on the first-round interaction association matrix, extracts feature association pairs with association strength meeting a preset threshold in the first-round interaction association matrix, and combines scene influence description to judge the clinical rationality of the feature association pairs to generate first run evaluation results and interaction adjustment suggestions; The feature interaction engine adjusts feature weight distribution according to the interaction adjustment suggestion, triggers a second round of feature interaction, emphasizes feature association pairs with clinical rationality in first round evaluation, introduces association features of different system attribution groups to perform cross-system interaction analysis, and generates a cross-system interaction association matrix; The decision feedback unit evaluates the cross-system interaction incidence matrix, verifies the logic consistency of cross-system feature association by combining the medical history information in the basic health record data, and generates two-round evaluation results and feature supplementing requirements; If the feature supplement requirement exists, the scene feature analysis module extracts the supplement features in the physical examination data scene association set in a targeted manner, the input feature interaction engine performs the final interaction perfection, and if the supplement is not needed, the multiple interaction results are directly integrated; and integrating the interaction incidence matrix, the evaluation result and the scene influence description of each round, and sequencing from high to low according to the feature incidence strength to generate a feature dynamic interaction result comprising feature incidence logic, scene adaptation analysis and clinical rationality verification results.
5. 5. The method for identifying abnormal physical examination data of an artificial intelligent main sample according to claim 1, wherein the steps of constructing an abnormal feature propagation chain based on a feature dynamic interaction result, screening feature nodes meeting a clinical abnormal association rule in the chain to form a physical examination abnormal feature aggregation set comprise: Analyzing feature association logic in the feature dynamic interaction result, taking each feature with independent clinical significance as an initial node, taking other features with association relation with the initial node as association nodes, and constructing a feature association network according to the order of the association strength from high to low; Identifying core feature nodes in a feature association network, wherein the core feature nodes are features with the largest number of association nodes and the largest sum of association strengths, and extend along the decreasing direction of the association strengths by taking the core feature nodes as starting points to construct a plurality of feature propagation paths; Carrying out continuity verification on each characteristic propagation path, checking whether the associated logic of adjacent characteristic nodes in the characteristic propagation paths accords with clinical abnormal development rules, reserving the characteristic propagation paths with logic continuity, and removing the characteristic propagation path fragments with logic fracture to form an initial abnormal characteristic propagation chain; Introducing a clinical abnormal association rule set, wherein the clinical abnormal association rule set comprises abnormal feature association modes summarized based on a large number of clinical cases, and each abnormal feature association mode determines a combination mode and an association sequence of feature nodes; matching the initial abnormal feature propagation chain with a clinical abnormal association rule set, screening out feature node combinations conforming to any abnormal association mode in the abnormal feature propagation chain, and marking feature nodes in the feature node combinations conforming to any abnormal association mode as candidate abnormal nodes; Carrying out association confirmation on the candidate abnormal nodes, analyzing the association relation between each candidate abnormal node and the scene influence description, judging whether scene factors possibly cause the characteristic nodes to present abnormal performance, and eliminating pseudo abnormal nodes caused by scene interference; carrying out association confirmation on the candidate abnormal nodes, analyzing the association relation between each candidate abnormal node and the scene influence description, judging whether scene factors possibly cause the feature to present abnormal performance, and eliminating pseudo abnormal nodes caused by scene interference; retaining candidate abnormal nodes which are not pseudo-abnormal, reconstructing a characteristic propagation path taking the candidate abnormal nodes as cores, and supplementing abnormal change description in time sequence evolution characteristics corresponding to the candidate abnormal nodes to form a complete abnormal characteristic propagation chain; Extracting all feature nodes in an abnormal feature propagation chain, classifying according to system attribution groups, sorting the feature nodes in the same group according to association strength, sorting different groups according to clinical importance, and adding abnormal expression description, scene influence analysis and associated node information for each feature node; And aggregating the classified and ordered feature nodes and the corresponding information, removing the repeated nodes and the redundant description, and forming a physical examination abnormal feature aggregation set.
6. 6. The method for identifying abnormal physical examination data of an artificial intelligent main examination object according to claim 4, wherein the feature interaction engine adjusts feature weight distribution according to interaction adjustment suggestions, triggers a second round of feature interaction, emphasizes feature association pairs with clinical rationality in first round evaluation, introduces association features of different system attribution groups to perform cross-system interaction analysis, and generates a cross-system interaction association matrix, comprising: Analyzing the interaction adjustment suggestions in the first round of evaluation results, extracting the feature association pairs to be enhanced and the feature association pairs to be weakened, which are determined in the interaction adjustment suggestions, distributing higher weight coefficients for the feature nodes corresponding to the feature association pairs to be enhanced, distributing weight coefficient values for the feature nodes corresponding to the feature association pairs to be weakened to be set as a set percentage of initial weight coefficient values, and keeping the initial weight coefficients for the feature nodes which are not mentioned; Updating a feature weight matrix in the feature interaction engine based on the adjusted weight coefficient, so that the importance degree and the associated priority of each feature node are accurately reflected by the weight matrix; Triggering first round of sub-interaction of the second round of feature interaction, inputting the same system attribution grouping feature after updating the weight into an interaction calculation module, and mainly calculating interaction coefficients between the feature association pairs to be enhanced, wherein the interaction coefficients are comprehensively generated by combining the weight coefficients, the association strength and the scene adaptation coefficients to generate enhanced interaction results; extracting feature interaction packages of different system attribution groups in the physical examination data scene association set, and screening features which have clinical association with the current system attribution groups to be used as cross-system association features; Preprocessing cross-system associated features, extracting feature elements related to the current system strengthening feature association pair, supplementing corresponding scene influence description and weight coefficients, and generating a cross-system feature input packet; triggering cross-system sub-interactions of the second round of feature interactions, fusing the enhanced interaction result with a cross-system feature input package, calculating the association strength between the current system enhanced feature association pair and the cross-system feature, and generating a cross-system association matrix first draft by combining weight coefficients and scene adaptation coefficients of the two parties; Invoking a logic verification sub-module in the feature interaction engine to verify clinical logic of feature association in the cross-system association matrix manuscript, and performing dimension normalization on the cross-system association matrix manuscript subjected to logic verification to generate a final cross-system interaction association matrix; Adding interaction process description for the cross-system interaction incidence matrix, and determining screening basis, weight adjustment condition and logic verification result of the cross-system characteristics.
7. 7. The method for identifying abnormal data for physical examination of an artificial intelligent master sample according to claim 5, wherein the introducing a set of clinical abnormality association rules comprises abnormality feature association patterns based on a plurality of clinical case summaries, each abnormality feature association pattern determining a combination manner and an association order of feature nodes, comprising: Collecting case data in a multi-center clinical physical examination abnormal case database, wherein each case data comprises physical examination abnormal feature combination, an abnormal diagnosis result, a pathological development process and treatment scheme information, performing duplication elimination processing on the case data, and reserving typical case data; Extracting the characteristics of the typical case data, extracting core characteristics and associated secondary characteristics which lead to abnormal diagnosis results from each typical case, determining the occurrence sequence of the core characteristics and the secondary characteristics and the mutual influence relationship between the core characteristics and the secondary characteristics, and generating a case characteristic sequence; Inputting the case feature sequences into a clinical rule mining module, analyzing the combination rule of features in a large number of case feature sequences by using a correlation rule mining algorithm, and extracting feature combinations with the occurrence frequency higher than a preset threshold value as candidate abnormal feature correlation modes; Evaluating the candidate abnormal characteristic association modes, judging the clinical effectiveness and applicability of each candidate mode by combining with the latest clinical diagnosis and treatment guide, removing the candidate modes which do not accord with clinical reality, and reserving the effective modes which are verified by experts; carrying out structural description on the effective modes, determining the names of the characteristic nodes, the combination modes of the characteristic nodes, the association sequence among the characteristic nodes and the association strength range contained in each effective mode, and forming a mode structural description; Adding a corresponding clinical case reference for each effective mode, determining a typical case from which the effective mode is derived, and diagnosing application effects of the effective mode in the typical case; The effective modes are classified according to the attribution of the physical examination system, and each classification is ordered according to the clinical common degree of the modes to form a clinical abnormality association rule set with ordered classification.
8. 8. The method for identifying abnormal physical examination data of an artificial intelligent main examination object according to claim 3, wherein the step of performing a time sequence analysis on the historical continuous detection data and the current detection value of the same detection item, extracting an evolution track of the value in a time dimension, and generating a time sequence evolution feature by combining a flow difference in a detection operation specification record comprises the following steps: Extracting historical continuous detection data and current detection values of the same detection item, and arranging the historical continuous detection data and the current detection values according to the sequence of the detection time to form a detection value time sequence, wherein each value corresponds to a unique detection time point and detection operation specification record; Comparing the numerical value differences of two adjacent detection time points in the time sequence, recording the change direction and the change amplitude of the differences, generating a numerical value change sequence, and simultaneously marking flow difference points in the detection operation specification record corresponding to each change; the numerical value change sequence is segmented, the flow difference points are taken as segmentation nodes, the numerical value change sequence is divided into a plurality of flow consistent sections, and the detection operation specifications in each flow consistent section have no obvious difference, so that the correlation between the numerical value change and the flow can be analyzed conveniently; Analyzing the change rule of the numerical value in each flow consistent section, calculating the change frequency and the average change amplitude of the numerical value in the section, identifying the change mode in the section, and generating the change characteristic in the section; Comparing the intra-segment change characteristics of the consistent segments of different processes, analyzing the influence of the process difference points on the numerical value change mode, recording the expression form and the association degree of the influence, and generating process influence characteristics; drawing a numerical evolution curve based on the detection numerical sequence, extracting the number of inflection points, the positions of the inflection points and the numerical variation condition at the inflection points of the curve, wherein the inflection points are time points at which the numerical variation direction is changed, and generating curve inflection point characteristics; Comparing the current detection value with the last historical detection value in the time sequence, analyzing the position and the change trend of the current value in the evolution curve, and judging whether the current value change is influenced by the current detection flow or not by combining the flow influence characteristics to generate current change characteristics; And integrating the intra-segment change feature, the flow influence feature, the curve inflection point feature and the current change feature, determining the association logic among the features, and generating the time sequence evolution feature comprising the numerical value long-term change trend, the short-term fluctuation mode and the change attribute caused by the flow difference.
9. 9. The method for identifying abnormal data of physical examination of an artificial intelligent main specimen according to claim 1, wherein the step of matching the aggregate set of abnormal characteristics of physical examination with the corresponding clinical diagnosis and treatment path template to generate an abnormal data identification report containing abnormal association logic, clinical reference basis and follow-up examination suggestion comprises the steps of: Extracting system attribution grouping information and core abnormal feature nodes in the physical examination abnormal feature aggregation set, taking the system attribution grouping as an index, and matching corresponding diagnosis and treatment path templates from a clinical diagnosis and treatment path template library, wherein each diagnosis and treatment path template comprises a processing flow and an inspection item of common abnormality of the system; analyzing the matched clinical diagnosis and treatment path template, extracting diagnosis and treatment links corresponding to core abnormal feature nodes in the physical examination abnormal feature aggregation set in the template, extracting clinical judgment standards and recommended examination items of the diagnosis and treatment links, and generating a template matching result; comparing the abnormal feature association logic in the physical examination abnormal feature aggregation set with clinical judgment standards in the template matching result, verifying whether the abnormal feature combination meets the abnormal judgment requirement of the diagnosis and treatment link, and generating logic verification description; extracting clinical research literature, diagnosis and treatment guide items and typical case abstracts corresponding to the core abnormal characteristic nodes and the abnormal association logic from a clinical reference basis database, and taking the clinical research literature, diagnosis and treatment guide items and typical case abstracts as clinical reference basis for abnormal judgment, so as to ensure the authority and the relativity of the reference basis; Based on recommended inspection items in the template matching result, combining scene influence description and cross-item interaction characteristics in the physical examination abnormal characteristic aggregation set, and pertinently adjusting the priority of the inspection items; generating a follow-up inspection proposal according to the adjusted inspection item priority, wherein the follow-up inspection proposal comprises an inspection item name, an inspection sequence, notes during inspection and an expected inspection purpose; Sorting feature nodes in the physical examination abnormal feature aggregation set according to the system attribution grouping, firstly listing core abnormal features under each grouping, then listing associated abnormal features, and attaching abnormal expression description and scene influence analysis; integrating system attribution grouping abnormal characteristics, abnormal association logic, logic verification description, clinical reference basis and follow-up examination suggestions, and generating a physical examination abnormal data identification report according to the order of clinical importance from high to low.
10. 10. A physical examination abnormal data identification system of an artificial intelligent main examination object, comprising a processor and a readable storage medium, wherein the readable storage medium stores a program which, when executed by the processor, implements the physical examination abnormal data identification method of the artificial intelligent main examination object according to any one of claims 1 to 9.

Description

Physical examination abnormal data identification method and system for artificial intelligent main examination Technical Field The application relates to the technical field of intelligent medical services, in particular to a physical examination abnormal data identification method and system of an artificial intelligent main sample. Background In the current field of physical examination abnormal data identification, the traditional identification method mainly focuses on simple analysis of the physical examination object sub-item detection data. For example, whether abnormality exists is judged only according to whether the single blood detection index value exceeds the normal range or not, or simple comparison is performed in combination with part of indexes in the historical continuous detection data so as to observe the change trend of the indexes. However, the above conventional approaches have a number of limitations. On the one hand, the important influence of the basic health record data on the physical examination result is ignored. The basic health record data contains key information such as past disease history, allergy history, family genetic disease history and the like of the physical examination object, and the information is important to accurately judge whether the current physical examination data is abnormal or not and the nature and severity of the abnormality. On the other hand, the conventional method does not sufficiently consider physical examination scene information. The running state of the detection equipment can cause deviation of detection data due to factors such as equipment aging, inaccurate calibration and the like, the change of detection environment parameters such as temperature, humidity and the like can also influence the results of certain detection items, the detection operation specification record reflects whether the detection process is strictly carried out according to a standard flow, and the detection data can be distorted due to the fact that the operation is not specification. In addition, when analyzing physical examination data, the traditional method lacks deep mining of coupling characteristics, time sequence evolution characteristics of the same detection item and interaction characteristics of cross-item data between detection data and scene information, is difficult to comprehensively and accurately identify physical examination abnormal data, and is further incapable of generating an abnormal data identification report with clinical reference value. Disclosure of Invention In view of the above, the present application aims to provide a method and a system for identifying abnormal data of physical examination of an artificial intelligent main sample. According to a first aspect of the present application, there is provided a physical examination abnormal data identification method of an artificial intelligence main specimen, the method comprising: Associating a full-dimensional physical examination data set of physical examination objects with a physical examination scene information set to generate a physical examination data scene association set, wherein the full-dimensional physical examination data set comprises current sub-item detection data, historical continuous detection data and basic health record data, and the physical examination scene information set comprises detection equipment running states, detection environment parameters and detection operation specification records; Performing feature scene association processing on the physical examination data scene association set to obtain a scene physical examination feature association group, wherein the scene physical examination feature association group comprises coupling features of detection data and scene information, time sequence evolution features of the same detection item and interaction features of cross-item data; invoking an artificial intelligent main inspection model with a dynamic interaction mechanism, performing multi-round feature interaction analysis on the scene physical examination feature association group, and generating a feature dynamic interaction result, wherein the artificial intelligent main inspection model comprises a scene feature analysis module, a feature interaction engine and a decision feedback unit; Constructing an abnormal feature propagation chain based on the feature dynamic interaction result, screening feature nodes meeting clinical abnormal association rules in the abnormal feature propagation chain, and forming a physical examination abnormal feature aggregation set; Matching the physical examination abnormal characteristic aggregation set with a corresponding clinical diagnosis and treatment path template to generate a physical examination abnormal data identification report containing abnormal association logic, clinical reference basis and follow-up examination suggestion. According to a second aspect of the present application, there is