Large model-based medical data compliance inspection method and system

Abstract

The invention relates to the technical field of data processing, in particular to a medical data compliance check method and system based on a large model, and the method comprises the steps: extracting a core target medical index from a medical compliance check statement through a preset large language model; the multi-source medical basic information, the basic examination rule and the target medical index are integrated through the preset large language model, the target medical index is automatically associated with the field name of the initial medical data, the targeted candidate examination rule is generated, the accuracy and applicability of the rule are improved, the target medical data are extracted from the initial medical data, and the accuracy and applicability of the candidate examination rule are improved. According to the method, the target examination rule is screened from the candidate examination rules, the target medical data and the target examination rule are input into the preset large language model, the compliance examination report is obtained, comprehensive consideration of resource consumption and the detection effect is achieved, and the resource consumption cost is reduced while the data anomaly detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a compliance inspection method and system for medical data based on large models. Background Art

[0002] In today's digital medical era, medical data contains a large amount of sensitive information of patients, such as personal identity, health status, treatment records, etc. The compliant use and storage of medical data not only relate to the privacy protection of patients, but also affect the standardized development of the medical industry. Therefore, it is of crucial significance to conduct compliance inspections on medical data.

[0003] In the prior art, the method for conducting compliance inspections on medical data is mainly to preset a series of compliance inspection rules based on relevant medical regulations and industry standards. The rule engine checks medical data one by one according to the set compliance inspection rules to determine whether the medical data meets the rule requirements. The above method can quickly and accurately conduct inspections for simple and clear compliance requirements, and the inspection results have high interpretability. However, it has poor flexibility. When encountering new regulatory requirements or complex business scenarios, it is necessary to manually add or modify rules, which is time-consuming and laborious. Moreover, for unstructured or semi-structured data, it is difficult to formulate comprehensive and accurate rules for inspection, resulting in poor accuracy of compliance inspections and difficulty in meeting the data compliance inspection requirements in medical scenarios.

[0004] Therefore, how to improve the accuracy of data anomaly detection has become an urgent problem to be solved. Summary of the Invention

[0005] In view of the above technical problems, on the one hand, the present invention provides a compliance inspection method for medical data based on large models. The compliance inspection method for medical data based on large models includes the following steps: S1, input a preset medical compliance inspection statement into a preset large language model to obtain target medical indicators for compliance inspection.

[0006] S2, input the medical basic information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library, and the target medical indicators into the preset large language model to obtain the target basic information corresponding to the target medical indicators and candidate inspection rules. Among them, the medical basic information includes all field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be inspected, and each candidate inspection rule is used to inspect the data corresponding to the corresponding several target field names in the initial medical data.

[0007] S3, extract the target medical data from the initial medical data according to the target basic information.

[0008] S4. Screen the target inspection rule from all candidate inspection rules according to the target medical data.

[0009] S5. Input the target medical data and the target inspection rule into a preset large language model to obtain a compliance inspection report corresponding to the initial medical data.

[0010] In a second aspect, the present invention provides a compliance inspection system for medical data based on a large model. The compliance inspection system for medical data based on a large model includes: An index acquisition module, configured to input a preset medical compliance inspection statement into a preset large language model to obtain target medical indexes for compliance inspection.

[0011] A data screening module, configured to input the medical basic information corresponding to the initial medical data, the basic inspection rules in a preset inspection rule library, and the target medical indexes into a preset large language model to obtain target basic information corresponding to the target medical indexes and candidate inspection rules. Among them, the medical basic information includes all field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be inspected, and each candidate inspection rule is used to inspect the data corresponding to the corresponding several target field names in the initial medical data.

[0012] A data extraction module, configured to extract the target medical data from the initial medical data according to the target basic information.

[0013] A rule screening module, configured to screen the target inspection rule from all candidate inspection rules according to the target medical data.

[0014] A compliance inspection module, configured to input the target medical data and the target inspection rule into a preset large language model to obtain a compliance inspection report corresponding to the initial medical data.

[0015] The present invention has at least the following beneficial effects: By presetting the natural language processing ability of the large language model, the core target medical indicators can be quickly and accurately extracted from the preset medical compliance inspection statements, clarifying the direction and focus for subsequent inspection work. By presetting the large language model to integrate multi-source medical basic information, basic inspection rules, and target medical indicators, automatically associating the field names of the target medical indicators with the initial medical data, and generating targeted candidate inspection rules, the accuracy and applicability of the rules are improved. Extracting the target medical data from the initial medical data and avoiding the interference of irrelevant data can reduce the amount of data for subsequent inspections, improve the inspection efficiency and the accuracy of inspection results. Screening the target inspection rules from all the candidate inspection rules, inputting the target medical data and the target inspection rules into the preset large language model, obtaining the compliance inspection report corresponding to the initial medical data, achieving a comprehensive consideration of resource consumption and detection effects, reducing the resource consumption cost while improving the accuracy of data anomaly detection, and facilitating medical data managers and decision-makers to quickly understand the compliance status of the data and make corresponding decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic diagram of a method for compliance inspection of medical data based on a large model provided in Embodiment 1 of the present invention; Figure 2 It is a module schematic diagram of a system for compliance inspection of medical data based on a large model provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0019] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It can be understood that, under appropriate circumstances, the above-mentioned terms used to distinguish similar objects can be interchanged, so that the present invention can also implement other embodiments other than the above-mentioned illustrated embodiments or described embodiments. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] Embodiment 1 Embodiment 1 provides a compliance inspection method for medical data based on a large model, as Figure 1 shown. The compliance inspection method for medical data based on a large model includes the following steps: S1. Input a preset medical compliance inspection statement into a preset large language model to obtain target medical indicators for compliance inspection.

[0021] Among them, the preset medical compliance inspection statement is a statement written in advance based on medical regulations, industry standards and business requirements, describing the requirements for medical data compliance inspection, and is used to generally indicate the inspection objects and inspection contents in this compliance inspection, and may include several statements such as "check whether the surgical process meets the specified requirements" and "confirm whether the use of drugs is within the specified range".

[0022] The preset large language model refers to an artificial intelligence model trained with a large amount of text data, such as models like GPT, which has powerful language understanding and generation capabilities, can perform semantic understanding and analysis on the input medical compliance inspection statement, identify the key medical information involved therein, and thus extract the target medical indicators.

[0023] The target medical indicator is a specific indicator extracted from the preset medical compliance inspection statement and used to measure whether the medical data is compliant. For example, in the above-mentioned medical compliance inspection statement "confirm whether the use of drugs is within the specified range", "the use of drugs" is a target medical indicator.

[0024] In a specific embodiment, a preset medical compliance inspection statement and a first preset prompt word are input into the preset large language model to obtain target medical indicators for compliance inspection.

[0025] Among them, the first preset prompt word is used to guide the preset large language model to generate specific results, that is, the target medical indicators for compliance inspection. The first preset prompt word can be set in advance according to the actual needs of the implementer. For example, the first preset prompt word can be "Give the key compliance indicators according to the input content" or "Give the risk assessment indicators according to the input content", etc., which are prompt words for obtaining the target medical indicators.

[0026] As described above, by using the intelligent analysis ability of the large language model, the target medical indicators can be quickly and accurately extracted from complex medical compliance inspection statements, avoiding the cumbersome and accuracy problems of manual extraction of indicators.

[0027] S2. Input the medical basic information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library, and the target medical indicators into the preset large language model to obtain the target basic information and candidate inspection rules corresponding to the target medical indicators. Among them, the medical basic information includes all the field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be inspected, and each candidate inspection rule is used to inspect the data corresponding to the corresponding several target field names in the initial medical data.

[0028] Among them, the initial medical data refers to the medical data that needs to be subject to compliance inspection, which can come from the hospital's information system, medical equipment, etc., and contains various medical information such as the patient's basic information, diagnosis results, treatment records, etc.

[0029] The medical basic information contains all the field names corresponding to the initial medical data. For example, the medical basic information can include field names such as "patient name", "patient age", "medical record number", "diagnosis result", "symptom manifestation", "disease code", treatment plan, "surgery name", "surgery record", "medication record", "attending doctor", etc.

[0030] The preset inspection rule library is a rule set established in advance according to medical regulations, industry standards and business requirements for conducting compliance inspections on medical data. The preset inspection rule library includes several basic inspection rules, and the basic inspection rules use SQL statements to conduct compliance inspections on medical data.

[0031] The preset large language model conducts comprehensive analysis and reasoning on the medical basic information, basic inspection rules and target medical indicators to find the target basic information and candidate inspection rules related to the target medical indicators. Among them, the target basic information refers to several target field names to be inspected that are related to the target medical indicators. For example, if the target medical indicator is "the usage situation of drugs", the target basic information can be the field name of "medication record".

[0032] The candidate inspection rules are a series of inspection rules generated based on the target medical indicators, medical basic information, and basic inspection rules. Each candidate inspection rule is used to inspect the corresponding data of several target field names in the initial medical data. For example, for the target medical indicator of "drug usage", the candidate inspection rules may include rules such as whether the drug type is correct, whether the drug usage time is correct, and whether the drug dosage is correct.

[0033] In a specific embodiment, the medical basic information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library, the target medical indicator, and the second preset prompt word are input into the preset large language model to obtain the target basic information and candidate inspection rules corresponding to the target medical indicator.

[0034] Among them, the second preset prompt word is used to guide the preset large language model to screen and generate the target basic information and candidate inspection rules corresponding to the target medical indicator from the medical basic information corresponding to the input initial medical data, the basic inspection rules in the preset inspection rule library, and the target medical indicator.

[0035] The second preset prompt word can be preset according to the actual needs of the implementer. For example, the second preset prompt word can be "According to the input content, screen and output the target basic information and candidate inspection rules corresponding to the target medical indicator".

[0036] As described above, by automatically associating the target medical indicator with the field names of the initial medical data through the large language model, targeted candidate inspection rules are generated, improving the accuracy and applicability of the rules, and enhancing the applicability to different medical data structures and compliance inspection requirements.

[0037] S3. Extract the target medical data from the initial medical data according to the target basic information.

[0038] Among them, the target medical data refers to the specific data extracted from the initial medical data and corresponding to the target basic information. For example, if the target basic information is the field name of "drug usage", the target medical data is all the data related to the drugs used in the initial medical data.

[0039] As described above, extracting the target medical data from the initial medical data can avoid the interference of irrelevant data, reduce the amount of data for subsequent inspections, and improve the inspection efficiency and the accuracy of inspection results.

[0040] S4. Screen the target inspection rules from all the candidate inspection rules according to the target medical data.

[0041] Among them, each candidate check rule corresponds to one, two or more target field names, which are used to comprehensively check the data corresponding to the corresponding target field names. Each target field name can be subject to compliance checks through one, two or more different candidate check rules. Therefore, after screening a number of candidate check rules from the preset check rule library in combination with the medical basic information, different candidate check rules can perform compliance checks on the data corresponding to the same target field name. When performing compliance checks according to all candidate check rules, there are redundant checks, resulting in consumption of computing resources, and further increasing the inspection cost and time.

[0042] Therefore, in this embodiment, the candidate check rules are further screened to obtain target check rules for performing compliance checks on the target medical data, so as to reduce the inspection cost and improve the inspection efficiency while ensuring the accuracy of the compliance checks.

[0043] In a specific implementation manner, S4 includes the following steps: S41, according to the several target field names corresponding to each candidate check rule, screen out several candidate check rule combinations from all candidate check rules, where each candidate check rule combination includes several candidate check rules, and each candidate check rule combination is used to check the target medical data.

[0044] S42, for any candidate check rule combination, according to the data volume corresponding to the several target field names corresponding to each candidate check rule in the current candidate check rule combination and the space-time processing parameters corresponding to each candidate check rule, obtain the space-time processing cost corresponding to the current candidate check rule combination.

[0045] S43, according to the data volume corresponding to the several target field names corresponding to each candidate check rule in the current candidate check rule combination and the execution efficiency parameters, obtain the execution efficiency score corresponding to the current candidate check rule combination.

[0046] S44, according to the space-time processing cost and the execution efficiency score corresponding to each candidate check rule combination, obtain the priority corresponding to each candidate check rule combination.

[0047] S45, determine the several candidate check rules corresponding to the candidate check rule combination with the highest priority as the target check rules.

[0048] Among them, each candidate check rule corresponds to several target field names. By analyzing the association and combination of the target field names, different rule combinations are screened out from all candidate check rules as the basis for screening out the target check rules.

[0049] The amount of data corresponding to each target field name, that is, the amount of data corresponding to each target field name in the target medical data.

[0050] The spatio-temporal processing parameters are parameters related to the time cost and space cost involved in the execution process of the candidate inspection rules. For example, the detection speed, occupied storage space, etc. when performing data compliance inspection using the candidate inspection rules, which are used to evaluate the spatio-temporal processing cost of the candidate inspection rule combination. The spatio-temporal processing cost represents the overall time and space resource cost required for the current candidate inspection rule combination to process the target medical data.

[0051] The execution efficiency parameters are parameters related to the execution effect of the candidate inspection rules. For example, indicators such as query accuracy, recall rate, and stability. The execution efficiency score of the candidate inspection rule combination is calculated through the execution efficiency parameters. The execution efficiency score represents the comprehensive execution efficiency and accuracy score of the current candidate inspection rule combination when processing the target medical data.

[0052] The priority of each candidate inspection rule combination can be determined by means of weighted calculation, combining the spatio-temporal processing cost and the execution efficiency score. The specific weighting method can be adjusted according to actual requirements and application scenarios.

[0053] The priority is used to represent the relative importance and applicability of each candidate inspection rule combination among all combinations. The higher the priority, the better the comprehensive performance of the candidate inspection rule combination in terms of spatio-temporal processing cost and execution efficiency.

[0054] The candidate inspection rule combination with the highest priority has the best comprehensive performance in terms of spatio-temporal processing cost and execution efficiency. Selecting the candidate inspection rule corresponding to the candidate inspection rule combination with the highest priority as the target inspection rule can maximize the inspection efficiency and reduce resource consumption while ensuring the inspection accuracy.

[0055] As described above, by calculating the spatio-temporal processing cost, the efficiency of each candidate inspection rule combination in resource utilization can be evaluated. By the execution efficiency score, the efficiency and accuracy of each candidate inspection rule combination in processing data can be reflected. Selecting the combination with a lower spatio-temporal processing cost and a higher execution efficiency score can reduce the burden on the system, avoid resource waste, and improve the inspection efficiency and the accuracy of inspection results.

[0056] In a specific implementation manner, S41 includes the following steps: S411, perform arbitrary combinations on all candidate inspection rules to obtain a number of initial inspection rule combinations.

[0057] S412. For any initial inspection rule combination, obtain the set of target field names corresponding to the current initial inspection rule combination according to the several target field names corresponding to each candidate inspection rule in the current initial inspection rule combination.

[0058] S413. If the set of target field names corresponding to the current initial inspection rule combination includes all the target field names to be inspected and there are no duplicate target field names, then determine the current initial inspection rule combination as a candidate inspection rule combination.

[0059] S414. Traverse all the initial inspection rule combinations and repeatedly execute steps S12 and S13 to obtain all the candidate inspection rule combinations.

[0060] Among them, the initial inspection rule combination refers to the rule set obtained by arbitrarily combining all the candidate inspection rules, providing a comprehensive selection range for subsequently screening out suitable candidate inspection rule combinations. Each initial inspection rule combination includes several candidate inspection rules.

[0061] The set of target field names is a set composed of the target field names corresponding to all the candidate inspection rules in the current initial inspection rule combination, which can clearly understand all the target fields involved in the current initial inspection rule combination and is used to subsequently determine whether the current initial inspection rule combination meets the requirements.

[0062] The set of target field names corresponding to the candidate inspection rule combination includes all the target field names to be inspected and there are no duplicate target field names, that is, all the candidate inspection rules in the candidate inspection rule combination can comprehensively detect the data corresponding to all the target field names to be inspected and will not perform duplicate detection, ensuring the integrity of the abnormal data detection and reducing the detection cost.

[0063] As described above, by screening out the initial inspection rule combinations that meet the conditions as candidate inspection rule combinations, it can be ensured that the final candidate inspection rule combination can comprehensively and efficiently inspect the target medical data, avoiding the problems of missing inspection fields or duplicate inspections, and improving the accuracy and efficiency of the inspection.

[0064] In a specific implementation manner, the spatio-temporal processing parameters include time processing parameters and space processing parameters, and S42 includes the following steps: S421. Obtain the first total data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the data volume corresponding to the several target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0065] S422. Obtain the time processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the total amount of the first data and the time processing parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0066] S423. Obtain the space-time processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the time processing cost and the space processing parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0067] The total amount of the first data refers to the sum of the data amounts included in all target field names corresponding to all candidate inspection rules in the current candidate inspection rule combination, which is used to clarify the data scale that each candidate inspection rule combination needs to process and provides basic data for subsequent calculation of the time processing cost and the space processing cost.

[0068] The time processing parameter can be the speed when performing data anomaly detection for the corresponding candidate inspection rule. According to the total amount of the first data to be detected and the time processing parameter, the time for the corresponding candidate inspection rule to process the data corresponding to the corresponding target field name can be calculated as the corresponding time processing cost.

[0069] The space processing parameter can be the resource occupation size of the corresponding candidate inspection rule for processing a unit data amount for resources such as CPU, memory, and disk, such as the disk space size occupied for storing a certain amount of data. According to the total amount of the first data to be detected and the space processing parameter, the resource size occupied by the data corresponding to the corresponding target field name by the corresponding candidate inspection rule can be calculated as the corresponding space processing cost.

[0070] Comprehensively calculate the time processing cost and the space processing cost of each candidate inspection rule. For example, the space-time processing cost corresponding to each candidate inspection rule is obtained by using the method of weighted summation, which reflects the processing cost of each candidate inspection rule during the execution process and serves as the basis for screening the candidate inspection rule combination.

[0071] As described above, by calculating the time processing cost and the space processing cost, the time required for each candidate inspection rule to process data and the resource size occupied can be quantified, which helps to evaluate the processing costs of different rule combinations in terms of time and space, and further provides a basis for selecting a better inspection rule combination.

[0072] In a specific implementation manner, the execution efficiency parameters include an accuracy parameter, a recall parameter, and a stability parameter. S43 includes the following steps: S431. Obtain the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the accuracy parameter, recall parameter, stability parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination, the first preset weight corresponding to the accuracy parameter, the second preset weight corresponding to the recall parameter, and the third preset weight corresponding to the stability parameter.

[0073] S432. Obtain the execution efficiency score corresponding to the current candidate inspection rule combination according to the first data volume and the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0074] Among them, the accuracy parameter represents the proportion of abnormal data correctly detected by the candidate inspection rule when inspecting data, reflecting the correctness of the inspection result of the candidate inspection rule. For example, among 100 actual abnormal data, the candidate inspection rule correctly detects 80, and its accuracy is 80%.

[0075] The recall parameter represents the proportion of abnormal data detected by the candidate inspection rule among all actual abnormal data, reflecting the ability of the candidate inspection rule to find all abnormal data. For example, if there are 100 actual abnormal data and the candidate inspection rule detects 90, the recall is 90%.

[0076] The stability parameter measures the consistency and reliability of the candidate inspection rule during multiple executions, and can be quantified by observing the fluctuation of the corresponding inspection results during multiple executions of the candidate inspection rule. For example, if the deviation of the inspection results is within a small range during multiple executions of the candidate inspection rule, the stability is high.

[0077] The first preset weight, the second preset weight, and the third preset weight are used to reflect the relative importance of the accuracy parameter, the recall parameter, and the stability parameter during comprehensive evaluation, and the specific values can be set by the implementer according to the actual situation.

[0078] As described above, comprehensively considering the accuracy parameter, recall parameter, and stability parameter of the candidate inspection rule can comprehensively reflect the efficiency of each candidate inspection rule during execution, providing a data basis for screening candidate inspection rule combinations with higher efficiency.

[0079] In a specific implementation manner, S432 includes the following steps: S4321. Obtain the second data volume corresponding to the current candidate inspection rule combination according to the first data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0080] S4322. Determine the ratio of the total amount of the first data corresponding to each candidate inspection rule in the current candidate inspection rule combination to the total amount of the second data corresponding to the current candidate inspection rule combination as the reference weight corresponding to the current candidate inspection rule combination.

[0081] S4323. Obtain the execution efficiency score corresponding to the current candidate inspection rule combination according to the reference weight and the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0082] Among them, the reference weight reflects the proportion of the data volume processed by each candidate inspection rule in the entire candidate inspection rule combination. The candidate inspection rule with a larger proportion of the data volume has a higher relative importance in the entire candidate inspection rule combination. Correspondingly, the weight occupied in the subsequent calculation of the execution efficiency score is larger, so that the execution efficiency score can more accurately reflect the actual efficiency of the candidate inspection rule combination in processing large-scale data, avoiding ignoring the impact of data volume differences on the overall efficiency during evaluation, and thus providing a more reliable basis for selecting the optimal candidate inspection rule combination.

[0083] In a specific implementation manner, S44 includes the following steps: S441. Obtain the cost threshold according to the spatio-temporal processing costs corresponding to all candidate inspection rule combinations.

[0084] S442. Obtain the first score threshold and the second score threshold according to the execution efficiency scores corresponding to all candidate inspection rule combinations.

[0085] S443. For any candidate inspection rule combination, if the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the second score threshold, then determine the current candidate inspection rule combination as the intermediate inspection rule combination.

[0086] S444. If the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the first score threshold and less than the second score threshold, and the corresponding spatio-temporal processing cost is less than or equal to the cost threshold, then determine the current candidate inspection rule combination as the intermediate inspection rule combination.

[0087] S445. Obtain the priority corresponding to each intermediate inspection rule combination according to the spatio-temporal processing cost, the execution efficiency score, the fourth preset weight corresponding to the spatio-temporal processing cost, and the fifth preset weight corresponding to the execution efficiency score of each intermediate inspection rule combination.

[0088] S446. If the execution efficiency score corresponding to the current candidate inspection rule combination is less than the first score threshold, or the corresponding spatio-temporal processing cost is greater than the cost threshold, then determine the priority corresponding to the current candidate inspection rule combination as the lowest priority.

[0089] Among them, the cost threshold refers to the boundary value of the spatio-temporal processing cost for screening candidate inspection rule combinations, which is obtained by analyzing the spatio-temporal processing costs of all candidate inspection rule combinations and is used to determine whether the spatio-temporal processing cost of a certain combination is within an acceptable range. The first scoring threshold and the second scoring threshold refer to the boundary values for dividing the execution efficiency levels of candidate inspection rule combinations, which are obtained by analyzing the execution efficiency scores of all candidate inspection rule combinations and are used to determine the level of execution efficiency of a certain combination. The first scoring threshold is less than the second scoring threshold.

[0090] Candidate inspection rule combinations with an execution efficiency score greater than or equal to the second scoring threshold perform excellently in terms of execution effect. Correspondingly, part of the requirement for low cost can be sacrificed, and the corresponding candidate inspection rule combinations can be screened as intermediate inspection rule combinations, which helps to prioritize combinations with better performance for subsequent priority calculation, thereby improving the accuracy of the detection results of the data to be detected.

[0091] Candidate inspection rule combinations with an execution efficiency score between the first scoring threshold and the second scoring threshold and a spatio-temporal processing cost less than or equal to the cost threshold perform well both in terms of execution effect and low cost. Screening the corresponding candidate inspection rule combinations as intermediate inspection rule combinations helps to comprehensively consider combinations with high accuracy and low cost for data anomaly detection and achieve a balance between accuracy and cost.

[0092] By comprehensively considering the spatio-temporal processing cost and the execution efficiency score in a weighted summation manner, a reasonable priority is quantified for each intermediate inspection rule combination, which is convenient for subsequently selecting the optimal combination for data anomaly detection.

[0093] For combinations with poor execution efficiency or too high spatio-temporal processing cost, their priorities are set to the lowest, reducing the possibility of being selected for data anomaly detection and avoiding negative impacts on system performance and detection effect.

[0094] As described above, by setting the cost threshold and two scoring thresholds, multi-dimensional screening and evaluation are performed on candidate inspection rule combinations, comprehensively considering two key factors of spatio-temporal processing cost and execution efficiency, and ensuring that the finally selected combination achieves a good balance between resource consumption and detection effect.

[0095] In a specific embodiment, S441 includes the following steps: S4411, according to the spatio-temporal processing costs corresponding to all candidate inspection rule combinations, fit to obtain a cost distribution function.

[0096] S4412, determine the spatio-temporal processing cost corresponding to the second quartile of the cost distribution function as the cost threshold.

[0097] Among them, the cost distribution function describes the distribution law of the spatio-temporal processing costs corresponding to all candidate inspection rule combinations. The second quartile is the median. Find the point where the cumulative distribution function value of the cost distribution function is 0.5, and the spatio-temporal processing cost corresponding to this point is the cost threshold.

[0098] Based on the distribution law of the spatio-temporal processing costs corresponding to all candidate inspection rule combinations, the appropriate cost threshold is selected, so that the selected combinations are representative and reasonable in terms of spatio-temporal resource consumption, and the screening accuracy of the subsequent optimal candidate inspection rule combinations is improved.

[0099] In a specific embodiment, S442 includes the following steps: S4421, according to the execution efficiency scores corresponding to all candidate inspection rule combinations, fit to obtain a score distribution function.

[0100] S4422, determine the execution efficiency score corresponding to the second quartile of the score distribution function as the first score threshold.

[0101] S4423, determine the execution efficiency score corresponding to the third quartile of the score distribution function as the second score threshold Among them, the score distribution function reflects the distribution of the execution efficiency scores corresponding to all candidate inspection rule combinations, which helps to understand the distribution characteristics of the execution efficiency of different combinations in the whole.

[0102] The second quartile of the score distribution function is the point where the cumulative distribution function value of the score distribution function is 0.5, and the third quartile of the score distribution function is the point where the cumulative distribution function value of the score distribution function is 0.75. Find the corresponding execution efficiency score of this point as the second score threshold.

[0103] Based on the distribution law of the execution efficiency scores corresponding to all candidate inspection rule combinations, setting appropriate first and second score thresholds can divide all candidate inspection rule combinations into different intervals according to the execution efficiency, which is convenient for subsequent taking different processing strategies according to different intervals and more reasonably screening out the optimal candidate inspection rule combinations.

[0104] As described above, by screening out the candidate inspection rule combinations that can detect the data corresponding to all target field names, the comprehensive inspection of each key part of the data to be detected is ensured. Based on the data volume of the target field names corresponding to each candidate inspection rule and the spatio-temporal processing parameters, the consumption of time and space resources by each candidate inspection rule combination during execution is accurately evaluated. And based on the data volume of the target field names corresponding to the candidate inspection rules and the execution efficiency parameters, the execution efficiency score of each candidate inspection rule combination is evaluated. Finally, by combining the spatio-temporal processing cost and the execution efficiency score, the priority of each candidate inspection rule combination is determined, which helps to intuitively compare the performance of different combinations in terms of detection accuracy, recall rate, stability, etc. Thus, in practical applications, according to the resource status and performance requirements, a combination with lower execution cost and excellent performance can be reasonably selected to achieve a comprehensive consideration of resource consumption and detection effect, while improving the accuracy of data anomaly detection and reducing the resource consumption cost.

[0105] S5. Input the target medical data and the target inspection rules into a preset large language model to obtain a compliance inspection report corresponding to the initial medical data.

[0106] Among them, when the target medical data and the target inspection rules are input into the preset large language model, the large language model will understand the target inspection rules, clarify the specific inspection requirements and logic, and then conduct a one-by-one inspection of the target medical data according to the target inspection rules. Specifically, during the inspection process, it will judge whether the data meets the rule requirements, and once data that does not meet the rules is found, it will be recorded. Finally, the large language model will sort out and analyze the inspection results, and predict potential risks and trends, and generate a compliance inspection report in the form of natural language, greatly reducing the workload of manual review and report writing, improving work efficiency, facilitating medical data managers and decision-makers to quickly understand the compliance status of the data, and making corresponding decisions.

[0107] In a specific embodiment, the target medical data, the target inspection rules, and a third preset prompt word are input into a preset large language model to obtain a compliance inspection report corresponding to the initial medical data.

[0108] Among them, the third preset prompt word is used to guide the preset large language model to output a compliance inspection report corresponding to the initial medical data.

[0109] The third preset prompt word can be set in advance according to the actual needs of the implementer. For example, the third preset prompt word can be "Conduct a compliance inspection of the target medical data according to the input target inspection rules, generate and output a compliance inspection report corresponding to the initial medical data, where the data that does not meet the rules is recorded in the compliance inspection report, and potential risks and trends are predicted".

[0110] As described above, by presetting the natural language processing ability of the large language model, the core target medical indicators are quickly and accurately extracted from the preset medical compliance inspection statements, clarifying the direction and focus for subsequent inspection work. By presetting the large language model to integrate multi-source medical basic information, basic inspection rules, and target medical indicators, automatically associating the field names of the target medical indicators with the initial medical data, generating targeted candidate inspection rules, improving the accuracy and applicability of the rules. Extracting the target medical data from the initial medical data and avoiding the interference of irrelevant data can reduce the amount of data for subsequent inspections, improve the inspection efficiency and the accuracy of inspection results. Screening the target inspection rules from all the candidate inspection rules, inputting the target medical data and the target inspection rules into the preset large language model, obtaining the compliance inspection report corresponding to the initial medical data, achieving a comprehensive consideration of resource consumption and detection effects, reducing the resource consumption cost while improving the accuracy of data anomaly detection, facilitating medical data managers and decision-makers to quickly understand the compliance status of the data and make corresponding decisions.

[0111] Embodiment 2 Embodiment 2 of the present invention provides a compliance inspection system for medical data based on a large model, as Figure 2 shown. The compliance inspection system for medical data based on a large model includes: An index acquisition module 21, configured to input the preset medical compliance inspection statements into the preset large language model to obtain the target medical indicators for compliance inspection.

[0112] A data screening module 22, configured to input the medical basic information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library, and the target medical indicators into the preset large language model to obtain the target basic information corresponding to the target medical indicators and the candidate inspection rules. Among them, the medical basic information includes all the field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be inspected, and each candidate inspection rule is used to inspect the corresponding data of the several target field names in the initial medical data.

[0113] A data extraction module 23, configured to extract the target medical data from the initial medical data according to the target basic information.

[0114] A rule screening module 24, configured to screen the target inspection rules from all the candidate inspection rules according to the target medical data.

[0115] A compliance inspection module 25, configured to input the target medical data and the target inspection rules into the preset large language model to obtain the compliance inspection report corresponding to the initial medical data.

[0116] In a specific embodiment, the rule screening module 24 includes: A rule screening sub-module, configured to screen out a number of candidate inspection rule combinations from all candidate inspection rules according to a number of target field names corresponding to each candidate inspection rule, where each candidate inspection rule combination includes a number of candidate inspection rules, and each candidate inspection rule combination is used to inspect target medical data.

[0117] A spatio-temporal processing cost acquisition sub-module, configured to, for any candidate inspection rule combination, obtain the spatio-temporal processing cost corresponding to the current candidate inspection rule combination according to the data volume corresponding to the number of target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination and the spatio-temporal processing parameters corresponding to each candidate inspection rule.

[0118] An execution efficiency score acquisition sub-module, configured to obtain the execution efficiency score corresponding to the current candidate inspection rule combination according to the data volume corresponding to the number of target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination and the execution efficiency parameters.

[0119] A priority acquisition sub-module, configured to obtain the priority corresponding to each candidate inspection rule combination according to the spatio-temporal processing cost and the execution efficiency score corresponding to each candidate inspection rule combination.

[0120] A target inspection rule acquisition sub-module, configured to determine the number of candidate inspection rules corresponding to the candidate inspection rule combination with the highest priority as the target inspection rules.

[0121] In a specific embodiment, the rule screening sub-module includes: A rule combination unit, configured to arbitrarily combine all candidate inspection rules to obtain a number of initial inspection rule combinations.

[0122] A target field name set acquisition unit, configured to, for any initial inspection rule combination, obtain the target field name set corresponding to the current initial inspection rule combination according to the number of target field names corresponding to each candidate inspection rule in the current initial inspection rule combination.

[0123] A first candidate inspection rule acquisition unit, configured to, if the target field name set corresponding to the current initial inspection rule combination includes all target field names to be inspected and there are no duplicate target field names, determine the current initial inspection rule combination as a candidate inspection rule combination.

[0124] A second candidate inspection rule acquisition unit, configured to traverse all initial inspection rule combinations and repeatedly execute the steps of the target field name set acquisition unit and the first candidate inspection rule acquisition unit to obtain all candidate inspection rule combinations.

[0125] In a specific embodiment, the spatio-temporal processing parameters include temporal processing parameters and spatial processing parameters. The spatio-temporal processing cost acquisition sub-module includes: The first data volume acquisition sub-unit is configured to obtain the first data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the data volumes corresponding to several target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0126] The temporal processing cost acquisition sub-unit is configured to obtain the temporal processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the first data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination and the temporal processing parameters.

[0127] The spatio-temporal processing cost acquisition sub-unit is configured to obtain the spatio-temporal processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the temporal processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination and the spatial processing parameters.

[0128] In a specific embodiment, the execution efficiency parameters include an accuracy parameter, a recall parameter, and a stability parameter. The execution efficiency score acquisition sub-module includes: The reference score acquisition unit is configured to obtain the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the accuracy parameter, recall parameter, stability parameter, first preset weight corresponding to the accuracy parameter, second preset weight corresponding to the recall parameter, and third preset weight corresponding to the stability parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0129] The execution efficiency score acquisition unit is configured to obtain the execution efficiency score corresponding to the current candidate inspection rule combination according to the first data volume and the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0130] In a specific embodiment, the execution efficiency score acquisition unit includes: The second data volume acquisition sub-unit is configured to obtain the second data volume corresponding to the current candidate inspection rule combination according to the first data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0131] The reference weight acquisition sub-unit is configured to determine the ratio of the first data volume corresponding to each candidate inspection rule in the current candidate inspection rule combination to the second data volume corresponding to the current candidate inspection rule combination as the reference weight corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0132] An execution efficiency score acquisition subunit, configured to obtain an execution efficiency score corresponding to the current candidate inspection rule combination according to the reference weight and reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

[0133] In a specific embodiment, the priority acquisition sub-module includes: A cost threshold acquisition unit, configured to obtain a cost threshold according to the spatio-temporal processing costs corresponding to all candidate inspection rule combinations.

[0134] A score threshold acquisition unit, configured to obtain a first score threshold and a second score threshold according to the execution efficiency scores corresponding to all candidate inspection rule combinations.

[0135] A first intermediate inspection rule combination acquisition unit, configured to, for any candidate inspection rule combination, if the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the second score threshold, determine the current candidate inspection rule combination as an intermediate inspection rule combination.

[0136] A second intermediate inspection rule combination acquisition unit, configured to, if the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the first score threshold and less than the second score threshold, and the corresponding spatio-temporal processing cost is less than or equal to the cost threshold, determine the current candidate inspection rule combination as an intermediate inspection rule combination.

[0137] A first priority acquisition unit, configured to obtain the priority corresponding to each intermediate inspection rule combination according to the spatio-temporal processing cost, execution efficiency score, fourth preset weight corresponding to the spatio-temporal processing cost, and fifth preset weight corresponding to the execution efficiency score of each intermediate inspection rule combination.

[0138] A second priority acquisition unit, configured to, if the execution efficiency score corresponding to the current candidate inspection rule combination is less than the first score threshold, or the corresponding spatio-temporal processing cost is greater than the cost threshold, determine that the priority corresponding to the current candidate inspection rule combination is the lowest priority.

[0139] In a specific embodiment, the cost threshold acquisition unit includes: A cost distribution function acquisition subunit, configured to fit a cost distribution function according to the spatio-temporal processing costs corresponding to all candidate inspection rule combinations.

[0140] A cost threshold acquisition subunit, configured to determine the spatio-temporal processing cost corresponding to the second quartile of the cost distribution function as the cost threshold.

[0141] In a specific embodiment, the score threshold acquisition unit includes: A scoring distribution function obtaining subunit, configured to fit a scoring distribution function according to the execution efficiency scores corresponding to all candidate inspection rule combinations.

[0142] A first scoring threshold obtaining subunit, configured to determine the execution efficiency score corresponding to the second quartile of the scoring distribution function as the first scoring threshold.

[0143] A second scoring threshold obtaining subunit, configured to determine the execution efficiency score corresponding to the third quartile of the scoring distribution function as the second scoring threshold.

[0144] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and not for limiting the scope of the present invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present invention. The scope disclosed by the present invention is defined by the appended claims.

Claims

1. A compliance inspection method for medical data based on a large model, characterized in that: The compliance checking method for medical data based on a big model includes the following steps: S1, inputting a preset medical compliance inspection sentence into a preset large language model to obtain target medical indicators for compliance inspection; S2, inputting the basic medical information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library and the target medical indicator into the preset large language model, and obtaining the target basic information and candidate inspection rules corresponding to the target medical indicator, wherein the basic medical information includes all field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be checked, and each candidate inspection rule is used to check the data corresponding to the corresponding several target field names in the initial medical data; S3, extracting target medical data from the initial medical data according to the target basic information; S4, screening a target inspection rule from all candidate inspection rules according to the target medical data; S5, inputting the target medical data and the target inspection rules into the preset large language model, and obtaining a compliance inspection report corresponding to the initial medical data.

2. The compliance checking method for medical data based on a large model according to claim 1, characterized in that: S4 includes the following steps: S41, according to the target field names corresponding to each candidate inspection rule, screening out a plurality of candidate inspection rule combinations from all candidate inspection rules, wherein each candidate inspection rule combination includes a plurality of candidate inspection rules, and each candidate inspection rule combination is used to inspect the target medical data; S42, for any candidate inspection rule combination, according to the data volume corresponding to several target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination and the spatiotemporal processing parameters corresponding to each candidate inspection rule, obtain the spatiotemporal processing cost corresponding to the current candidate inspection rule combination; S43, obtaining an execution efficiency score corresponding to the current candidate inspection rule combination according to the data volume and execution efficiency parameters corresponding to a number of target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination; S44, obtaining the priority corresponding to each candidate inspection rule combination according to the spatiotemporal processing cost and execution efficiency score corresponding to each candidate inspection rule combination; S45, determining several candidate inspection rules corresponding to the candidate inspection rule combination with the highest priority as target inspection rules.

3. The compliance checking method for medical data based on a large model according to claim 2, characterized in that: S41 includes the following steps: S411, arbitrarily combining all candidate inspection rules to obtain a number of initial inspection rule combinations; S412, for any initial inspection rule combination, according to a number of target field names corresponding to each candidate inspection rule in the current initial inspection rule combination, obtain a target field name set corresponding to the current initial inspection rule combination; S413, if the target field name set corresponding to the current initial check rule combination includes all the target field names to be checked and there are no repeated target field names, then the current initial check rule combination is determined as a candidate check rule combination; S414, traverse all initial inspection rule combinations, and repeat steps S412 and S413 to obtain all candidate inspection rule combinations.

4. The compliance checking method for medical data based on a large model according to claim 2, characterized in that: The spatiotemporal processing parameters include time processing parameters and space processing parameters, and S42 includes the following steps: S421, acquiring a first total amount of data corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the amount of data corresponding to a number of target field names corresponding to each candidate inspection rule in the current candidate inspection rule combination; S422, acquiring the time processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the first data volume and the time processing parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination; S423, according to the time processing cost and space processing parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination, the time and space processing cost corresponding to each candidate inspection rule in the current candidate inspection rule combination is obtained.

5. The compliance checking method for medical data based on a large model according to claim 4 is characterized in that: The execution performance parameters include accuracy parameters, recall parameters and stability parameters. S43 includes the following steps: S431, obtaining a reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination according to the accuracy parameter, recall parameter, stability parameter, first preset weight corresponding to the accuracy parameter, second preset weight corresponding to the recall parameter, and third preset weight corresponding to the stability parameter corresponding to each candidate inspection rule in the current candidate inspection rule combination; S432, obtaining an execution efficiency score corresponding to the current candidate inspection rule combination according to the total amount of first data and the reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

6. The compliance checking method for medical data based on a large model according to claim 5, characterized in that: S432 includes the following steps: S4321, acquiring a second total amount of data corresponding to the current candidate inspection rule combination according to the first total amount of data corresponding to each candidate inspection rule in the current candidate inspection rule combination; S4322, determining the ratio of the first total amount of data corresponding to each candidate inspection rule in the current candidate inspection rule combination to the second total amount of data corresponding to the current candidate inspection rule combination as the reference weight corresponding to each candidate inspection rule in the current candidate inspection rule combination; S4323, obtaining the execution efficiency score corresponding to the current candidate inspection rule combination according to the reference weight and reference score corresponding to each candidate inspection rule in the current candidate inspection rule combination.

7. The compliance checking method for medical data based on a large model according to claim 2, characterized in that: S44 includes the following steps: S441, obtaining a cost threshold according to the spatiotemporal processing costs corresponding to all candidate inspection rule combinations; S442, obtaining a first scoring threshold and a second scoring threshold according to the execution efficiency scores corresponding to all candidate inspection rule combinations; S443, for any candidate inspection rule combination, if the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the second score threshold, determine the current candidate inspection rule combination as an intermediate inspection rule combination; S444, if the execution efficiency score corresponding to the current candidate inspection rule combination is greater than or equal to the first score threshold and less than the second score threshold, and the corresponding spatiotemporal processing cost is less than or equal to the cost threshold, then the current candidate inspection rule combination is determined as an intermediate inspection rule combination; S445, obtaining the priority corresponding to each intermediate inspection rule combination according to the spatiotemporal processing cost corresponding to each intermediate inspection rule combination, the execution efficiency score, the fourth preset weight corresponding to the spatiotemporal processing cost, and the fifth preset weight corresponding to the execution efficiency score; S446: If the execution efficiency score corresponding to the current candidate inspection rule combination is less than the first score threshold, or the corresponding spatiotemporal processing cost is greater than the cost threshold, determine that the priority corresponding to the current candidate inspection rule combination is the lowest priority.

8. The compliance checking method for medical data based on a large model according to claim 7, characterized in that: S441 includes the following steps: S4411, fitting a cost distribution function according to the spatiotemporal processing costs corresponding to all candidate inspection rule combinations; S4412: Determine the spatiotemporal processing cost corresponding to the second quartile of the cost distribution function as a cost threshold.

9. The compliance checking method for medical data based on a large model according to claim 7, characterized in that: S442 includes the following steps: S4421, fitting a score distribution function according to the execution efficiency scores corresponding to all candidate inspection rule combinations; S4422, determining the execution efficiency score corresponding to the second quartile of the score distribution function as a first score threshold; S4423: Determine the execution performance score corresponding to the third quartile of the score distribution function as a second score threshold.

10. A compliance checking system for medical data based on a large model, characterized in that: The compliance inspection system for medical data based on a large model includes: An indicator acquisition module, used to input a preset medical compliance inspection sentence into a preset large language model to obtain a target medical indicator for compliance inspection; A data screening module, used for inputting the basic medical information corresponding to the initial medical data, the basic inspection rules in the preset inspection rule library and the target medical indicator into the preset large language model, and obtaining the target basic information and candidate inspection rules corresponding to the target medical indicator, wherein the basic medical information includes all the field names corresponding to the initial medical data, the basic inspection rules are in the form of SQL statements, the target basic information includes several target field names to be checked, and each candidate inspection rule is used for checking the data corresponding to the corresponding several target field names in the initial medical data; A data extraction module, used for extracting target medical data from the initial medical data according to the target basic information; A rule screening module, used for screening a target inspection rule from all candidate inspection rules according to the target medical data; A compliance checking module is used to input the target medical data and the target checking rules into the preset large language model to obtain a compliance checking report corresponding to the initial medical data.

Images