A medical data quality inspection method based on an internet of things

Abstract

The application discloses a medical data quality inspection method based on an Internet of Things, which comprises the following steps: based on the metadata uploaded by the inspection equipment of the Internet of Things, constructing an event graph, fusing equipment bias and network delay disturbance, and calculating a space-time consistency index; based on the inspection data identified through the consistency index and a physiological coupling rule library, calculating a physiological reasonableness score through strong coupling of physiological rules and explainability weight; based on multi-source state data, obtaining a man-machine stability factor by matching man-machine state snapshots and a scene perception rule library; based on the event graph and inspection metadata, obtaining a group response vector through simulation of an immune mechanism, and obtaining a relative credibility index according to an adjustment of an abnormal tolerance threshold value according to an epidemiological background; based on the above four items and multi-source feedback logs, obtaining a three-state quality arbitration result through multi-level arbitration grid decision and strategy optimization. The application improves the authenticity of medical data sources and results through data quality inspection.

Description

Technical Field

[0001] This invention relates to the field of the Internet of Things (IoT), and more particularly to a method for inspecting the quality of medical data based on the IoT. Background Technology

[0002] Current medical data quality inspection methods largely rely on static reference ranges and isolated numerical verification, which are insufficient to address the complexities of real-world clinical scenarios. On one hand, there is a lack of effective verification of the authenticity of test data sources; in the high-turnover environment of emergency rooms, mislabeling of samples or missynchronization of equipment are common occurrences. On the other hand, the reliability of the testing process is neglected, failing to incorporate operator condition, equipment quality control trends, and environmental factors into the assessment. In ICUs, operators under high-intensity work conditions may introduce sample addition errors, but these factors are often overlooked. Furthermore, existing methods lack intelligent arbitration mechanisms at the population level that combine similar patient distributions, epidemiological backgrounds, and medical knowledge rules, making it difficult to distinguish between genuine critical values ​​and technical artifacts, and they are insufficiently adapted to the individualized physiological baselines of patients with chronic diseases. Based on these shortcomings, this invention proposes a medical data quality inspection method based on the Internet of Things (IoT). Summary of the Invention

[0003] This invention provides a method for medical data quality inspection based on the Internet of Things, characterized by comprising: S10. Based on the test metadata uploaded by IoT medical data testing equipment, construct a four-dimensional spatiotemporal event map and integrate equipment bias and network latency disturbances to calculate the spatiotemporal consistency index to identify underlying data misalignment. S20. Based on the test data identified through the consistency index and the continuously evolving medical and physiological coupling rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights. S30. Based on multi-source state data of equipment, people and environment, by matching human-machine state snapshots with the context-aware rule base and integrating equipment quality control and operational errors, a human-machine stability factor reflecting the current inspection behavior is obtained. S40. Based on the four-dimensional spatiotemporal event map and patient metadata of test records, the immune mechanism is simulated to obtain the population response vector, and the abnormal tolerance threshold is adjusted according to the epidemiological background to obtain the relative confidence index. S50, based on the spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index and multi-source feedback logs, obtains a three-state quality arbitration result through multi-level arbitration grid decision-making and strategy optimization.

[0004] The above-described IoT-based medical data quality inspection method involves constructing a four-dimensional spatiotemporal event map based on the inspection metadata uploaded by IoT medical data inspection devices, fusing device bias and network latency disturbances, and calculating a spatiotemporal consistency index to identify underlying data misalignments. Specifically, this method comprises the following sub-steps: By integrating IoT medical device testing data from the laboratory, emergency room, and wards, along with metadata such as patient sampling, device location, and network transmission logs, a four-dimensional spatiotemporal event graph is constructed in the graph database. By calculating the transportation time variance and network transmission variance corresponding to the spatial distance between the sampling point and the testing equipment, the maximum acceptable turnaround time is calculated, providing a benchmark for judging the time reasonableness of the test results. The spatiotemporal consistency index formula is used to quantify the credibility of the physical source of each inspection record, and the consistency index is used to identify anomalies in the underlying data.

[0005] The above-described IoT-based medical data quality inspection method, which calculates a physiological rationality score based on inspection data identified through a consistency index and a continuously evolving medical-physiological coupling rule base, involves strongly coupling physiological rules with interpretability weights. The method comprises the following sub-steps: By integrating authoritative guidelines and literature, a medical physiology coupling rule base is constructed, from which strongly coupled physiological rules are extracted, and a structured rule matching list is generated for each test record; When a patient is diagnosed with a rare disease that includes an ICD-10 code, the exception tolerance mechanism is triggered and the interpretability weight is calculated, thereby reasonably accommodating special but real test anomalies caused by rare diseases while maintaining the universality of the rules. Based on all triggered strongly coupled physiological rules, and combined with interpretability weights, a physiological rationality score is calculated to determine whether the test results conform to known medical principles, thus avoiding misjudging real abnormalities as technical errors.

[0006] The above-described IoT-based medical data quality inspection method, based on multi-source state data of devices, people, and environment, obtains a human-machine stability factor reflecting the current inspection behavior by matching human-machine state snapshots with a context-aware rule base and fusing device quality control and operational errors. Specifically, it comprises the following sub-steps: The system synchronously collects the rate of change of the coefficient of variation of equipment quality control, operator shift and workload, laboratory environmental parameters and physiological indicators monitored by wearable devices through a sliding window, and generates a human-machine status snapshot after time alignment and encapsulation. Based on the human-machine state snapshot matching rules in the context-aware rule base and the assessment of the impact of equipment failure, human risk scores and equipment risk scores are calculated respectively, and these two scores guide the construction of a structured causal reasoning chain. Based on the hazard weights of activated high-risk scenarios, equipment quality control data, and operator operational errors, a human-machine stability factor is generated through function mapping to quantify the credibility of the source of test results.

[0007] The above-described IoT-based medical data quality inspection method, which uses a four-dimensional spatiotemporal event map and patient metadata from inspection records to simulate an immune mechanism and obtain a population response vector, and adjusts the anomaly tolerance threshold according to the epidemiological background to obtain a relative reliability index, is specifically divided into the following sub-steps: Based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates is dynamically constructed using multi-hop traversal. The current test sample is regarded as an antigen, and the neighborhood of similar patients is regarded as an immune cell pool. The test distance between the sample and each neighboring member is converted into antibody activation intensity through Gaussian kernel function to form a population response vector. The immune recognition mechanism is used to suppress isolated abnormalities. The relative credibility index is calculated using the herd immunity recognition entropy formula to accurately distinguish between clinically true critical values ​​and technically false anomalies.

[0008] The above-described IoT-based medical data quality inspection method, based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, dynamically constructs a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates through multi-hop traversal. This method comprises the following sub-steps: In the four-dimensional spatiotemporal event graph, the node representing the current record to be examined is accurately retrieved and anchored. This node has been associated with its unique patient identifier and basic demographic attributes. Perform a multi-hop traversal from the graph. The first hop starts from the current node and traverses along predefined semantic relationships to its clinical feature nodes such as diagnostic code, age group, gender, and underlying diseases to extract the individualized medical background. The second hop traces back to the specific sampling location or the location of the detection equipment by following the relationship between sampling and processing, based on the most recently tested event node associated with the current node, and obtains the spatial context. The third step involves filtering all other test event nodes that occur within a preset window before and after the current test time in the graph to ensure that the group comparison has timeliness consistency and avoid interference from cross-seasonal or long-term trends. The fourth hop further traverses other patient nodes that meet all of the following conditions: they share the primary diagnosis with the current patient, are in the same or similar clinical region, their age difference does not exceed the threshold, and they have recently completed the same test items, thereby dynamically constructing a neighborhood of similar patients with high clinical similarity.

[0009] The above-described IoT-based medical data quality inspection method, based on spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index, and multi-source feedback logs, undergoes multi-level arbitration grid decision-making and strategy optimization to obtain a three-state quality arbitration result. Specifically, it comprises the following sub-steps: By inputting four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index—into a multi-threshold decision grid with dynamically loaded weights and hierarchical thresholds according to the type of inspection item, intelligent three-level arbitration driven by comprehensive arbitration score is achieved. By embedding a rule engine after a multi-threshold decision grid, precise corrections can be made for a few high-risk, highly specific medical exception scenarios. By automatically feeding back the clinical confirmation results after arbitration, the system dynamically adjusts the equipment stability profile, risk scenario weights, and population anomaly tolerance thresholds, enabling incremental parameter updates and continuous self-correction and adaptive evolution in real-world diagnostic and treatment environments.

[0010] This invention also provides an Internet of Things-based medical data quality inspection system, comprising: Consistency Module: Based on the test metadata uploaded by IoT medical data testing equipment, a four-dimensional spatiotemporal event map is constructed and the device bias and network latency disturbances are integrated to calculate the spatiotemporal consistency index and identify underlying data misalignment; Physiological rationality module: Based on the test data identified through the consistency index and the continuously evolving medical physiology coupling rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights; Human-machine stability module: Based on multi-source state data of equipment, people and environment, it obtains human-machine stability factors that reflect the current inspection behavior by matching human-machine state snapshots with the context-aware rule base and integrating equipment quality control and operational errors. Relative credibility module: Based on the four-dimensional spatiotemporal event map and patient metadata of test records, the immune mechanism is simulated to obtain the population response vector, and the abnormality tolerance threshold is adjusted according to the epidemiological background to obtain the relative credibility index; Quality arbitration module: Based on spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index and multi-source feedback logs, a three-state quality arbitration result is obtained through multi-level arbitration grid decision-making and strategy optimization.

[0011] The beneficial effects achieved by this invention are as follows: This invention improves the authenticity of medical data sources and results through data quality inspection. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0013] Figure 1This is a flowchart of a medical data quality inspection method based on the Internet of Things provided in Embodiment 1 of this application.

[0014] Figure 2 This is a schematic diagram of a medical data quality inspection system based on the Internet of Things provided in Embodiment 2 of this application. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1

[0017] like Figure 1 As shown, Embodiment 1 of this application provides a method for medical data quality inspection based on the Internet of Things, including: S10. Based on the test metadata uploaded by IoT medical data testing equipment, construct a four-dimensional spatiotemporal event map and integrate equipment bias and network latency disturbances to calculate the spatiotemporal consistency index to identify underlying data misalignment.

[0018] S11. Integrate IoT medical device testing data from the laboratory, emergency room, and wards, along with metadata such as patient sampling, device location, and network transmission logs, to construct a four-dimensional spatiotemporal event graph in the graph database.

[0019] This process receives real-time raw test data streams from IoT medical devices deployed in the laboratory, emergency room, and wards. IoT medical devices refer to all networked analytical instruments throughout the hospital capable of automatically generating and uploading structured quantitative test results. Results from purely vital sign monitoring devices, imaging equipment, and manually recorded or manually entered results from paper reports are not included in this quality control process. Simultaneously, the process obtains the corresponding patient's unique EMPI number, the sampling timestamp entered by the nurse station's LIS system, the sample container barcode information, and the fixed workstation code and unique device ID registered during installation. The fixed workstation code uses a naming convention of area-floor-functional area-number. The network middleware records a complete transmission log (including sending time, receiving time, hop count, and latency) for each message from the device to the cloud. After the above multi-source heterogeneous metadata is uniformly parsed, a spatiotemporal event graph is dynamically constructed in the graph database, with test events as nodes. Node attributes include fields such as test item name, numerical result, device unique ID, patient EMPI number, sampling time, device physical location, and transmission delay. Semantic relationships such as continuous testing of the same patient, adjacent operations on the same device, and cross-device association within the same time window are used as edges. This graph supports millisecond-level backtracking of the full-cycle test trajectory for any patient and can quickly identify concurrent operations on the same patient by multiple devices within the same time window, providing a high-fidelity, traceable structured context foundation for subsequent spatiotemporal consistency verification.

[0020] S12. By calculating the variance of transportation time and network transmission corresponding to the spatial distance between the sampling point and the testing equipment, the maximum acceptable turnaround time is calculated, providing a benchmark for judging the time reasonableness of the test results.

[0021] Based on the constructed four-dimensional spatiotemporal event map, three key disturbance sources are integrated to characterize the real-world uncertainties in the data generation process: First, the built-in or connected quality control modules of the LIS are queried to automatically retrieve the quality control results for each testing device over the past 30 days for each testing item. These results are then validated using Westgard rules to obtain valid quality control data. Based on this valid data, the average system bias value for each item on a given device is calculated—the percentage difference between the quality control measurement mean and the target value. This bias value serves as a static disturbance factor for the current measurement accuracy of the device, used to subsequently reduce the credibility weight of its output results. Second, the fixed-station codes of sampling points and testing devices are parsed, and corresponding path categories are matched in the hospital building topology database, including same-floor, same-floor, direct-connection across buildings, non-direct-connection across buildings, special areas, and POCT local testing. Special areas include high-control areas such as ICU, operating rooms, NICU, and isolation wards, where sample transportation follows a dedicated process. Subsequently, the delivery time statistics table for each path category generated from the sample delivery logs was queried to obtain the historical average delivery time and its standard deviation for that path category. A normal distribution model was then constructed using these two values ​​to represent the expected time delay from sampling to loading and its natural fluctuation range. The delivery time variance was calculated based on the standard deviation of the delivery time. Third, using the transmission logs recorded by the network middleware, the message transmission delay distribution from each device to the cloud over the past 7 days was statistically analyzed. The mean and 95th percentile of the delay distribution were extracted, and an empirical distribution was fitted to calculate the standard deviation. The network transmission variance was then calculated based on this standard deviation.

[0022] Subsequently, based on the workstation codes of the sampling points and testing equipment, the path category of the current inspection event is automatically determined. The corresponding transportation time variance and network transmission variance are then queried. The two variances are superimposed using the Gaussian error propagation method to obtain the transportation time uncertainty variance. .based on The maximum acceptable turnaround time is preset for each inspection item. ,in The necessary processing time for the current testing item is determined by the laboratory's standard operating procedures. g is the clinical confidence coefficient, with g=2 corresponding to a 95% confidence level. The maximum acceptable turnaround time provides a dynamic, quantifiable, and clinically acceptable benchmark for the timeliness of test results, thereby supporting intelligent assessment and arbitration decisions regarding data quality.

[0023] S13. The spatiotemporal consistency index formula is used to quantify the credibility of the physical source of each inspection record, and the consistency index is used to identify underlying data anomalies.

[0024] The spatiotemporal consistency index formula is used to quantitatively assess the reliability of the spatiotemporal source of each test record, verifying whether the test data truly originates from the patient, time, location, and equipment claimed in the data record. The specific formula is as follows: ,in, As a spatiotemporal consistency index, It is the standard normal cumulative distribution function, used to transform the time deviation after turnover threshold correction into the cumulative probability of the deviation occurring under the ideal Gaussian perturbation model, that is, a probabilistic measure of the degree of time inconsistency. The LIS system sampling time for the current test record serves as the authoritative sampling timestamp. To verify the time when the result was successfully transmitted to the LIS system and confirmed to have been received, This is the absolute value of the actual turnaround time from sampling to reception. For the maximum acceptable turnaround time, This represents the variance of the uncertainty in transportation time. This represents the average system bias, expressed as a percentage. This is the absolute value of the bias. The maximum allowable bias in clinical practice is set at 5%. The test result represents the actual end-to-end network delay from the device sending the data to the LIS receiving it. The baseline network latency is the median of the network transmission latency of the hospital's testing equipment under normal communication conditions. A value greater than 0.95 indicates that the test record is highly reliable. A value within the range of [0.7, 0.95] indicates that the test record is basically reliable, but it needs to be judged comprehensively in conjunction with other steps. If the value is less than 0.7, the test record has significant sample confusion, mislabeling, or data misalignment caused by equipment clock drift and serious inaccuracy, and should be intercepted or manually reviewed.

[0025] S20. Based on the test data identified through the consistency index and the continuously evolving medical and physiological coupling rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights.

[0026] S21. By integrating authoritative guidelines and literature, a medical physiology coupling rule base is constructed, from which strongly coupled physiological rules are extracted, and a structured rule matching list is generated for each test record.

[0027] First, a continuously updated medical-physiological coupling rule base is loaded. The data for this rule base is obtained periodically from the UpToDate clinical advisor website via an API interface, which provides the latest clinical guidelines and evidence-based recommendations. The latest versions of national and international treatment guidelines are downloaded and parsed into structured or unstructured data. Abstracts or full texts of literature related to laboratory medicine are also downloaded in batches from the PubMed database. For structured data obtained from UpToDate clinical advisors and national and international treatment guidelines, the returned fields are standardized in terms of terminology and directly parsed into subject-relational-object triples. For PDF / webpage versions of national and international treatment guidelines and unstructured data obtained from the PubMed database, a pre-trained medical natural language processing model is used to identify target entities, including laboratory indicators, pathological states, and clinical manifestations. Dependency parsing is used to extract subject-verb-object structures, resulting in subject-relational-object triples. These triples are stored using a graph database, forming a semantic network centered on three node types: test indicators, pathological states, and clinical manifestations. Each node type in the network contains the following metadata fields: standardized unique identifier, readable name, entity type, reference range for test indicators, unit, list of data sources, and last update timestamp. Each edge contains the following metadata: relation type, whether it is directed, triggering condition, level of evidence (1-5), pathological hazard coefficient, list of supporting literature PMIDs, confidence score, applicable context, and main source. The triggering conditions are divided into simple threshold type and complex logic type. The simple threshold type directly extracts values, operators, and units from the text returned by the structured guidelines or API through regular expressions and terminology standardization. The complex logic type requires processing the text in the structured clinical guidelines. The data, including the multivariate joint judgment rules described in the description, the patient's historical test results, diagnostic records and timestamps stored in the hospital information system, and the composite relationships extracted by the natural language processing model, are generated through symbol parsing and rule template matching using methods such as clinical semantic understanding, baseline variable binding, time window inference, and logical connector recognition. The evidence level is obtained by converting the edge source type into a level according to preset rules. The pathological hazard coefficient is the score given by experts based on conditions such as disease mortality rate and intervention urgency, ranging from [0.1,1]. The confidence score is the output of the natural language processing model. The applicable context is the commonly occurring limiting conditions automatically summarized from a large number of documents or guidelines supporting the same medical relationship. When there are conflicting descriptions, the conflict resolution algorithm based on the evidence level coordinates and integrates them.

[0028] Strongly coupled physiological rules with explicit logical constraints or high co-occurrence probability are extracted from this semantic network. These rules include pairwise rules, group rules, and dynamic threshold rules. Pairwise rules refer to explicit logical constraints between two test indicators or between a test indicator and a clinical manifestation. Group rules refer to the association between multiple test indicators. Dynamic threshold rules are threshold conditions set based on time series or baseline changes. The rule extraction method is as follows: Pattern matching is used to search for relationship chains in the semantic network that conform to three predefined rule patterns: pairwise rules, grouped rules, and dynamic threshold rules. Statistical analysis is then used to test the co-occurrence frequency and significance of these relationship chains with historical hospital test data. This verifies whether the rule relationship chain actually occurs frequently within the hospital, avoiding the misclassification of theoretically valid but locally rare associations as strongly coupled rules. Furthermore, the co-occurrence and testing process identifies strong association patterns unique to the hospital, thereby selecting relationship chains with stable clinical associations as high-confidence rule candidates. Finally, clinical experts review each candidate rule, considering pathophysiological mechanisms and actual clinical scenarios to confirm their scientific validity, rationality, and applicability boundaries, ensuring that the rules ultimately included in the database are both evidence-based and clinically feasible.

[0029] When a new test record enters the quality assessment process, all relevant strongly coupled physiological rules are searched in the medical-physiological coupling rule base based on the included test items. All corresponding rule nodes are activated, and each matching rule is labeled with two key metadata pre-defined during the rule base construction: First, the evidence level: Level 1 indicates the rule originates from UpToDate clinical advisor recommendation strength or from PubMed literature containing RCT types in the medical subject terms and recommended by treatment guidelines; Level 2 indicates high-quality cohort studies or systematic reviews confirmed by expert sampling; Level 3 indicates case series of more than 5 cases or retrospective observational studies; Level 4 indicates expert consensus, textbooks, or clinical experience summaries; Level 5 indicates single case reports, inference rules, or weak associations automatically mined by natural language processing models. Second, the pathological hazard coefficient, reflecting the severity of the potential clinical risk corresponding to the rule violation. Finally, a structured rule matching list is output for each test record, including rule ID and description, whether it has been violated, evidence level, and pathological hazard coefficient, serving as the prior knowledge basis for subsequent physiological consistency scoring.

[0030] S22. When a patient is diagnosed with a rare disease that includes an ICD-10 code, the exception tolerance mechanism is triggered, and the interpretability weight is calculated, so as to reasonably accommodate special but real test anomalies caused by rare diseases while maintaining the universality of the rules.

[0031] When a patient's primary or secondary diagnosis contains a rare disease code as defined by ICD-10, an exception tolerance mode is automatically triggered: First, the strong constraint rule violated by the current test result is identified from the rule matching list. Then, using the rare disease name and the violated rule as joint query conditions, a search is conducted in a pre-built evidence-based medicine knowledge base (integrating clinical cases and expert consensus literature from Orphanet, OMIM, and PubMed) to determine if there is any literature evidence that clearly supports the disease as a reasonable explanation for this abnormal combination. If at least one high-quality article (case report, review, or...) exists... As mentioned in the guidelines, the strength of evidence is extracted, categorized as high, medium, and low. High strength requires that the rare disease be explicitly included in Orphanet as having typical laboratory manifestations, supported by case series (≥5 cases), and ≥10 PubMed citations. Medium strength requires ≥2 independent case reports or 1 case series (≥3 cases) describing the association, and ≥3 PubMed citations. Low strength requires a single case report or inferential description, <3 PubMed citations, or no citations. The interpretability weight of this rule in the current patient context is then calculated based on this strength of evidence. The value range is (0,1]. These are the weighting coefficients for Orphanet inclusion, document type level, and citation count, respectively, with values ​​set to 0.5, 0.3, and 0.2. These represent Orphanet inclusion, document type level, and citation count, respectively. Being included in Orphanet means... The value is 1, and the value is 0 otherwise. The following reference levels were used: RCT = 1.0, cohort = 0.8, case series = 0.6, individual cases = 0.3. The value is 1 for references ≥ 50, 0.7 for references 10–49, and 0.4 for references < 10. This indicates that it is fully interpretable. If no supporting evidence is found, this weight is then used to dynamically reduce the penalty for violating the rule in subsequent physiological consistency scores, thereby ensuring the rule's universality while preserving clinical inclusivity and interpretability for the special pathophysiological manifestations of rare diseases.

[0032] S23. Based on all triggered strongly coupled physiological rules, and combined with interpretability weights, calculate the physiological rationality score to determine whether the test results conform to known medical laws, and avoid misjudging real abnormalities as technical errors.

[0033] By combining the violation status of all triggered strongly coupled physiological rules in the comprehensive rule matching list, and considering whether the patient suffers from a rare disease that could explain these abnormalities, a physiological rationality score is output to determine whether the current abnormality is a true pathological manifestation or a potential data quality issue. The specific formula is as follows: ,in, This is a score for physiological rationality, with a value range of [0,1]. The closer the value is to 1, the more reasonable the test data is; the closer it is to 0, the more serious and inexplicable the physiological contradiction, and the questionable the data quality. R represents the number of strongly coupled rules activated in the current test record. Let 'harm' represent the pathological hazard coefficient of the r-th rule, where 'harm' indicates the hazard. denoted as the original degree of violation of the r-th rule, representing the absolute or relative amount by which the test value deviates from the rule threshold. For static threshold rules, the absolute deviation value is used; for dynamic / baseline dependent rules, the relative percentage change is used. The standard deviation of the population for the indicator corresponding to the r-th rule is denoted as , and . Within the same institution, estimates are made from historical hospital data or publicly available databases to standardize the degree of anomalies, making different indicators comparable. is the exponential compression sensitivity parameter, controlling the nonlinear response of the violation intensity, with a value range of [1,3]. exp() is an exponential function that achieves saturated compression, preventing extreme anomalies from dominating the score. Let be the interpretability weight of the r-th rule. This is a rare disease indicator function. It takes a value of 1 if the patient's primary / secondary diagnosis contains an ICD-10 rare disease code associated with rule r, and a value of 0 otherwise. x represents the currently evaluated test record and its associated patient clinical context. A value greater than or equal to 0.85 indicates that the data is highly reasonable. Values ​​in the range [0.6, 0.84] indicate mild or partially explainable anomalies, but without high-risk contradictions. These should be recorded as low-priority for review or recorded for observation. A value less than 0.6 indicates a significant and unexplained physiological inconsistency, suggesting a potential data quality problem.

[0034] S30. Based on multi-source state data of equipment, people and environment, by matching human-machine state snapshots with the context-aware rule base and integrating equipment quality control and operational errors, a human-machine stability factor reflecting the current inspection behavior is obtained.

[0035] S31. The equipment quality control variation coefficient change rate, operator shift and workload, laboratory environmental parameters and physiological indicators monitored by wearable devices are collected synchronously through a sliding window, and a human-machine status snapshot is generated after time alignment and encapsulation.

[0036] Using a 10-minute sliding window, the following multi-source heterogeneous data were synchronously collected and time-aligned: First, historical quality control data was extracted from the quality control management system of the testing equipment. For each key testing item, the mean and standard deviation of the quality control results within the sliding window were calculated, and the coefficient of variation was calculated based on these two values. The relative change rate of the coefficient of variation compared to one week ago was used as the coefficient of variation change rate to quantify the stability trend of instrument performance. Second, information such as the identity, on-duty hours, continuous working hours, shift records, and job qualifications of the current operator within the window were obtained from the hospital's scheduling and human resources information system to assess human error risks. Third, environmental parameters such as temperature, humidity, atmospheric pressure, cleanliness, and electromagnetic field strength were collected in real time by environmental sensors deployed in the testing room to identify external physical factors that interfere with the test results. Fourth, physiological indicators such as heart rate variability, skin conductance, and hand micro-tremor frequency were continuously recorded using wearable devices worn by operators as objective proxy variables for operator fatigue, decreased attention, or stress. All four types of data are aligned with a unified timestamp using a 10-minute sliding window, and the four data items are structured and encapsulated for each time window as a snapshot of human-machine state, providing high-dimensional contextual features for anomaly attribution and root cause analysis in subsequent sub-steps.

[0037] S32. Based on the human-machine state snapshot, match the rules in the context-aware rule base and evaluate the impact of equipment failure. Calculate the human risk score and the equipment risk score respectively, and use these two scores to guide the construction of the structured causal reasoning chain.

[0038] First, a predefined scenario-aware rule base is loaded. This base stores multi-factor joint rules in a condition-conclusion format. Each rule contains multiple preconditions and a conclusion, along with an expert-defined confidence level. For example, the combination of low heart rate variability and high temperature suggests decreased operator hand stability and increased risk of operational errors, or the combination of prolonged continuous work and increased hand microtremors suggests an increased probability of pipetting errors. Then, a Boolean logic match is performed between the generated 10-minute window snapshot of the human-machine interface and the rules in the scenario-aware rule base. If the current state combination meets the preconditions of a scenario rule, the corresponding high-risk scenario label is activated, and its confidence level is recorded. Simultaneously, the system retrieves the fault logs and performance deviation records of the testing equipment for the past 7 days from the equipment maintenance database. These records include, but are not limited to, light source intensity attenuation, needle blockage, and temperature control drift. Combined with the testing principles of the current test item, the system queries the corresponding sensitivity coefficients in a predefined sensitivity matrix of equipment faults and testing methods. This matrix is ​​jointly calibrated by the clinical engineering team, laboratory experts, and the equipment manufacturer and is fixed in the form of a database table or configuration file. The human risk score is obtained by weighted summation of the confidence levels of activated high-risk scenario rules. Simultaneously, the severity of each equipment failure item is multiplied by its sensitivity coefficient and summed to obtain the equipment impact score. The construction of causal paths is guided by the human risk and equipment impact scores: First, an activation threshold is set, and only risk scenarios and equipment exceeding the threshold are included in the causal chain to avoid interference from low-relevance noise. Second, among multiple potential paths, priority is given to constructing and displaying high-scoring primary causal paths. Third, the score is used as a confidence strength label for causal nodes to enhance the credibility and transparency of the explanation. Fourth, when the human and equipment paths both point to the same test anomaly, the fusion method is determined by the ratio of their scores. If the human score is significantly higher than the equipment score, the human condition factor is highlighted in the causal chain, and operational intervention is recommended; conversely, the equipment defect is emphasized, and maintenance procedures are triggered. Based on this, entities and causal predicates in all high-risk scenario rules are extracted and linked together according to a preset logical topology using environmental factors, personnel status, equipment status, and test items. The resulting causal reasoning chain is presented in the form of a structured directed graph, with each link associated with the original evidence source and corresponding quantitative score support, thereby generating a scenario-aware causal reasoning chain that is clearly structured, traceable in evidence, measurable in confidence, and clinically operable.

[0039] S33. Based on the hazard weights of activated high-risk scenarios, equipment quality control data, and operator operation errors, a human-machine stability factor is generated through function mapping to quantify the credibility of the source of the test results.

[0040] Based on the quality control principle that the default human-machine collaboration state is in a highly reliable state, and the reliability is only reduced when there is an identifiable risk, the human-machine stability factor is calculated as follows: Based on the high-risk scenarios in step S32, only the preset clinical hazard weights of these activated scenarios are added together and subtracted from the initial stability baseline value 1 to obtain the baseline stability score. The preset clinical hazard weight refers to the negative impact coefficient of high-risk clinical scenarios on the stability of test results, with a value range of [0.02, 0.2]. A value of 0.05 indicates mild risk, a value of 0.1 indicates moderate risk, and a value of 0.15 to 0.2 indicates severe risk. The specific score is determined by the expert panel. Simultaneously, the historical quality control performance of the testing equipment over the past 30 days was retrieved, including the coefficient of variation compliance rate, frequency of out-of-control events, and calibration pass rate. These three items were weighted and averaged to generate an equipment reliability score. The equipment reliability score was then multiplied by an equipment scoring sensitivity coefficient and added to the baseline stability score to obtain the equipment reliability bonus. The equipment scoring sensitivity coefficient ranged from [0.2, 0.3], with a default initial value of 0.25, determined through consultation among multidisciplinary experts based on typical clinical scenarios. This ensured that the historical performance of the equipment provided appropriate, reasonable, and non-overreaching positive compensation, neither excessively penalizing reliable equipment nor condoning high-risk operations. Furthermore, operator proficiency was introduced as a proportional adjustment factor for the overall score. The operator proficiency score was multiplied by the equipment reliability bonus to obtain an adjusted score. This proficiency was calculated from the average operational error of the operator performing similar testing items over the past 6 months; the smaller the error, the higher the adjustment weight, thereby amplifying or suppressing the current total score. The adjusted scores are nonlinearly compressed using an S-shaped monotonic function, and the original scores are smoothly mapped to [0,1] to generate a human-machine stability factor. The closer the factor is to 1, the more reliable the current human-machine collaboration is, and the higher the source credibility of the test results. The closer it is to 0, the more significant the human or equipment risk is, and the lower the confidence weight of the data needs to be in subsequent arbitration.

[0041] S40. Based on the four-dimensional spatiotemporal event map and patient metadata of the test records, the immune mechanism is simulated to obtain the population response vector, and the abnormal tolerance threshold is adjusted according to the epidemiological background to obtain the relative confidence index.

[0042] S41. Based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates is dynamically constructed using multi-hop traversal.

[0043] First, the clinical characteristics of the patient corresponding to the current test sample to be evaluated are obtained in real time from the hospital's electronic medical record system, inpatient management platform, and medical order records. This includes de-identified clinical metadata information such as age, primary diagnosis code, inpatient department, and type of medication currently being used. Next, the nodes representing the current test record are precisely retrieved and anchored in the four-dimensional spatiotemporal event graph generated in step S11. This node has been associated with its unique patient identifier and basic demographic attributes. A multi-hop traversal is performed from the graph to form a neighborhood of similar patients: the first hop starts from... Starting from the beginning, the system traverses along predefined semantic relationships to clinical feature nodes such as diagnostic codes, age groups, gender, and underlying diseases to extract personalized medical context. The second hop passes through... The most recently associated test event node is traced back to the specific sampling location or testing equipment location along the relationship between sampling and processing to obtain the spatial context. The third hop filters all other test event nodes that occurred within a preset window before and after the current test time in the graph to ensure that the group comparison has timeliness consistency and avoids interference from cross-seasonal or long-term trends. The fourth hop further traverses other patient nodes that meet all of the following conditions: sharing the primary diagnosis with the current patient, being in the same or similar clinical area, having an age difference not exceeding a threshold, and having recently completed the same test items, thereby dynamically constructing a neighborhood of similar patients with high clinical similarity. The numerical attributes of the target items are extracted from the corresponding test event nodes in the neighborhood and aggregated into a structured numerical set as the basic data source for subsequent calculation of the group distribution. This neighborhood is not a static cache, but is automatically updated daily based on dynamic events such as new visits, follow-up visits, medication discontinuation, diagnosis changes, and test requests, and is reconstructed each time a test rationality assessment is triggered to ensure that the selected neighborhood truly reflects the hospital's current disease admission structure, epidemiological distribution, and treatment patterns.

[0044] S42. Treat the current test sample as an antigen and the neighborhood of similar patients as an immune cell pool. Use a Gaussian kernel function to convert the test distance between the sample and each neighborhood member into antibody activation intensity, forming a population response vector. Use the immune recognition mechanism to suppress isolated abnormalities.

[0045] The current test record sample to be evaluated is considered an antigen, and its multidimensional test results (a vector composed of indicators such as liver function, kidney function, and electrolytes) are used as input. Simultaneously, patients in the neighborhood of similar patients are considered an immune cell pool, with each neighborhood member using its own test result vector as an antibody template. The Euclidean distance between the test record sample and each neighboring patient in the standardized test space is calculated, and this distance is converted into antibody activation intensity using a monotonically decreasing Gaussian kernel response function. The closer the distance, the stronger the activation intensity, indicating that the neighboring member recognizes the current sample as its own or a normal variant. The activation intensities of all patients in the neighborhood of similar patients are combined to form a population response vector. Because this immune mechanism relies on population consensus, if the current sample is a true abnormality, its test vector will significantly deviate from the entire neighborhood distribution, resulting in extremely low activation intensity values ​​for the vast majority of samples and a weak overall response vector. Conversely, if it is a reasonable physiological fluctuation, it will be effectively recognized by multiple neighboring members, producing a strong response. This simulated immune recognition process naturally possesses the ability to suppress isolated outliers; a single abnormality cannot obtain a coordinated response from the population antibodies.

[0046] S43. Calculate the relative credibility index using the herd immunity recognition entropy formula to accurately distinguish between clinically true critical values ​​and technically false abnormalities.

[0047] The herd immunity recognition entropy formula is used to quantify the consistency and acceptability of current test results in herd immunity recognition. Epidemiological background is incorporated to dynamically modulate the anomaly assessment and calculate the relative confidence index, thereby more accurately distinguishing between clinically true critical values ​​and technical artifacts. The specific formula is as follows: Where K is the number of patients in the neighborhood of similar patients. For the group response vector, This represents the antibody response entropy. The lower the entropy value, the more concentrated the population response and the more consistent the judgment. The normalized activation probability. Let be the antibody activation intensity of the b-th neighboring patient in response to the current sample. This involves summing the activation strengths of all K neighboring members. For indicator functions, when It is 1 if it is true, otherwise it is 0. This is the multidimensional test result vector for the current sample. Let b be the vector of test results from the b-th neighboring patient. Let be the Euclidean distance between the current sample test result vector and the test result vector of the b-th neighboring patient in the test space. This is the distance scaling factor, with an initial value of 1.96, corresponding to the 95% confidence boundary. Let be the standard deviation of the corresponding item in the test sample for the b-th neighboring patient in their historical test results. This is the epidemiological moderating sensitivity coefficient, ranging from [0.1, 0.3], with an initial value of 0.2. This means that when the epidemiological index is 1 (i.e., a pandemic of a disease), the anomaly detection threshold can be relaxed by a maximum of 0.2. . This is an epidemiological index, with values ​​ranging from [0,1]. A value of 0 indicates that there is currently no specific disease associated with the tested sample that is in an epidemic state. A value of 1 indicates that the specific disease associated with the test sample is in a highly prevalent state. The epidemiological index calculation process is as follows: For the target disease associated with the current test record, dynamically relevant and available data from three categories—real-time in-hospital data, regional public health data, and official warning levels—are selected. These data are then stratified and weighted according to the clinical reliability of each type of data to generate an epidemiological index reflecting the intensity of disease prevalence. This index is used to intelligently adjust the leniency of test anomaly judgment. The output EM-IRE value is the relative reliability index, ranging from [0,1]. An EM-IRE value close to 1 indicates a consistent population response and acceptance of the result by most neighboring members, classifying it as a true anomaly or reasonable variation. An EM-IRE value close to 0 indicates a scattered response with almost no acceptance, classifying it as an isolated false anomaly, requiring retesting or manual review.

[0048] S50, based on the spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index and multi-source feedback logs, obtains a three-state quality arbitration result through multi-level arbitration grid decision-making and strategy optimization.

[0049] S51. By inputting four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index—into a multi-threshold decision grid with dynamically loaded weights and hierarchical thresholds according to the type of inspection item, intelligent three-level arbitration driven by comprehensive arbitration score is achieved.

[0050] The generated spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index are used as inputs and fed into a pre-set multi-threshold decision grid. This grid is not based on fixed rules but dynamically loads corresponding weight configurations and stratified threshold strategies according to the type of the test item: First, the type of the current test item is identified and classified into critical value items, routine clinical test items, or scientific research exploratory items according to pre-set classification rules. Then, based on the item type, a weight vector and a stratified decision threshold set strictly bound to that category are dynamically loaded from a pre-set strategy configuration library. The weight vector defines the linear combination coefficients of the four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index—in the comprehensive score, ensuring that the contribution of each dimension of evidence is differentiated under different risk categories; the stratified decision threshold set contains three ordered thresholds. These correspond to three arbitration actions: automatic release, restricted access (requiring manual review), and forced interception. Automatic release is for highly reliable test data, which is directly included in the clinical report. Restricted access is for questionable test results, which are pushed to the physician for confirmation along with explanatory evidence (causal chain). Forced interception is for high-probability technical artifacts, triggering a re-examination process. The comprehensive arbitration score is obtained by weighting and summing the four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative reliability index—according to their respective weights. Then Compare with the threshold set corresponding to the current category: If If so, the result will be automatically approved; if If it is marked as restricted access and pushed to the manual review queue, then... If the result is not met, a mandatory interception is triggered, preventing it from entering the clinical reporting process. All weights and thresholds are determined by multidisciplinary experts based on historical misjudgment rates, the severity of clinical consequences, and operational feasibility, and are continuously optimized with a closed-loop feedback mechanism, thereby achieving dynamic adaptive arbitration driven by risk level and executed within a multi-threshold grid framework.

[0051] S52. By embedding a rule engine after the multi-threshold decision grid, precise correction can be performed on a few high-risk, highly specific medical exception scenarios.

[0052] This step, based on the multi-threshold decision grid, specifically addresses high-risk exceptions that cannot be accurately arbitrated by general weighted scoring due to the unique nature of medical logic. The specific implementation process is as follows: During the deployment phase, a clinical expert team identifies several typical scenarios with high historical misjudgment frequencies, severe clinical consequences, or contradictions with conventional statistical models. For example, rhabdomyolysis leads to a significant increase in creatine kinase, but there are no similar cases in the neighborhood of similar patients. A high-priority structured rule is pre-defined for each scenario. Each rule is solidified in the form of a rule engine. The conditions of the high-priority structured rule not only use the four indicators from step S51 but also forcibly bind specific clinical contexts such as diagnosis, medication, or test combinations, forming highly specific judgment logic. For example, if the relative confidence index is low (<0.6) but the primary diagnosis includes rhabdomyolysis and the test item is creatine kinase, then the grid judgment is overridden, and it is marked as restricted use rather than blocked. The rule engine is located after step S51 and before the result output in the arbitration process. Only when a rule is fully matched does it override the preliminary conclusion of step S51; otherwise, the arbitration result of step S51 is retained. This mechanism does not re-arbitrate all samples, but rather serves as a precise exception handling layer to ensure that key medical common sense is not lost while relying on the data-driven backbone, thereby improving the overall clinical safety and interpretability of the arbitration.

[0053] S53. By automatically feeding back the clinical confirmation results after arbitration, the system dynamically adjusts the equipment stability profile, risk scenario weights, and population anomaly tolerance thresholds, thereby achieving incremental parameter updates and enabling the system to continuously self-correct and adaptively evolve in real-world diagnostic and treatment environments.

[0054] After arbitration of each test record, the final confirmation result is automatically collected, including manual review conclusions, clinical follow-up feedback, and subsequent test verification. This confirmation feedback information is then structured and fed back to the underlying knowledge base and parameter configuration. If a result is forcibly intercepted but subsequently confirmed as a genuine clinical abnormality, the stability profile of the relevant equipment or operational procedure is modified. For example, the historical stability baseline of the equipment for the corresponding test item is lowered to prevent excessive punishment of reasonable abnormalities in similar scenarios in the future. Conversely, if a test result is automatically allowed but subsequently confirmed as a technical artifact or misjudgment, the pathological hazard coefficient of the high-risk scenario item that triggered the path in step S32 is automatically increased. This enhances the penalty for similar risks in the future. Simultaneously, the tolerance threshold er used for identifying anomalies in the vicinity of similar patients is dynamically fine-tuned based on seasonal disease monitoring data. It is appropriately relaxed during epidemic periods to avoid over-intercepting reasonable pathological changes, and tightened during non-epidemic periods to improve specificity. The entire process, through incremental updates of rule weights, device profiles, and threshold parameters, enables the system to continuously self-correct and adaptively evolve in real clinical environments, thereby constructing a data-driven, feedback-closed-loop, interpretable, and continuously optimizing intelligent testing mechanism.

[0055] This method achieves a triple leap in the paradigm of medical data quality inspection: First, it moves from static thresholds to dynamic context awareness, abandoning the rigid judgment that relies on fixed reference ranges and instead integrating real-time human-machine status and clinical context, making quality assessment more scenario-adaptive. Second, it upgrades from isolated detection to a herd immunity mechanism, constructing similar patient neighborhoods based on real-time data from across the hospital to simulate the immune system's logic of recognizing and tolerating abnormalities, effectively distinguishing between technical artifacts and real pathological signals, and significantly improving discrimination specificity. Finally, it evolves from one-way verification to closed-loop self-optimization, automatically converting multi-source feedback such as manual review and clinical follow-up into incremental updates of equipment files, rule weights, and herd thresholds, enabling continuous system evolution without model retraining. These three leaps together support a highly automated, robust, and reliable intelligent quality control system, significantly reducing the false alarm rate of technical artifacts and the missed alarm rate of critical values ​​while ensuring efficiency, comprehensively improving the clinical usability, security, and physician trust of laboratory data.

[0056] Example 2

[0057] like Figure 2 As shown, Embodiment 2 of this application provides a medical data quality inspection system based on the Internet of Things, including: Consistency Module: Based on the test metadata uploaded by IoT medical data testing equipment, a four-dimensional spatiotemporal event map is constructed and device bias and network latency disturbances are integrated to calculate the spatiotemporal consistency index and identify underlying data misalignments. Specifically, it is divided into the following sub-modules: Sub-module construction: Integrate IoT medical equipment test data from the laboratory, emergency room and wards, as well as metadata such as patient sampling, equipment location, and network transmission logs, to construct a four-dimensional spatiotemporal event graph in the graph database.

[0058] Maximum Turnover Module: By calculating the variance of transportation time and network transmission corresponding to the spatial distance between the sampling point and the testing equipment, the maximum acceptable turnover time is calculated, providing a benchmark for judging the time reasonableness of the test results.

[0059] The index calculation submodule uses the spatiotemporal consistency index formula to quantify the credibility of the physical source of each inspection record, and uses this consistency index to identify anomalies in the underlying data.

[0060] The physiological rationality module: Based on test data identified through the consistency index and a continuously evolving medical physiology coupling rule base, it calculates the physiological rationality score through strongly coupled physiological rules and interpretability weights. Specifically, it is divided into the following sub-modules: Rule List Submodule: By integrating authoritative guidelines and literature, a medical physiology coupled rule base is constructed, from which strongly coupled physiological rules are extracted, and a structured rule matching list is generated for each test record.

[0061] Interpretation weight submodule: When a patient is diagnosed with a rare disease that includes an ICD-10 code, the exception tolerance mechanism is triggered and interpretability weights are calculated, thereby reasonably accommodating special but real test anomalies caused by rare diseases while maintaining the universality of the rules.

[0062] The calculated molecular module: Based on all triggered strongly coupled physiological rules, combined with interpretability weights, a physiological rationality score is calculated to determine whether the test results conform to known medical laws, thus avoiding misjudging real abnormalities as technical errors.

[0063] Human-Machine Stability Module: Based on multi-source state data of equipment, human, and environment, this module matches human-machine state snapshots with a context-aware rule base and integrates equipment quality control and operational errors to obtain a human-machine stability factor reflecting the current inspection behavior. It is specifically divided into the following sub-modules: The snapshot generation submodule synchronously collects the rate of change of the equipment quality control coefficient of variation, operator shift and workload, laboratory environmental parameters and physiological indicators monitored by wearable devices through a sliding window, and generates a human-machine status snapshot after time alignment and encapsulation.

[0064] Constructing a causal chain submodule: Based on human-machine state snapshots, matching rules in the context-aware rule base and evaluating the impact of equipment failure, calculating human risk scores and equipment risk scores respectively, and using these two to guide the construction of a structured causal reasoning chain.

[0065] The stability factor generation submodule generates a human-machine stability factor based on the hazard weights of activated high-risk scenarios, equipment quality control data, and operator operation errors, which is used to quantify the credibility of the source of test results.

[0066] The relative reliability module, based on a four-dimensional spatiotemporal event map and patient metadata from test records, simulates the immune mechanism to obtain a population response vector. It then adjusts the abnormality tolerance threshold according to the epidemiological background to obtain a relative reliability index. Specifically, it consists of the following sub-modules: The patient neighborhood submodule: Based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates is dynamically constructed through multi-hop traversal.

[0067] The population response submodule treats the current test sample as an antigen and the neighborhood of similar patients as an immune cell pool. It uses a Gaussian kernel function to convert the test distance between the sample and each neighboring member into antibody activation intensity, forming a population response vector. It then uses the immune recognition mechanism to suppress isolated abnormalities.

[0068] The immune recognition submodule calculates the relative confidence index using the herd immunity recognition entropy formula to accurately distinguish between clinically true critical values ​​and technically false anomalies.

[0069] The quality arbitration module, based on the spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index, and multi-source feedback logs, obtains a three-state quality arbitration result through multi-level arbitration grid decision-making and strategy optimization. It is specifically divided into the following sub-modules: Decision-making grid submodule: By inputting four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index—into a multi-threshold decision grid with dynamically loaded weights and hierarchical thresholds according to the type of inspection item, intelligent three-level arbitration driven by comprehensive arbitration score is realized.

[0070] Correction Submodule: By embedding a rule engine after the multi-threshold decision grid, it can accurately correct a few high-risk and highly specific medical exception scenarios.

[0071] The calibration and optimization submodule automatically feeds back the clinical confirmation results after arbitration, dynamically adjusts the equipment stability profile, risk scenario weights, and population anomaly tolerance thresholds, and achieves incremental parameter updates, enabling the system to continuously self-calibrate and adaptively evolve in real diagnosis and treatment environments.

[0072] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for medical data quality inspection based on the Internet of Things, characterized in that, include: S10. Based on the test metadata uploaded by IoT medical data testing equipment, construct a four-dimensional spatiotemporal event map and integrate equipment bias and network latency disturbances to calculate the spatiotemporal consistency index to identify underlying data misalignment. S20. Based on the test data identified through the consistency index and the continuously evolving medical physiology coupling rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights. S30. Based on multi-source state data of equipment, people and environment, by matching human-machine state snapshots with the context-aware rule base and integrating equipment quality control and operational errors, a human-machine stability factor reflecting the current inspection behavior is obtained. S40. Based on the four-dimensional spatiotemporal event map and patient metadata of test records, the immune mechanism is simulated to obtain the population response vector, and the abnormal tolerance threshold is adjusted according to the epidemiological background to obtain the relative confidence index. S50, based on the spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index and multi-source feedback logs, obtains a three-state quality arbitration result through multi-level arbitration grid decision-making and strategy optimization.

2. The method for medical data quality inspection based on the Internet of Things as described in claim 1, characterized in that, Based on the test metadata uploaded by IoT medical data testing equipment, a four-dimensional spatiotemporal event map is constructed and device bias and network latency disturbances are integrated. The spatiotemporal consistency index is calculated to identify underlying data misalignment. The specific steps are as follows: By integrating IoT medical device testing data from the laboratory, emergency room, and wards, along with metadata such as patient sampling, device location, and network transmission logs, a four-dimensional spatiotemporal event graph is constructed in the graph database. By calculating the transportation time variance and network transmission variance corresponding to the spatial distance between the sampling point and the testing equipment, the maximum acceptable turnaround time is calculated, providing a benchmark for judging the time reasonableness of the test results. The spatiotemporal consistency index formula is used to quantify the credibility of the physical source of each inspection record, and the consistency index is used to identify anomalies in the underlying data.

3. The method for medical data quality inspection based on the Internet of Things as described in claim 1, characterized in that, Based on the test data identified through the consistency index and the continuously evolving medical physiology coupled rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights, specifically in the following sub-steps: By integrating authoritative guidelines and literature, a medical physiology coupling rule base is constructed, from which strongly coupled physiological rules are extracted, and a structured rule matching list is generated for each test record; When a patient is diagnosed with a rare disease that includes an ICD-10 code, the exception tolerance mechanism is triggered and the interpretability weight is calculated, thereby reasonably accommodating special but real test anomalies caused by rare diseases while maintaining the universality of the rules. Based on all triggered strongly coupled physiological rules, and combined with interpretability weights, a physiological rationality score is calculated to determine whether the test results conform to known medical principles, thus avoiding misjudging real abnormalities as technical errors.

4. The method for medical data quality inspection based on the Internet of Things as described in claim 1, characterized in that, Based on multi-source state data of equipment, people, and environment, by matching human-machine state snapshots with a context-aware rule base and integrating equipment quality control and operational errors, a human-machine stability factor reflecting the current inspection behavior is obtained. This process is divided into the following sub-steps: The system synchronously collects the rate of change of the coefficient of variation of equipment quality control, operator shift and workload, laboratory environmental parameters and physiological indicators monitored by wearable devices through a sliding window, and generates a human-machine status snapshot after time alignment and encapsulation. Based on the human-machine state snapshot matching rules in the context-aware rule base and the assessment of the impact of equipment failure, human risk scores and equipment risk scores are calculated respectively, and these two scores guide the construction of a structured causal reasoning chain. Based on the hazard weights of activated high-risk scenarios, equipment quality control data, and operator operational errors, a human-machine stability factor is generated through function mapping to quantify the credibility of the source of test results.

5. The method for medical data quality inspection based on the Internet of Things as described in claim 1, characterized in that, Based on a four-dimensional spatiotemporal event map and patient metadata from test records, the immune mechanism is simulated to obtain a population response vector. Then, the abnormality tolerance threshold is adjusted according to the epidemiological background to obtain a relative confidence index. This process is divided into the following sub-steps: Based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates is dynamically constructed using multi-hop traversal. The current test sample is regarded as an antigen, and the neighborhood of similar patients is regarded as an immune cell pool. The test distance between the sample and each neighboring member is converted into antibody activation intensity through Gaussian kernel function to form a population response vector. The immune recognition mechanism is used to suppress isolated abnormalities. The relative credibility index is calculated using the herd immunity recognition entropy formula to accurately distinguish between clinically true critical values ​​and technically false anomalies.

6. The method for medical data quality inspection based on the Internet of Things as described in claim 5, characterized in that, Based on real-time clinical metadata and a four-dimensional spatiotemporal event graph, a patient neighborhood with high clinical similarity, spatiotemporal alignment, and daily updates is dynamically constructed using multi-hop traversal. This process is divided into the following sub-steps: In the four-dimensional spatiotemporal event graph, the node representing the current record to be examined is accurately retrieved and anchored. This node has been associated with its unique patient identifier and basic demographic attributes. Perform a multi-hop traversal from the graph. The first hop starts from the current node and traverses along predefined semantic relationships to its clinical feature nodes such as diagnostic code, age group, gender, and underlying diseases to extract the individualized medical background. The second hop traces back to the specific sampling location or the location of the detection equipment by following the relationship between sampling and processing, based on the most recently tested event node associated with the current node, and obtains the spatial context. The third step involves filtering all other test event nodes that occur within a preset window before and after the current test time in the graph to ensure that the group comparison has timeliness consistency and avoid interference from cross-seasonal or long-term trends. The fourth hop further traverses other patient nodes that meet all of the following conditions: they share the primary diagnosis with the current patient, are in the same or similar clinical region, their age difference does not exceed the threshold, and they have recently completed the same test items, thereby dynamically constructing a neighborhood of similar patients with high clinical similarity.

7. The method for medical data quality inspection based on the Internet of Things as described in claim 1, characterized in that, Based on the spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index, and multi-source feedback logs, a three-state quality arbitration result is obtained through multi-level arbitration grid decision-making and strategy optimization, specifically divided into the following sub-steps: By inputting four indicators—spatiotemporal consistency index, physiological rationality score, human-machine stability factor, and relative credibility index—into a multi-threshold decision grid with dynamically loaded weights and hierarchical thresholds according to the type of inspection item, intelligent three-level arbitration driven by comprehensive arbitration score is achieved. By embedding a rule engine after a multi-threshold decision grid, precise corrections can be made for a few high-risk, highly specific medical exception scenarios. By automatically feeding back the clinical confirmation results after arbitration, the system dynamically adjusts the equipment stability profile, risk scenario weights, and population anomaly tolerance thresholds, enabling incremental parameter updates and continuous self-correction and adaptive evolution in real-world diagnostic and treatment environments.

8. A medical data quality inspection system based on the Internet of Things, characterized in that, include: Consistency Module: Based on the test metadata uploaded by IoT medical data testing equipment, a four-dimensional spatiotemporal event map is constructed and the device bias and network latency disturbances are integrated to calculate the spatiotemporal consistency index and identify underlying data misalignment; Physiological rationality module: Based on the test data identified through the consistency index and the continuously evolving medical physiology coupling rule base, the physiological rationality score is calculated by strongly coupling physiological rules and interpretability weights; Human-machine stability module: Based on multi-source state data of equipment, people and environment, it obtains human-machine stability factors that reflect the current inspection behavior by matching human-machine state snapshots with the context-aware rule base and integrating equipment quality control and operational errors. Relative credibility module: Based on the four-dimensional spatiotemporal event map and patient metadata of test records, the immune mechanism is simulated to obtain the population response vector, and the abnormality tolerance threshold is adjusted according to the epidemiological background to obtain the relative credibility index; Quality arbitration module: Based on spatiotemporal consistency index, physiological rationality score, human-machine stability factor, relative credibility index and multi-source feedback logs, a three-state quality arbitration result is obtained through multi-level arbitration grid decision-making and strategy optimization.

Images