A blood detection equipment performance evaluation method and system based on big data

Abstract

The present application relates to the technical field of device performance evaluation, in particular to a blood detection device performance evaluation method and system based on big data, comprising collecting real sample detection data of the detection device within a preset time period; determining an upper critical region and a lower critical region, and establishing a critical sample database; calculating a critical resolution index based on the critical sample database, wherein the critical resolution index is obtained by weighted calculation of a critical precision index, a boundary separation index and a distribution continuity index; grouping the critical sample database according to matrix types, calculating the critical resolution index, and comparing and analyzing the sensitivity difference; setting a sliding time window, obtaining the critical resolution index of different time periods, obtaining the time series of the analysis sensitivity, and performing trend analysis to output a performance degradation curve and trigger a performance degradation warning; and fusing the critical resolution index, the analysis sensitivity difference and the performance degradation curve to output a device performance evaluation report.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of device performance evaluation, in particular to a blood detection device performance evaluation method and system based on big data. BACKGROUND

[0002] Currently, when evaluating the performance of a detection device, the evaluation indicators specified in industry standards and guidelines are mainly used, and a small number of standard samples or quality control products are usually used for functional sensitivity testing.

[0003] The traditional evaluation method is based on a small number of artificially prepared quality control samples. The matrix characteristics of these samples are significantly different from real clinical samples, and they cannot fully reflect the actual detection performance of the device when processing real samples of different matrix types and different interference factors. Although a large amount of real detection data has been accumulated, there is a lack of effective data mining and analysis methods, and these data resources cannot be fully utilized for device performance evaluation.

[0004] In detection, samples with detection results near the boundary of the reference interval are in the transition zone between normal and abnormal. These critical samples are of great significance for early diagnosis of diseases and identification of slight abnormalities. The resolution capability of the device in the critical region directly affects the accuracy of clinical diagnosis, but the existing evaluation methods often mix critical samples with other samples for analysis, lack targeted evaluation indicators and calculation methods, and cannot accurately quantify the analytical sensitivity of the device in the key detection region.

[0005] From the perspective of data processing technology, the existing methods have the following deficiencies: first, there is a lack of effective critical sample screening and database construction algorithms, which cannot accurately extract key information from massive data; second, there is a lack of quantitative evaluation models of comprehensive multi-dimensional performance characteristics, and the existing indicators are too single to fully reflect the device performance; third, there is a lack of dynamic monitoring algorithms based on time window sliding and trend fitting, which cannot realize real-time tracking of performance changes; fourth, there is a lack of intelligent diagnosis methods for multi-factor correlation analysis and root cause tracing, which cannot provide accurate guidance for solving performance problems.

[0006] Therefore, it is urgent to construct a critical sample database, establish a critical resolution index calculation model, realize matrix-specific grouping analysis, use a sliding time window to monitor performance trends, and apply machine learning for predictive maintenance optimization data processing technology to realize intelligent evaluation of blood detection device performance, thereby improving detection quality and device management level.

[0007] Therefore, a blood detection device performance evaluation method and system based on big data are proposed. SUMMARY

[0008] The application aims to provide a blood detection equipment performance evaluation method and system based on big data, which collects a large amount of sample detection data in the real running process of the equipment, focuses on the samples whose detection results are located at the critical position of the reference interval, establishes a critical sample database, and calculates the critical resolution index based on the distribution characteristics of the critical samples, so as to realize the accurate evaluation of the analysis sensitivity of the equipment in the key detection area.

[0009] To achieve the above-mentioned purpose, the application provides the following technical scheme.

[0010] A blood detection equipment performance evaluation method based on big data, comprising:

[0011] Collecting real sample detection data of the detection equipment within a preset time period;

[0012] Determining the upper critical region and the lower critical region of the detection item, screening out sample data whose detection results fall into the upper critical region or the lower critical region from the detection data, and establishing a critical sample database;

[0013] Calculating the critical resolution index based on the critical sample database, and the critical resolution index is obtained by weighting the following sub-indicators: the coefficient of variation of the sample measurement value in the critical region as the critical precision index, the standardized distance between the critical sample mean and the boundary of the reference interval as the boundary separation degree index, and the probability density distribution of the fitted critical region sample and the evaluation of the smoothness as the distribution continuity index;

[0014] Grouping the critical sample database according to the matrix type, calculating the critical resolution index under each matrix type respectively, and identifying the analysis sensitivity difference under different matrix conditions through comparative analysis;

[0015] Setting a sliding time window, obtaining the critical resolution index in different time periods, obtaining the analysis sensitivity time sequence, and performing trend analysis to output the performance degradation curve and trigger the performance degradation warning;

[0016] Fusing the critical resolution index, the analysis sensitivity difference and the performance degradation curve, and outputting the equipment performance evaluation report.

[0017] Preferably, the detection data includes sample identification, detection result and sample matrix type information, and the detection data is cleaned to eliminate quality control samples and outliers;

[0018] The acquisition process of the upper critical region and the lower critical region includes: extracting normal sample data falling within the reference interval and abnormal sample data falling outside the reference interval in the historical detection data, respectively calculating the standard deviation of the normal sample data and the average distance of the abnormal sample data from the boundary of the reference interval; determining the value range of the preset width based on the standard deviation and the average distance, so that the critical region can cover the transition sample from normal to slight abnormal; obtaining the lower boundary of the lower critical region by subtracting the preset width from the lower limit of the reference interval, taking the lower limit of the reference interval as the upper boundary of the lower critical region, and forming the lower critical region; taking the upper limit of the reference interval as the lower boundary of the upper critical region, and obtaining the upper boundary of the upper critical region by adding the preset width to the upper limit of the reference interval, and forming the upper critical region.

[0019] Preferably, the acquisition process of the critical precision index and the boundary separation index includes: extracting the detection result values of all critical samples in the critical sample database, calculating the coefficient of variation of the detection result values, converting the coefficient of variation into the critical precision index by inverse transformation, and obtaining the critical precision index; calculating the absolute distance between the average value and the boundary of the reference interval, and normalizing the absolute distance by dividing by the critical region width to obtain the boundary separation index.

[0020] Preferably, the acquisition process of the distribution continuity index includes: dividing the critical region into multiple sub-intervals according to the detection result values; counting the number of samples in each sub-interval to form a sample frequency distribution; when the total number of critical samples is greater than a sample total number threshold, performing kernel density estimation and / or polynomial fitting on the sample frequency distribution to obtain a smoothed probability density curve; calculating a goodness-of-fit index between the probability density curve and the actual sample frequency distribution, calculating a smoothness coefficient of the curve, and evaluating the continuity and regularity of the distribution; and converting the evaluation result into a distribution continuity index.

[0021] Preferably, the acquisition process of the analysis sensitivity difference includes: extracting matrix type annotation information from the critical sample database, and classifying the critical samples according to the matrix type; obtaining the critical resolution index corresponding to each matrix group; calculating the difference between the critical resolution index of each matrix group and the critical resolution index of the normal matrix group to obtain the sensitivity deviation value of the interference matrix relative to the normal matrix; when the sensitivity deviation value exceeds a preset threshold, identifying the matrix type with a critical resolution index lower than that of the normal matrix group, and marking it as the analysis sensitivity difference of the blood detection equipment.

[0022] Preferably, the performance degradation early warning acquisition process comprises: setting the length and sliding step of a sliding time window, the time window length being a preset number of days; extracting critical sample data in each time window in turn from the current time point forward by the sliding step, calculating the critical resolution index of each time window respectively to form a time-ordered critical resolution index sequence; performing trend analysis on the critical resolution index sequence, fitting a performance trend curve, and setting a mutation detection mechanism by comparing the original data with the baseline value to identify sudden changes in performance; calculating the slope of the performance trend curve, and determining a continuous downward trend when the slope is negative and the absolute value exceeds a preset threshold; comparing the critical resolution index of the latest time window with that of the baseline time window, and triggering a performance degradation early warning signal when the decline exceeds a preset percentage and / or a mutation is detected.

[0023] Preferably, the device performance evaluation report acquisition process comprises: aggregating the overall critical resolution index and comparing it with the historical optimal value of the device to generate an overall analysis sensitivity rating; obtaining the critical resolution index of each matrix type to construct a matrix-specific sensitivity comparison matrix and identify the matrix type with a sensitivity short board; obtaining performance trend curve data to generate a time series trend chart and mark the performance degradation early warning time point; generating targeted improvement suggestions based on the identified sensitivity short board and performance degradation, including suggestions for instrument calibration, reagent replacement and / or maintenance.

[0024] A blood detection device performance evaluation system based on big data, comprising:

[0025] A data acquisition module for acquiring real sample detection data of a detection device within a preset time period;

[0026] A critical region division module for determining the upper and lower critical regions of a detection item and establishing a critical sample database;

[0027] A sensitivity analysis module for calculating a critical resolution index based on the critical sample database, the critical resolution index being calculated by weighting a critical precision index, a boundary separation index and a distribution continuity index;

[0028] A matrix-specific analysis module for grouping the critical sample database by matrix type, calculating the critical resolution index of each matrix, and identifying the analysis sensitivity difference under different matrix conditions;

[0029] A dynamic monitoring module for setting a sliding time window, obtaining critical resolution indexes in different time periods, obtaining an analysis sensitivity time sequence, performing trend analysis, outputting a performance degradation curve, and triggering a performance degradation early warning;

[0030] A report generation module is configured to fuse the critical resolution index, the analysis sensitivity difference and the performance degradation curve to output a device performance evaluation report.

[0031] Compared with the prior art, the present application has the following advantages:

[0032] 1. The present application introduces "critical region samples" for performance evaluation of blood detection equipment. The actual analysis sensitivity of the equipment under low difference and subtle change conditions is measured by long-term and large-scale collection and analysis of real clinical samples near the boundary of the reference interval of the detection results. By constructing a critical sample database, the equipment can obtain continuous monitoring capability in the interval where errors are most likely to occur, and the critical resolution index based on the variation coefficient, boundary deviation degree and probability density continuity is quantified. Compared with the traditional evaluation method based on full range samples, the identification sensitivity of the boundary region performance problem is improved, the discovery period of performance abnormalities is shortened from monthly to weekly, and the actual detection performance is accurately evaluated, so that the small performance fluctuation of the detection equipment can be identified in time, and the reliability of the detection results is improved.

[0033] 2. The present application introduces sample matrix type as a performance difference evaluation dimension, which can automatically screen critical samples under different matrices from real detection data, and calculate the corresponding critical resolution index of each matrix condition. The present application realizes the quantitative detection and presentation of sensitivity decline caused by matrix interference, which helps equipment manufacturers or laboratories to optimize reagents, improve methodology, and perfect pretreatment process for precise intervention measures, and finally improves the adaptability and overall detection accuracy of the equipment in diversified clinical samples.

[0034] 3. The present application realizes the prediction and early warning mechanism of performance degradation by setting a sliding time window to construct an analysis sensitivity time series, continuously tracking and trend analyzing the long-term detection performance change. The present application can automatically identify the performance continuous decline or sudden degradation by combining the slope of the critical resolution index over time, mutation detection and historical baseline comparison, and trigger an early warning signal in advance, thereby improving the equipment maintenance capability, reducing the error results, clinical risks and maintenance costs caused by performance degradation not found in time. At the same time, the performance degradation curve and trend chart generated by the system can provide an interpretable quantitative basis for equipment management, support precise maintenance, life assessment and reagent consumable replacement strategy optimization, and make the performance management of blood detection equipment more intelligent and sustainable. BRIEF DESCRIPTION OF DRAWINGS

[0035] Fig. 1 A blood detection equipment performance evaluation method flowchart based on big data is provided in the present application.

[0036] Fig. 2A blood detection equipment performance evaluation system structure diagram based on big data is provided for the present application.

[0037] Fig. 3 An equipment performance evaluation report output flowchart is provided for the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0039] Example 1

[0040] Please refer to Figs. 1-2 The present application provides a blood detection equipment performance evaluation method based on big data, which is applied to a blood detection equipment performance evaluation system based on big data. The system specifically includes the following modules: a data acquisition module, a critical region division module, a sensitivity analysis module, a matrix specificity analysis module, a dynamic monitoring module and a report generation module. The technical scheme is as follows:

[0041] Acquire real sample detection data of the detection equipment within a preset time period;

[0042] Determine the upper critical region and the lower critical region of the detection item; screen out sample data whose detection result falls into the upper critical region or the lower critical region from the detection data, and establish a critical sample database;

[0043] The principle of evaluating equipment performance by using critical samples is that, from a statistical point of view, the resolution capability of the equipment is most sensitive near the boundary of the reference interval. Small measurement errors here will cause changes in the judgment result (normal judgment as abnormal or abnormal judgment as normal). Therefore, the critical region is the region where equipment performance problems are most likely to be exposed. By continuously collecting a sufficient number of critical samples (usually more than 200), the law of large numbers ensures the stability and representativeness of statistical indicators. At the same time, through the symmetry evaluation of the upper critical region and the lower critical region, and the hierarchical analysis of multiple matrix types, the detection performance of the equipment under various clinical scenarios can be comprehensively reflected.

[0044] Further, the detection data includes sample identification, detection result and sample matrix type information, and the detection data is cleaned to eliminate quality control samples and outliers;

[0045] The obtaining process of the upper critical region and the lower critical region includes: extracting normal sample data falling within the reference interval and abnormal sample data falling outside the reference interval in the historical detection data, calculating the standard deviation of the normal sample data and the average distance of the abnormal sample data from the boundary of the reference interval respectively; determining the value range of the preset width based on the standard deviation and the average distance, so that the critical region can cover the transition sample from normal to slight abnormal; subtracting the preset width from the lower limit of the reference interval to obtain the lower boundary of the lower critical region, taking the lower limit of the reference interval as the upper boundary of the lower critical region to form the lower critical region; taking the upper limit of the reference interval as the lower boundary of the upper critical region, and adding the preset width to the upper limit of the reference interval to obtain the upper boundary of the upper critical region to form the upper critical region. The preset width is determined according to the standard deviation, and preferably, the preset width is between 1.5 times and 2.5 times of the standard deviation of the normal sample, and the specific value is determined according to the clinical sensitivity requirement of the detection item, and a smaller value (1.5 times) is taken for an item with high boundary discrimination requirement, and a larger value (2.5 times) is taken for an item with higher fault tolerance.

[0046] In the embodiment, the sensitive evaluation effect of the blood detection equipment performance is realized by constructing the critical sample database. By cleaning the real detection data and extracting the samples in the critical region, the key data close to the normal and abnormal boundary is focused, and the detection accuracy and stability of the equipment in the edge case are effectively reflected. The width of the critical region is determined by using the standard deviation and the deviation distance, so that it can cover the range from normal to slight abnormal, thereby improving the refinement and reliability of the performance evaluation, and providing a strong basis for equipment quality control and early warning.

[0047] The critical resolution index is calculated based on the critical sample database, and the critical resolution index is obtained by weighting the following sub-indicators: the coefficient of variation of the measurement value of the sample in the critical region as the critical precision index, the standardized distance between the mean value of the critical sample and the boundary of the reference interval as the boundary separation index, and the probability density distribution of the fitting critical region sample and the evaluation of the smoothness as the distribution continuity index; the calculation of the critical resolution index follows the following principles: the three sub-indicators (critical precision index, boundary separation index, and distribution continuity index) need to be normalized first, so that the numerical range is uniform between 0 and 1, and then weighted summation is performed according to the weight.

[0048] Further, the obtaining process of the critical precision index and the boundary separation index includes: extracting the detection result values of all critical samples in the critical sample database, calculating the coefficient of variation of the detection result values, converting the coefficient of variation into the critical precision index by using the reciprocal transformation to obtain the critical precision index; calculating the absolute distance between the mean value and the boundary of the reference interval, and normalizing the absolute distance by dividing by the width of the critical region to obtain the boundary separation index.

[0049] Further, the acquisition process of the distribution continuity index comprises: dividing the critical region into a plurality of subintervals according to the detection result value; counting the number of samples in each subinterval to form a sample frequency distribution; when the total number of critical samples is greater than the total sample number threshold, performing kernel density estimation and / or polynomial fitting on the sample frequency distribution to obtain a smooth probability density curve; calculating a goodness-of-fit index between the probability density curve and the actual sample frequency distribution, calculating the smoothness coefficient of the curve, and evaluating the continuity and regularity of the distribution; and converting the evaluation result into a distribution continuity index. The number of subintervals is determined according to the total number of critical samples n: when n < 200, the distribution continuity is not evaluated, and the index is directly set to a default value of 0.5; when 200 ≤ n < 1000, the number of subintervals is set to 10-20; when n ≥ 1000, the number of subintervals is set to 20-30; the Gaussian kernel density estimation method is preferably used, and the bandwidth parameter in the kernel density estimation is automatically determined by the empirical rule; the order of polynomial fitting is determined according to the sample size: when 200 ≤ n < 500, a cubic polynomial is used; and when n ≥ 500, a quartic polynomial is used. First, two methods are used for fitting respectively, the goodness-of-fit (R² value) is compared, and the method with a higher R² value is selected as the final result.

[0050] In the embodiment, the critical resolution index is calculated to evaluate the performance of the detection equipment in the critical region, three sub-indices of critical precision, boundary separation degree and distribution continuity are introduced, the detection stability, discrimination ability and data regularity of the equipment for edge samples are comprehensively measured, the critical sample data and the relationship with the reference interval boundary are statistically analyzed, the distribution smoothness is evaluated by using the probability density fitting, the evaluation result is more objective and quantifiable, and finally the critical resolution index can accurately reflect the sensitive monitoring ability of the equipment for normal and slightly abnormal transition intervals, and provide a reliable basis for equipment performance optimization.

[0051] The critical sample database is grouped according to the matrix type, the critical resolution index under each matrix type is calculated, and the differences in analysis sensitivity under different matrix conditions are identified by comparative analysis;

[0052] Further, the obtaining process of the analysis sensitivity difference comprises: extracting matrix type labeling information from the critical sample database, classifying the critical samples according to the matrix types; obtaining the critical resolution index corresponding to each matrix group; calculating the difference between the critical resolution index of each matrix group and the critical resolution index of the normal matrix group to obtain the sensitivity deviation value of the interference matrix relative to the normal matrix; when the sensitivity deviation value exceeds the preset threshold, identifying the matrix type with the lower critical resolution index than the normal matrix group and marking it as the analysis sensitivity difference of the blood detection equipment. The layered threshold judgment method is used to determine whether the sensitivity deviation value exceeds the preset threshold: the critical resolution index of the normal matrix group is I_normal, the critical resolution index of the interference matrix group is I_matrix, and the relative decline amplitude AI% = (I_normal-I_matrix) / I_normal*100%. When AI% ≥ 5%, it is marked as slight sensitivity difference; when AI% ≥ 10%, it is marked as moderate sensitivity difference; when AI% ≥ 15%, it is marked as significant sensitivity difference.

[0053] In the embodiment, the critical samples are analyzed by matrix type grouping, and the influence of different matrix conditions on the detection performance of the equipment is quantified. By comparing the difference in critical resolution index between each matrix group and the normal matrix, the interference matrix causing the sensitivity decline can be quickly identified, thereby providing a basis for equipment suitability evaluation. The quantitative analysis of the sensitivity deviation value visualizes and traces the performance difference of the equipment in a complex sample environment, which is conducive to optimizing the detection process, improving the reagent system or adjusting the algorithm parameters, thereby improving the detection reliability and stability of the equipment in diversified clinical samples.

[0054] Further, the identification of the analysis sensitivity difference also comprises an abnormal pattern identification step: performing distribution pattern analysis on the critical sample data corresponding to the matrix type marked as the sensitivity difference, identifying three typical abnormal patterns by calculating the skewness, kurtosis and multi-peak indexes of the samples: systematic shift pattern (the whole sample is biased to one side of the reference interval, and the absolute value of the skewness is greater than 0.5), dispersion increase pattern (kurtosis <-0.5, indicating flat distribution), and bimodal distribution pattern (judged as bimodal when there are more than two significant peaks according to the number of local maximum points of the frequency distribution). The identification of these patterns is based on statistical distribution characteristics and can objectively reflect different physical interference mechanisms.

[0055] Meanwhile, through analysis of a large amount of actual detection data, it is found that the hemolysis, lipemia and jaundice matrix interference factors will cause the critical resolution index to decrease systematically. The three typical abnormal patterns (systematic shift pattern, dispersion increase pattern and bimodal distribution pattern) correspond to different physical interference mechanisms: the systematic shift is usually caused by optical interference, the dispersion increase reflects unstable reaction kinetics, and the bimodal distribution indicates batch difference or interference threshold effect.

[0056] Setting a sliding time window, obtaining critical resolution indexes of different time periods, obtaining an analysis sensitivity time sequence, and performing trend analysis to output a performance degradation curve and trigger a performance degradation warning;

[0057] Further, the performance degradation warning acquisition process includes: setting the length and sliding step of the sliding time window, the time window length is a preset number of days; from the current time point, the critical sample data in each time window is extracted in turn according to the sliding step, the critical resolution index of each time window is calculated respectively, and a time-ordered critical resolution index sequence is formed; trend analysis is performed on the critical resolution index sequence, a performance trend curve is fitted, and a sudden change detection mechanism is set by comparing the original data with the baseline value to identify sudden changes in performance (specifically, a linear regression method is used to fit the performance trend curve, the determination coefficient R² is calculated, when R²≥0.7, a linear model is used, when R²<0.7, a 3-day moving average is smoothed and then linear regression is performed; the sudden change detection is performed by comparing the original critical resolution index sequence with the baseline value (the average value of the historical optimal period) point by point, when the critical resolution index of a certain time point decreases by more than 20% relative to the baseline, it is determined as a sudden change); the slope of the performance trend curve is calculated, when the slope is negative and the absolute value exceeds a preset threshold (when the slope is negative and the absolute value exceeds the preset threshold, it is determined as a continuous downward trend), it is determined as a continuous downward trend; the slope threshold is determined according to the time window length W and the allowed decline amplitude: the maximum monthly decline amplitude allowed is 7% (the median of 5%-10%), when the actual slope is less than the slope threshold, the performance degradation warning is triggered, the critical resolution index of the latest time window is compared with the critical resolution index of the baseline time window, when the decline amplitude exceeds a preset percentage or a sudden change is detected, a performance degradation warning signal is triggered.

[0058] In this embodiment, the critical resolution index time sequence is constructed by a sliding time window, which realizes dynamic tracking of the change of device performance over time. Through trend fitting and slope analysis, slow performance degradation or sudden decline can be identified in time, a performance degradation curve is formed and a warning is given. When the critical resolution index continuously declines or the decline amplitude exceeds a set threshold, the system can automatically trigger an alarm, which helps to discover potential problems of device aging, calibration deviation or environmental impact in advance, so as to support timely maintenance and calibration, and improve the long-term stability and quality assurance capability of blood detection equipment.

[0059] Further, the performance degradation warning further includes a predictive maintenance scheduling optimization step:

[0060] A device historical maintenance record database is established to record the type, time point, critical resolution index change before and after maintenance, maintenance cost and device downtime of each maintenance;

[0061] Based on the performance trend curve and historical maintenance records, a machine learning algorithm is used to train a performance degradation prediction model, inputting the current critical resolution index sequence, the cumulative use time of the device, and the sample detection amount, to predict the decline trend of the critical resolution index in the future preset time period;

[0062] A performance acceptable lower threshold is set, and the predicted time to reach the threshold is calculated as the starting time point of the maintenance window period according to the predicted performance degradation rate;

[0063] A maintenance scheduling optimization model is constructed: the predicted maintenance window period, the detection task priority of the device, and the availability of maintenance personnel and spare parts are used as constraint conditions; the optimization goal is to minimize the performance risk of the device; and the optimal maintenance time point and maintenance type are obtained by solving;

[0064] When performance degradation warning is triggered, a predictive maintenance suggestion is generated synchronously;

[0065] The performance recovery of the device after maintenance is tracked and evaluated, and the actual maintenance effect is fed back to the prediction model to realize continuous optimization and adaptive adjustment of the model.

[0066] Specifically, a device historical maintenance record database is established to record detailed information of each maintenance, including maintenance type (preventive, corrective, emergency maintenance), maintenance time point, maintenance content, change in critical resolution index before and after maintenance, maintenance cost, and device downtime.

[0067] A machine learning algorithm is used to train a performance degradation prediction model, and the input features include the current critical resolution index sequence, the cumulative use time of the device, the sample detection amount, the time since the last maintenance, and the historical performance degradation pattern. The output is a sequence of predicted critical resolution index values in the future preset time period. Methods such as ARIMA model, exponential smoothing, multiple regression, and neural network can be used, and LSTM network is preferably used to process time series data. A performance acceptable lower threshold is set, and the time point when the critical resolution index first falls below the threshold is found as the starting time of the maintenance window period according to the performance curve output by the prediction model. A maintenance scheduling optimization model is constructed, considering constraint conditions and optimization goals. The constraint conditions include the predicted maintenance window period, the detection task priority of the device, and the availability of maintenance resources. The optimization goals include minimizing performance risk, reducing downtime, and reducing maintenance cost. Linear programming, genetic algorithm, or weighted scoring method can be used to solve the optimal maintenance time point and maintenance type.

[0068] The predictive maintenance suggestion is generated when the performance degradation early warning is triggered, including the maintenance time suggestion, maintenance item priority ranking, predicted performance improvement range, maintenance cost budget and maintenance preparation; after the maintenance is completed, the actual effect is tracked and evaluated, the new maintenance record is fed back to the prediction model, the model parameters are updated by using the incremental learning method, the model is continuously optimized and adaptively adjusted, and a closed loop mechanism of maintenance effect tracking and model optimization is formed.

[0069] By establishing a predictive maintenance scheduling optimization mechanism, passive responsive maintenance is upgraded to active predictive maintenance, which can intervene in advance before the device performance has been seriously degraded, avoiding the decline in detection result reliability and patient safety risk caused by insufficient performance. This method combines performance prediction and resource scheduling optimization to reasonably arrange the maintenance time window under the premise of ensuring detection quality, reduces the impact on normal detection process, and improves device utilization. At the same time, by quantifying the maintenance benefit and cost, the scientific and fine management of maintenance decision is realized, effectively avoiding the two extreme situations of over-maintenance and insufficient maintenance, prolonging the service life of the device, reducing the overall operating cost of the laboratory, and improving the quality management and operating efficiency level of the medical laboratory.

[0070] The device performance evaluation report is output by fusing the critical resolution index, the analysis sensitivity difference and the performance degradation curve.

[0071] Further, the device performance evaluation report acquisition process includes: summarizing the overall critical resolution index and comparing it with the historical optimal value to generate the overall analysis sensitivity rating (excellent if the current value is greater than or equal to 95% of the historical optimal value, good if 90%-95%, qualified if 80%-90%, needs improvement if 70%-80%, and unqualified if less than 70%); obtaining the critical resolution index of each matrix type, constructing a matrix of matrix-specific sensitivity comparison, and identifying the matrix type with sensitivity short board; obtaining performance trend curve data, generating a time series trend graph, and marking the performance degradation warning time point; based on the identified sensitivity short board and performance degradation, generating targeted improvement suggestions, including suggestions for instrument calibration, reagent replacement and / or maintenance.

[0072] In this embodiment, by fusing the critical resolution index, the matrix sensitivity difference and the performance degradation trend, comprehensive evaluation of the performance of the blood detection device is realized. The generated performance evaluation report not only gives the overall sensitivity rating, but also locates the weak link of sensitivity through the matrix-specific matrix; and identifies the performance degradation risk combined with the time series trend graph, with clear visual performance. At the same time, the system can automatically give optimization suggestions such as device calibration, reagent replacement or maintenance according to the detection short board and degradation, helping the laboratory to intervene and make decisions in time, and ensuring the long-term stable operation of the device and the reliability of the detection quality.

[0073] Further, a multi-device transverse comparison analysis step is further included:

[0074] A device performance benchmark database is established, and critical resolution index, analysis sensitivity difference and performance degradation curve data of multiple devices under the same detection item are collected;

[0075] Performance consistency evaluation is performed on devices of the same type, the dispersion degree of the critical resolution index of each device is calculated, and devices with abnormal performance deviation are identified;

[0076] Transverse performance comparison is performed on devices of different types, the critical resolution index of each device is standardized and sorted to generate a device performance benchmark curve;

[0077] Based on the device performance benchmark curve, the performance deviation and performance percentile ranking of each evaluated device are calculated, and a transverse comparison analysis module is added to the device performance evaluation report to mark the performance positioning of the device among the same type of devices;

[0078] When it is detected that the critical resolution index of a device continuously falls below a preset percentage of the average value of devices of the same type, an abnormal device identification is triggered, and special inspection is recommended.

[0079] Specifically, a device performance benchmark database is established, and performance data of multiple devices under the same detection item are collected. The database collection range includes devices of the same type in the same laboratory, the same medical institution or the same region, and ensures that the detection items and reference interval standards are consistent. The critical resolution index, critical resolution index of each matrix type, and performance trend curve data are collected from each device and stored after standardization.

[0080] Performance consistency evaluation is performed on devices of the same type, the critical resolution index of devices of the same type is extracted, the average value, standard deviation and coefficient of variation are calculated, the smaller the coefficient of variation, the better the consistency, and statistical methods are used to identify abnormal devices. Devices with critical resolution index lower than the average value minus twice the standard deviation are marked as low-performance abnormalities and need to be investigated for device failure, improper operation or environmental factors, etc.

[0081] Transverse performance comparison is performed on devices of different types, the critical resolution index of devices of different types is normalized to achieve comparability by converting to a zero to one interval; the device performance benchmark curve is generated by sorting according to the standardized value, and the performance ranking of devices of different types is intuitively displayed to provide a basis for device selection and procurement.

[0082] The performance deviation and performance percentile ranking of each evaluated device are calculated, the deviation reflects the difference between the device and the average level of the same type, and the percentile ranking indicates the position of the device among the same type of devices. A transverse comparison analysis module is added to the evaluation report, including device performance positioning overview, comparison chart of devices of the same type, cross-type benchmark curve and performance gap analysis.

[0083] In this embodiment, by establishing a multi-device performance benchmark database and conducting a horizontal comparative analysis, the performance deviation of a single device relative to the group can be identified, and performance problems caused by individual differences, installation environment or use habits of the device can be effectively found. This method expands the single-point evaluation to group reference evaluation, improves the accuracy of performance anomaly detection, avoids misjudging group performance decline as single device failure, and provides quantitative performance comparison basis for laboratory device configuration optimization and device procurement decision, promoting the standardization and normalization of device performance management.

[0084] The present application realizes comprehensive, dynamic and intelligent evaluation of blood detection device performance through multi-dimensional technologies such as constructing critical sample database, calculating critical resolution index, analyzing matrix sensitivity difference, trend monitoring performance degradation and predictive maintenance scheduling. This method can accurately identify the detection capability of the device in the normal and abnormal boundary region, and quantify the influence of different matrix conditions on sensitivity. At the same time, through sliding time window and trend analysis, performance degradation warning is realized, and machine learning is combined to predict maintenance time, improving the stability of device operation and the scientificity of maintenance management. In addition, the horizontal multi-device comparison mechanism further enhances the performance anomaly positioning ability, and provides a reliable basis for device management and configuration optimization. The overall scheme upgrades the traditional quality management to an intelligent quality control system integrating monitoring, early warning, tracing and optimization, effectively guarantees the accuracy of detection results and the safety of clinical use, improves the laboratory operation efficiency and the level of device whole life cycle management, and the specific process is referred to Fig. 3 .

[0085] Embodiment two:

[0086] The present application provides a blood detection device performance evaluation method based on big data, which collects a large amount of sample detection data in the real running process of the device, focuses on the samples whose detection results are located at the critical position of the reference interval, establishes a critical sample database, and calculates the critical resolution index based on the distribution characteristics of the critical samples, so as to realize the analysis of sensitivity of the device in the key detection area.

[0087] Firstly, the real sample detection data of the detection device in a preset time period needs to be collected. The preset time period is determined according to the sample quantity of the detection project, and is preferably one month to three months continuously, so as to ensure that a sufficient number of samples are collected for statistical analysis. For the conventional detection project with large sample quantity, a shorter time period can be selected; for the special detection project with small sample quantity, the collection time needs to be prolonged to accumulate sufficient sample data.

[0088] The collected detection data includes but is not limited to the following information:

[0089] Sample ID: used to uniquely identify each detection sample, which can be a sample number or other unique identifier;

[0090] Detection result: the quantitative or qualitative detection result value of the sample under a specific detection item;

[0091] Sample matrix type information: including serum, plasma, whole blood matrix type, and whether there are hemolysis, lipemia, jaundice interference factors.

[0092] After data collection is completed, the detection data needs to be cleaned. The purpose of data cleaning is to eliminate atypical samples that may affect the accuracy of performance evaluation. Specifically, the quality control sample data needs to be eliminated because the quality control sample is a standard sample artificially prepared, and its detection result cannot represent the distribution characteristics of the real clinical sample. At the same time, outliers need to be identified and eliminated. The identification of outliers can use statistical methods, such as based on the mean and standard deviation of the detection result distribution, data points deviating from the mean by more than a certain multiple of the standard deviation are marked as outliers. Preferably, the box plot method can be used to identify data points outside a certain range of the upper and lower boundaries of the box as outliers. In addition, obvious error data caused by equipment failure or operation error also needs to be eliminated.

[0093] The critical region refers to a specific numerical range near the boundary of the reference interval. The samples in this region contain both normal samples and slightly abnormal samples, and are the transition zone from normal to abnormal. The resolution capability of the device for samples in the critical region is directly related to the early diagnosis of diseases and the timely discovery of slight abnormalities.

[0094] The specific implementation process is as follows:

[0095] First, extract the normal sample data falling within the reference interval and the abnormal sample data falling outside the reference interval from the historical detection data. The reference interval is determined according to the reference value range provided by the clinical medical standard or the device manufacturer, and is usually represented by the lower limit value and the upper limit value of a certain detection item;

[0096] Second, statistical analysis is performed on the normal sample data and the abnormal sample data respectively. For normal sample data, the standard deviation of the detection result is calculated, and the preset width is determined based on the standard deviation. Preferably, the preset width is 1.5 to 2.5 times the standard deviation of the normal sample, and the standard configuration is 2.0 times the standard deviation. This value is based on statistical principles: the natural fluctuation of normal samples at the edge of the reference interval is usually within 1-2 times the standard deviation, and the distance of early abnormal samples deviating from the reference interval boundary is usually within 0.5-1.5 times the standard deviation. Therefore, the critical region width of 2 times the standard deviation can effectively cover the transition region of these two types of samples.

[0097] The lower critical region is determined in the following manner: the lower boundary of the lower critical region is obtained by subtracting the preset width from the lower limit of the reference interval, and the upper limit of the reference interval is taken as the upper boundary of the lower critical region, thereby determining the numerical range of the lower critical region; the upper critical region is determined in the following manner: the upper limit of the reference interval is taken as the lower boundary of the upper critical region, and the upper boundary of the upper critical region is obtained by adding the preset width to the upper limit of the reference interval, thereby determining the numerical range of the upper critical region.

[0098] After the upper critical region and the lower critical region are determined, sample data whose detection results fall into the upper critical region or the lower critical region is screened from the cleaned detection data, and the screening process is realized through numerical comparison: for each detection data record, it is judged whether the numerical value of the detection result falls within the range of the lower critical region or the range of the upper critical region, and if the condition is met, the sample data is included in the critical sample database; the critical sample database is a structured data set for storing and managing sample information of all critical positions, and each record in the database contains complete information of the sample, including sample identification, detection result, matrix type, detection time, reagent batch, and equipment information; preferably, the database adopts a relational database structure, which is convenient for subsequent query, grouping, and statistical analysis operations.

[0099] The critical resolution index is used to quantify the resolution capability of the device for the critical region sample, and the index is obtained by the weighted combination of three sub-indices; the critical precision index reflects the repeatability and stability of the measurement results of the device in the critical region; the calculation process of the critical precision index is as follows: first, the detection result values of all critical samples are extracted from the critical sample database; second, the average value μ and the standard deviation σ of the value sequence are calculated, and the coefficient of variation CV = σ / μ; then, the coefficient of variation is converted into a precision index in the range of 0-1 by using the reciprocal transformation formula I_precision = 1 / (1+CV), which ensures that when CV = 0, the precision index is 1, and when CV increases, the index tends to 0; the boundary separation degree index reflects the distance relationship between the overall level of the critical sample and the boundary of the reference interval. The calculation process of the boundary separation degree index is as follows: using the average value μ of the critical sample detection results; determining the corresponding reference interval boundary value (for the lower critical region, the lower limit of the reference interval, for the upper critical region, the upper limit of the reference interval); calculating the absolute distance d between the average value and the reference interval boundary; the distance is normalized by dividing by the critical region width to obtain the boundary separation degree index; the distribution continuity index reflects the regularity and smoothness of the distribution of the critical region sample, and the calculation process of the distribution continuity index is as follows: when the total number of critical samples n ≥ 200, the distribution is analyzed, and when n < 200, the index is defaulted to 0.5; the critical region is evenly divided into √n sub-intervals; the number of samples in each sub-interval is counted; the sample frequency distribution is estimated by kernel density estimation (using Gaussian kernel, bandwidth h is automatically calculated by Scott rule) or polynomial fitting (3 or 4 times), and the fitting goodness R² of the two methods is compared, and the method with higher R² is selected; according to the R² value, the distribution continuity index is converted: when R² > 0.85, the index is 0.9, when 0.7 ≤ R² ≤ 0.85, the index is 0.7, and when R² < 0.7, the index is 0.5.

[0100] The weights of the three sub-indices can be adjusted according to the clinical needs of different detection items, and in a preferred embodiment, the weights of the critical precision index, the boundary separation degree index and the distribution continuity index are 0.4, 0.3 and 0.3 respectively; for detection items with high boundary judgment sensitivity, the weight of the boundary separation degree index can be increased.

[0101] The critical precision index is used to evaluate the measurement precision of the device in the critical region, and the precision refers to the consistency between the measurement results when the same sample is measured multiple times under the same conditions; the calculation process of the critical precision index is as follows:

[0102] First, the detection result values of all critical samples are extracted from the critical sample database to form a value sequence, which includes all sample detection results falling into the upper critical region and the lower critical region;

[0103] Secondly, the average value and the standard deviation of the numerical sequence are calculated, the average value represents the central tendency of the critical sample detection result, and the standard deviation represents the dispersion degree of the detection result around the average value, the smaller the standard deviation, the more concentrated the detection result of the critical sample, and the higher the measurement precision of the equipment; the coefficient of variation is the standard deviation divided by the average value.

[0104] The critical precision index is represented by the coefficient of variation, which is the ratio of the standard deviation to the average value, eliminating the influence of dimension, so that the precision of different detection items is comparable; the smaller the coefficient of variation, the smaller the relative variation, and the higher the precision; preferably, the coefficient of variation can be standardized or inverted, so that the index value is positively correlated with the precision, facilitating subsequent comprehensive scoring.

[0105] The boundary separation index is used to evaluate the distance relationship between the overall level of the critical sample and the boundary of the reference interval, the larger the boundary separation, the farther the detection result of the critical sample from the boundary of the reference interval, and the stronger the ability of the equipment to distinguish between normal and abnormal samples; the calculation process of the boundary separation index is as follows:

[0106] Firstly, the average value of the critical sample detection result calculated by the foregoing is used; the average value represents the central level of all critical samples;

[0107] Secondly, the corresponding reference interval boundary value is determined, for the samples in the lower critical region, the reference interval boundary is the lower limit value of the reference interval; for the samples in the upper critical region, the reference interval boundary is the upper limit value of the reference interval;

[0108] Then, the absolute distance between the average value and the reference interval boundary is calculated, the absolute distance represents the degree of deviation of the central level of the critical sample from the reference interval boundary.

[0109] In order to eliminate the influence of the dimension of different detection items, the distance is preferably standardized, and the standardization can adopt various ways, for example, the distance is divided by the width of the critical region to obtain the relative distance; or the distance is divided by the standard deviation of the normal sample to obtain the standardized distance; the standardized distance value can be compared between different detection items.

[0110] The distribution continuity index is used to evaluate the regularity and smoothness of the distribution of the critical region samples, poor distribution continuity may indicate that the equipment has detection instability, systematic deviation or quality control problems in the critical region, and the calculation process of the distribution continuity index is as follows:

[0111] Firstly, the critical region is evenly divided into multiple sub-intervals according to the detection result values, and the number of sub-intervals needs to be determined according to the total number of critical samples. The larger the sample size, the more sub-intervals can be divided. Preferably, the number of sub-intervals can be set to the integer part of the square root of the total number of critical samples, or fixed to ten to fifty sub-intervals, each sub-interval having an equal numerical span.

[0112] Secondly, the number of samples in each sub-interval is counted to form a sample frequency distribution, which describes the distribution of critical samples in different numerical intervals in the form of a histogram.

[0113] Thirdly, the sample frequency distribution is smoothed to obtain a smoothed probability density curve. The purpose of smoothing is to eliminate the frequency fluctuations caused by sample randomness and extract the overall trend of sample distribution. The smoothing can be performed using a kernel density estimation method or a polynomial fitting method. Kernel density estimation is a non-parametric density estimation method that estimates the overall probability density distribution by setting a kernel function at each sample point and superimposing all kernel functions. The kernel function can be Gaussian, triangular, or rectangular, and the Gaussian kernel is preferred. The polynomial fitting method uses a polynomial function to fit the sample frequency distribution. The order of the polynomial determines the complexity of the fitted curve. The higher the order, the more the fitted curve can reflect the detailed characteristics of the distribution, but it is also prone to overfitting. Preferably, a cubic to quintic polynomial is used for fitting.

[0114] Fourthly, the quality of the smoothed curve is evaluated, and a distribution continuity index is calculated. The evaluation includes two aspects:

[0115] First, the goodness of fit between the probability density curve and the actual sample frequency distribution is calculated. The goodness of fit reflects the representativeness of the smoothed curve for the actual distribution. Commonly used goodness of fit indicators include the coefficient of determination and the root mean square error. The higher the goodness of fit, the more accurately the smoothed curve can describe the actual sample distribution.

[0116] Secondly, the smoothness coefficient of the curve is calculated. The smoothness can be evaluated by the second derivative of the curve. The smaller the change in the second derivative, the smoother the curve. The smoothness can also be evaluated by calculating the slope change between adjacent points of the curve. High smoothness indicates that the sample distribution is continuous and uniform, without abnormal mutations or discontinuities.

[0117] Fifthly, the goodness of fit and the smoothness coefficient are converted into a distribution continuity index. The goodness of fit and the smoothness can be combined by weighted averaging, or the goodness of fit and the smoothness can be converted into standardized scores according to the scoring rules, and then combined. The higher the value of the distribution continuity index, the better the continuity and regularity of the critical region sample distribution.

[0118] After obtaining the critical precision index, the boundary separation index and the distribution continuity index, the critical resolution index is calculated by weighted summation; the higher the value of the critical resolution index, the stronger the resolution capability of the equipment in the critical region, and the better the performance.

[0119] The matrix condition of the blood sample has an important influence on the test result, and different matrix types may cause test interference and affect the analysis sensitivity of the equipment; common matrix types include serum, plasma and whole blood, and common matrix interference factors include hemolysis, lipemia and jaundice; the performance of the equipment under different matrix conditions may differ, so the critical samples need to be grouped according to the matrix type, and the performance under each matrix condition is evaluated respectively.

[0120] The sample matrix type labeling information is extracted from the critical sample database, and the matrix type information has been recorded in the database during the data acquisition stage; according to the matrix type information, the critical samples are classified, and the classification can be performed according to a single dimension, such as classification according to serum, plasma and whole blood only.

[0121] For each matrix group, all critical sample data in the group are extracted, and the critical precision index, the boundary separation index and the distribution continuity index of the matrix group are calculated respectively according to the foregoing method, and then the critical resolution index of the matrix group is calculated.

[0122] After obtaining the critical resolution index of each matrix group, comparative analysis is needed to identify the analysis sensitivity difference under different matrix conditions; the normal matrix group is taken as the reference standard for comparative analysis, and the normal matrix group refers to the standard sample matrix without interference factors, such as serum samples without hemolysis, lipemia and jaundice; the critical resolution index of each matrix group is calculated by difference with the critical resolution index of the normal matrix group, and the sensitivity deviation value is obtained; the positive sensitivity deviation value indicates that the analysis sensitivity of the matrix group is higher than that of the normal matrix, and the negative deviation value indicates that it is lower than that of the normal matrix; when the sensitivity deviation value exceeds the preset threshold, the matrix type with lower critical resolution index than the normal matrix group is identified, and these matrix types are marked as the analysis sensitivity difference of the equipment; preferably, a threshold value is set, and only when the decline amplitude of the critical resolution index exceeds the threshold value, it is marked as there is a significant sensitivity difference; through matrix type grouping analysis, it can be identified that the equipment has performance short board under which matrix conditions, and provides basis for targeted quality improvement, for example, if it is found that the critical resolution index of the equipment under the lipemia sample is significantly reduced, it indicates that the anti-interference ability of the equipment to lipid interference is insufficient, and the reagent formula needs to be optimized or the detection method needs to be adjusted.

[0123] In order to dynamically monitor the performance trend of the equipment over time, a sliding time window is adopted; the sliding time window is a time series analysis method, by setting a fixed length of time window, sliding on the time axis, obtaining data in different time periods for analysis; the setting of the length of the time window needs to consider the amount of data and timeliness; if the time window is too short, the sample size in the window is insufficient, and the statistical result is unstable; if the time window is too long, the performance change cannot be reflected in time; the sliding step determines the time interval between the adjacent two time windows, the smaller the sliding step, the higher the time resolution, but the larger the calculation amount.

[0124] From the current time point, the critical sample data in each time window is extracted in turn according to the set sliding step; for each time window, the following operations are performed:

[0125] First, all critical samples whose detection time falls within the time window range are screened out from the critical sample database;

[0126] Secondly, the critical precision index, boundary separation index and distribution continuity index of the critical samples in the time window are calculated, and then the critical resolution index of the time window is calculated.

[0127] The trend analysis is performed on the critical resolution index sequence, and the performance trend curve is fitted, the purpose of the trend analysis is to extract the long-term change trend of the equipment performance from the time series data containing noise; the fitting of the trend curve can adopt various methods, the linear regression method is suitable for the case that the performance changes linearly, a straight line is fitted by the least square method to describe the performance trend, the polynomial regression is suitable for the case that the performance changes more complex, a quadratic or cubic polynomial can be used for fitting, the moving average method smoothes the noise by calculating the average value of adjacent points in the sequence to extract the trend.

[0128] In addition to long-term trend analysis, a mutation detection mechanism needs to be set up to identify sudden changes in performance, which can be caused by factors such as equipment failure, reagent batch replacement, maintenance operation, etc.; mutation detection can use statistical process control methods, specifically, calculate the mean and standard deviation of the critical resolution index sequence, set control limits; when the critical resolution index at a certain time point deviates from the mean by a certain multiple of the standard deviation, it is determined that a mutation has occurred; when a mutation is detected, the system should record the time point and the magnitude of the mutation, and mark it on the performance trend curve; calculate the slope of the performance trend curve, the slope reflects the speed of performance change over time, when the slope is negative, it means that the performance is in a downward trend; the larger the absolute value of the slope, the faster the decline; set a preset threshold, when the slope is negative and the absolute value exceeds the preset threshold, it is determined to be a continuous downward trend; the setting of the slope threshold needs to be determined according to the requirements of the detection project and the normal fluctuation range of the equipment. Preferably, the slope threshold can be set so that under the threshold condition, the device performance decreases by 5% to 10% of the initial performance within a month.

[0129] The triggering of performance degradation warning is based on two conditions:

[0130] The first condition is to compare the critical resolution index of the recent time window with the critical resolution index of the baseline time window, the baseline time window refers to the time window when the device performance is good, which can be selected as the initial period after the installation and commissioning of the device, or the period with the best historical performance, calculate the decline amplitude of the critical resolution index of the recent time window relative to the baseline time window; set a decline amplitude threshold, when the decline amplitude exceeds the threshold, trigger the performance degradation warning;

[0131] The second condition is to detect the aforementioned performance mutation, when the critical resolution index is detected to have a mutation and the mutation direction is performance decline, immediately trigger the performance degradation warning;

[0132] When any of the above conditions are met, the system generates a performance degradation warning signal, notifying the laboratory management personnel that the device performance has degraded and needs to be checked and maintained.

[0133] The device performance evaluation report first summarizes the overall critical resolution index of the device, which is a comprehensive performance indicator calculated based on all critical samples in the entire evaluation period; compare the overall critical resolution index with the historical optimal value of the device, which can be extracted from the historical performance database of the device, representing the performance level of the device in the best state; through comparison, the gap between the current performance of the device and the optimal state can be evaluated.

[0134] An overall analysis sensitivity rating is generated based on the comparison results, and the rating can adopt excellent, good, qualified, needs to be improved, and unqualified grading standards; the rating standards can be determined according to the percentage of the critical resolution index relative to the historical optimal value.

[0135] The critical resolution index of each matrix type is obtained, a matrix-specific sensitivity comparison matrix is constructed, and the comparison matrix is presented in the form of a table, with rows representing different matrix types and columns representing the corresponding critical resolution index, deviation value relative to the normal matrix, and rating information; through the comparison matrix, the matrix type with a sensitivity short board can be directly identified, and the sensitivity short board refers to the matrix type with a critical resolution index significantly lower than the normal matrix or lower than the qualified standard; in the comparison matrix, the matrix type with a sensitivity short board is marked with a special identifier (such as red highlighting, warning symbols), reminding the management personnel to pay attention.

[0136] Performance trend curve data is obtained, and a time series trend graph is generated, with time as the horizontal axis and the critical resolution index as the vertical axis, and the critical resolution index curve over time is plotted.

[0137] Important time nodes are marked on the trend graph, including: the time point when the performance degradation warning is triggered, which is highlighted with a warning mark; the time point of reagent batch replacement and equipment maintenance operation, which is marked with different symbols; the time point when the performance mutation is detected.

[0138] Based on the identified sensitivity short board and performance degradation, targeted improvement suggestions are generated; the generation of improvement suggestions is based on expert rule library and historical treatment experience.

[0139] For different performance problems, corresponding measures are suggested:

[0140] When the overall critical resolution index decreases but each matrix type performs consistently, instrument calibration is recommended, which can correct systematic deviation and restore the accuracy of the device;

[0141] When a performance mutation is detected and is related to the reagent batch replacement time, reagent quality inspection or reagent replacement is recommended;

[0142] When the performance shows a continuous slow downward trend, equipment maintenance and maintenance are recommended, including cleaning optical components, replacing consumables, and checking mechanical components;

[0143] When there is a sensitivity short board for a specific matrix type, special optimization for that matrix type is recommended, which may include adjusting the detection method parameters, replacing the reagent formula suitable for that matrix.

[0144] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for evaluating the performance of blood testing equipment based on big data, characterized in that, The method comprises the following steps: Collecting real sample detection data of the detection equipment within a preset time period; Determining the upper critical region and the lower critical region of the detection item; Screening sample data with detection results falling into the upper critical region or the lower critical region from the detection data to establish a critical sample database; Calculating a critical resolution index based on the critical sample database, and the critical resolution index is obtained by weighting the following sub-indicators: the coefficient of variation of the measurement values of the samples in the critical region as the critical precision indicator, the standardized distance between the mean value of the critical samples and the boundary of the reference interval as the boundary separation degree indicator, and the probability density distribution of the fitting critical region samples and the evaluation of the smoothness as the distribution continuity indicator; Grouping the critical sample database according to the matrix types, calculating the critical resolution index under each matrix type respectively, and identifying the analysis sensitivity difference under different matrix conditions through comparative analysis; Setting a sliding time window to obtain the critical resolution index in different time periods, obtaining the time sequence of the analysis sensitivity, and performing trend analysis to output the performance degradation curve and trigger the performance degradation warning; Fusing the critical resolution index, the analysis sensitivity difference and the performance degradation curve to output the equipment performance evaluation report.

2. The blood detection equipment performance evaluation method based on big data according to claim 1, wherein: The detection data includes sample identification, detection result and sample matrix type information, and the detection data is cleaned to remove quality control samples and outliers; The acquisition process of the upper critical region and the lower critical region includes: extracting normal sample data falling into the reference interval and abnormal sample data falling outside the reference interval from the historical detection data, calculating the standard deviation of the normal sample data and the average distance of the abnormal sample data from the boundary of the reference interval respectively; determining the value range of the preset width based on the standard deviation and the average distance, so that the critical region can cover the transition samples from normal to slight abnormal; the lower boundary of the lower critical region is obtained by subtracting the preset width from the lower limit of the reference interval, the upper limit of the reference interval is taken as the upper boundary of the lower critical region, and the lower critical region is formed; the upper limit of the reference interval is taken as the lower boundary of the upper critical region, and the upper boundary of the upper critical region is obtained by adding the preset width to the upper limit of the reference interval, and the upper critical region is formed.

3. The blood detection equipment performance evaluation method based on big data according to claim 1, wherein: The acquisition process of the critical precision indicator and the boundary separation degree indicator includes: extracting the detection result values of all critical samples in the critical sample database, calculating the coefficient of variation of the detection result values, converting the coefficient of variation into the critical precision indicator by inverse transformation, and obtaining the critical precision indicator; calculating the absolute distance between the mean value and the boundary of the reference interval, and normalizing the absolute distance by dividing by the width of the critical region to obtain the boundary separation degree indicator.

4. The blood detection equipment performance evaluation method based on big data according to claim 1, wherein: The acquisition process of the distribution continuity index includes: dividing the critical region into multiple subintervals according to the detection result value; counting the number of samples in each subinterval to form a sample frequency distribution; when the total number of critical samples is greater than the total sample threshold, performing kernel density estimation and / or polynomial fitting on the sample frequency distribution to obtain a smooth probability density curve; calculating the goodness-of-fit index between the probability density curve and the actual sample frequency distribution, calculating the smoothness coefficient of the curve, and evaluating the continuity and regularity of the distribution; and converting the evaluation result into a distribution continuity index.

5. The method of claim 1, wherein: The acquisition process of the analysis sensitivity difference includes: extracting matrix type labeling information from the critical sample database, and classifying the critical samples according to the matrix type; obtaining the critical resolution index corresponding to each matrix group; calculating the difference between the critical resolution index of each matrix group and the critical resolution index of the normal matrix group to obtain the sensitivity deviation value of the interference matrix relative to the normal matrix; when the sensitivity deviation value exceeds the preset threshold, identifying the matrix type with a lower critical resolution index than the normal matrix group, and marking it as an analysis sensitivity difference of the blood detection equipment.

6. The blood detection equipment performance evaluation method based on big data according to claim 1, characterized in that: The acquisition process of the performance degradation warning includes: setting the length and sliding step of the sliding time window, the length of the time window being a preset number of days; extracting the critical sample data in each time window in turn from the current time point backward according to the sliding step, calculating the critical resolution index of each time window respectively, and forming a time-ordered critical resolution index sequence; performing trend analysis on the critical resolution index sequence, fitting a performance trend curve, and setting a mutation detection mechanism by comparing the original data with the baseline value to identify sudden changes in performance; calculating the slope of the performance trend curve, and determining a continuous downward trend when the slope is negative and the absolute value exceeds a preset threshold; comparing the critical resolution index of the latest time window with the critical resolution index of the baseline time window, and triggering a performance degradation warning signal when the decline amplitude exceeds a preset percentage and / or a mutation is detected.

7. The blood detection equipment performance evaluation method based on big data according to claim 1, characterized in that: The device performance evaluation report acquisition process includes: comparing the overall critical resolution index with the historical optimal value of the device to generate an overall analysis sensitivity rating; obtaining the critical resolution index of each matrix type, constructing a matrix-specific sensitivity comparison matrix, and identifying the matrix type with a short sensitivity; obtaining performance trend curve data, generating a time series trend graph, and marking the performance degradation warning time point; generating targeted improvement suggestions based on the identified sensitivity short board and performance degradation, including suggestions for instrument calibration, reagent replacement, and / or maintenance. 8.A big data based blood testing device performance evaluation system, characterized in that, The method comprises the following steps: a data acquisition module is used to collect real sample detection data of the detection equipment within a preset time period; a critical region division module is used to determine upper and lower critical regions of the detection item and establish a critical sample database; a sensitivity analysis module is used to calculate a critical resolution index based on the critical sample database, wherein the critical resolution index is obtained by weighted calculation of a critical precision index, a boundary separation index and a distribution continuity index; a matrix specificity analysis module is used to group the critical sample database according to matrix types, calculate the critical resolution index of each matrix and identify the analysis sensitivity difference under different matrix conditions; a dynamic monitoring module is used to set a sliding time window, obtain the critical resolution index in different time periods, obtain an analysis sensitivity time sequence and perform trend analysis to output a performance degradation curve and trigger a performance degradation warning; and a report generation module is used to fuse the critical resolution index, the analysis sensitivity difference and the performance degradation curve and output an equipment performance evaluation report.

Images