A digital thread driven intelligent management system for clinical laboratories

Abstract

The application discloses a digital thread driven intelligent management system for a clinical laboratory, which comprises a system body, and the system body is operated through the following method, and the specific method comprises the following steps: obtaining patient historical records and same batch sample data from a pre-established database by collecting current test result data, and obtaining multimodal information containing original images, detection curves and trend deviations; determining potential abnormal patterns by processing the original images and the detection curves through an abnormality detection algorithm according to the multimodal information; if the abnormal patterns are inconsistent with historical comparison results, comparing previous results of the patient through a trend analysis algorithm to judge the trend deviation degree; obtaining critical project specific threshold values according to the trend deviation degree to obtain key clue recognition results; and the application aims to solve the problem that in the prior art, when facing critical value projects, comprehensive verification and rapid release cannot be balanced, resulting in poor quality control and efficiency of the clinical laboratory.

Description

Technical Field

[0001] This invention relates to the technical field of information technology, specifically to a digital thread-driven intelligent management system for clinical laboratories. Background Technology

[0002] As a crucial link between clinical diagnosis and treatment decisions in medical institutions, the accuracy and reliability of laboratory test results directly impact patient safety and the quality of medical care. With the rapid increase in the number and complexity of laboratory tests, traditional methods of laboratory management relying on manual experience are no longer sufficient to meet the demands of modern medicine for efficiency, accuracy, and traceability. There is an urgent need to introduce systematic digital management tools to improve overall operational efficiency.

[0003] Currently, most inspection management systems still rely on a single review path, primarily depending on manual review by inspectors. While this method ensures result quality to some extent, it is prone to errors due to fatigue, distraction, or inexperience when workload surges. Furthermore, relying solely on manual review makes it difficult to perform real-time comparisons and anomaly pattern identification on massive amounts of historical data, resulting in some potential risks going undetected. These limitations make the verification process for inspection results both time-consuming and inherently unstable.

[0004] A deeper technical challenge lies in the fact that verifying the accuracy of test results requires balancing two interrelated requirements: comprehensiveness and timeliness. On the one hand, key test data often involves multimodal information such as raw images and detection curves. Only by combining these raw records can a reliable judgment be made, but manually checking this complex information item by item is time-consuming and inefficient. On the other hand, test results must also maintain a high degree of consistency with patient history records and data from the same batch of samples. Any subtle abnormalities may indicate instrument drift, sample problems, or operational errors, and it is difficult for humans to complete such multi-dimensional in-depth comparisons in a short period of time. An inherent tension exists between these two requirements: pursuing comprehensive verification will significantly slow down the speed of result release, while emphasizing timeliness may sacrifice the thoroughness of verification.

[0005] For example, when dealing with critical values ​​such as blood gas analysis or cardiac marker testing, laboratory technicians need to complete the review and confirmation within a very short time. However, subtle abnormalities in the original waveform and deviations from the trend of the patient's previous results often require repeated comparisons to confirm whether they are truly abnormal. Relying solely on quick manual review can easily miss crucial clues; repeated verification may delay clinical intervention. This contradiction between comprehensive verification and rapid release has become a key issue restricting the quality control and efficiency improvement of the laboratory. Summary of the Invention

[0006] This invention provides a digital thread-driven intelligent management system for clinical laboratories, aiming to solve the problem in existing technologies where it is impossible to balance comprehensive verification and rapid release when facing critical value items, which leads to a decline in the quality control and efficiency of clinical laboratories.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A digital thread-driven intelligent management system for clinical laboratories includes a system ontology. The system ontology operates through the following methods: First, it collects current test result data and retrieves patient historical records and sample data from the same batch from a pre-established database to obtain multimodal information including original images, detection curves, and trend deviations. Second, based on the multimodal information, it processes the original images and detection curves using an anomaly detection algorithm to identify potential anomaly patterns. Third, if the anomaly pattern is inconsistent with historical comparison results, it compares the patient's previous results using a trend analysis algorithm to determine the degree of trend deviation. Fourth, based on the degree of trend deviation, it obtains a specific threshold for critical items to obtain key clue identification results. Fifth, based on the key clue identification results, it integrates comprehensive verification elements using a real-time data fusion method to determine the verification completeness. Sixth, if the verification completeness is higher than a preset threshold, it outputs a timely release signal through a rapid review process to obtain final test result confirmation. Finally, based on the final test result confirmation, it obtains feedback data on anomaly patterns and trend deviations, updates the pre-established database, and obtains an optimized historical comparison basis.

[0009] In one aspect of the invention, the step of acquiring patient historical records and sample data from the same batch from a pre-established database by collecting current test result data, and obtaining multimodal information including original images, detection curves, and trend deviations, includes:

[0010] Acquire the original image and detection curve data of the current sample;

[0011] Extract images and curve sequences from historical test records in the database using patient identification;

[0012] Obtain the detection curve data of all samples in the same batch based on the sample identifier;

[0013] The current detection curve is compared point by point with the corresponding curve sequence in the historical test records to obtain the deviation value sequence at each time point;

[0014] The moving average method is used to process the deviation numerical series to obtain a smoothed trend deviation curve;

[0015] If multiple consecutive segments of the trend deviation curve exceed the preset range, they are marked as abnormal trend segments.

[0016] The original image, detection curve, trend deviation curve, and abnormal trend segment positions are stored together to form a multimodal comprehensive record of the current test.

[0017] In one aspect of the present invention, the step of processing the original image and detection curve using an anomaly detection algorithm based on multimodal information to determine potential anomaly patterns includes:

[0018] The raw image data and detection curve data in the multimodal information are acquired and classified and stored separately through a pre-established storage structure to obtain a preliminary image set and curve set;

[0019] The initially organized image set was then standardized using image processing tools to obtain a standardized image dataset.

[0020] Based on a standardized image dataset, an anomaly detection algorithm is applied to scan the image data layer by layer to identify potential anomaly regions in the image.

[0021] For the initially sorted set of curves, the curve data is preprocessed using a data smoothing tool to obtain a smoothed curve dataset;

[0022] If some segments in the smoothed curve dataset deviate from the preset threshold range, they are marked as abnormal curve segments, and the annotated curve anomaly information is obtained.

[0023] By integrating the anomaly information of the labeled curves and the potential anomaly regions in the image, an information matching method is used to perform correlation analysis to determine the correspondence between the two.

[0024] Based on the results of the correlation analysis, a comprehensive anomaly pattern record is generated to determine the final anomaly detection output.

[0025] In one aspect of the invention, if the abnormal pattern is inconsistent with historical comparison results, the degree of trend deviation is determined by comparing the patient's previous results using a trend analysis algorithm, including:

[0026] The original image data and detection curve data in the multimodal information are obtained and stored separately according to the pre-established storage structure to obtain the original image set and the original curve set.

[0027] For the original image set, image enhancement tools are used to adjust the contrast and correct the brightness to obtain the enhanced image set;

[0028] Based on the enhanced image set, the isolated forest algorithm is applied to scan the image pixel distribution, mark the isolated abnormal pixel regions, and obtain the isolated abnormal pixel marking results.

[0029] For the original set of curves, the curves are smoothed using the moving average method to obtain a smoothed set of curves;

[0030] If the slope of a certain curve in the set of smooth curves exceeds the preset slope range, it is marked as a slope abnormal segment, and the slope abnormal segment marking result is obtained.

[0031] By comparing the timestamps and locations of isolated abnormal pixel marking results and slope abnormal segment marking results, a matching operation is performed to obtain the successfully matched abnormal corresponding groups.

[0032] Based on the successfully matched abnormal corresponding groups and the patient's historical detection sequence, a trend analysis algorithm is used to calculate the abnormal position offset of multiple consecutive detections, determine the direction and magnitude of the trend deviation, and obtain the final abnormal trend pattern record.

[0033] In one aspect of the invention, obtaining a specific threshold for a critical project based on the degree of trend deviation to obtain a key clue identification result includes:

[0034] From the pre-established database, based on the critical project classification, historical data related to trend deviations are obtained to determine the initial project classification range;

[0035] Based on the initial project classification scope, the data groups that match specific thresholds are filtered out through the extraction process to obtain the threshold set related to critical projects;

[0036] Based on the threshold set, correspondence analysis is used, combined with the dynamic changes in trend deviation, to determine the potential correlation of key clues and obtain preliminary clue screening results.

[0037] Based on the initial clue screening results, if the matching degree of key clues exceeds the preset threshold range, they are prioritized through judgment logic to determine the set of high-priority clues.

[0038] Based on the high-priority set of clues and the contextual information of the project classification, feature data related to the recognition results are obtained to obtain the final list of clues to be recognized.

[0039] For the final list of identified clues, the key clues and their association with critical projects are stored in a structured manner through result recording, which serves as the input basis for subsequent analysis.

[0040] In one aspect of the present invention, the step of integrating comprehensive verification elements and determining verification completeness through a real-time data fusion method based on key clue identification results includes:

[0041] Based on the identification results of key clues, real-time data associated with the clues are obtained from a pre-established database to determine the initial scope of data integration;

[0042] Based on the initial data integration scope, a fusion method is used to process the real-time data in multiple dimensions to obtain a structured set of verification elements;

[0043] Based on the structured set of verification elements, extract the core information related to comprehensive verification. If the coverage of the core information meets the preset threshold, it is determined that it meets the requirements of the verification process.

[0044] For the set of elements that meet the requirements of the verification process, obtain contextual data related to the result analysis and determine the priority order of the verification elements;

[0045] By prioritizing the verification elements, a logical matching method is used to deeply mine the clue associations and obtain data mappings related to complete judgment.

[0046] Based on the data mapping related to the completeness judgment, integrate the final information of the comprehensive verification, and determine the achievement of the verification completeness.

[0047] In one aspect of the invention, if the verification completeness is higher than a preset threshold, a timely release signal is output through a rapid review process to obtain final verification result confirmation, including:

[0048] Based on the clue matching results, extract the corresponding real-time information from the associated resource library;

[0049] For the extracted real-time information, define the data collection boundaries and perform the collection operation;

[0050] Based on the collected real-time information, multi-source synchronous processing is performed to obtain ordered verification elements;

[0051] For ordered verification elements, calculate the completeness of the match and compare it with a preset standard;

[0052] If the complete matching degree reaches the preset standard, the ordered verification elements are prioritized according to the context association information to obtain a priority sequence;

[0053] Based on the priority sequence, the sequence alignment method is used to compare the clues layer by layer, uncover the implicit correspondences and form the basis for integrity assessment.

[0054] Based on the integrity assessment criteria, summarize the verification conclusions, determine the verification integrity, and output the determination results.

[0055] If the judgment result shows that the verification completeness is higher than the preset threshold, the rapid review process is initiated, a timely release signal is generated, and the final inspection result confirmation is obtained.

[0056] In one aspect of the invention, the step of confirming the final test results, obtaining feedback data on abnormal patterns and trend deviations, updating the pre-established database, and obtaining an optimized historical comparison basis includes:

[0057] Based on the optimized historical comparison benchmark, anomaly patterns are initially screened, and related historical records are retrieved from the database to determine the initial classification of the anomaly patterns.

[0058] For the initial classification of abnormal patterns, a hierarchical comparison method is used to analyze the specific manifestations of trend deviations and obtain a detailed description of the deviation distribution.

[0059] Based on the detailed description of the deviation distribution, key points in the feedback data are extracted and matched with a pre-established rule base to determine whether the deviation conforms to a known pattern.

[0060] If the deviation matches a known pattern, the corresponding solution is obtained from historical comparisons, a targeted adjustment strategy is generated, and the direction of subsequent processing is determined.

[0061] If the deviation does not conform to the known pattern, the test feedback is deeply mined through pattern analysis tools to identify potential new patterns and obtain new classification criteria.

[0062] Based on the new classification criteria, the optimization benchmark in the database is updated, and the feedback data is reorganized in conjunction with the data extraction process to determine the updated comparison basis.

[0063] Based on the updated comparison basis, a dynamic monitoring mechanism for abnormal patterns is generated. The applicability of the monitoring mechanism is determined through continuous data collection and result confirmation.

[0064] In one aspect of the invention, the system includes:

[0065] The data acquisition and processing module is used to collect current test result data and obtain patient history records and sample data from the same batch from a pre-established database to obtain multimodal information including original images, test curves and trend deviations.

[0066] The anomaly detection module is used to process the original image and detection curve based on multimodal information and anomaly detection algorithms to determine potential anomaly patterns.

[0067] The trend analysis module is used to compare the patient's previous results with those of the past few times to determine the degree of trend deviation if the abnormal pattern is inconsistent with the historical comparison results.

[0068] The key clue identification module is used to obtain specific thresholds for critical projects based on the degree of trend deviation, and to obtain key clue identification results.

[0069] The data fusion and verification module is used to integrate comprehensive verification elements based on the results of key clue identification using real-time data fusion methods to determine the completeness of verification.

[0070] The result confirmation and release module is used to output a timely release signal through a rapid review process if the verification completeness is higher than a preset threshold, so as to obtain the final test result confirmation.

[0071] The database update module is used to confirm the final test results, obtain feedback data on abnormal patterns and trend deviations, update the pre-established database, and obtain an optimized historical comparison basis.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] This invention collects current test result data, retrieves patient historical records and sample data from a pre-established database, and integrates multimodal information such as original images, detection curves, and trend deviations. An anomaly detection algorithm processes the original images and detection curves to identify potential anomaly patterns. These anomaly patterns are then compared with the patient's historical results for consistency assessment. If inconsistencies are found, a trend analysis algorithm is used to compare the patient's previous results to quantify the degree of trend deviation. Based on the degree of deviation, a specific threshold for critical items is applied to obtain key clue identification results. Subsequently, a real-time data fusion method is used to integrate comprehensive verification elements and calculate verification completeness. When the verification completeness exceeds a preset threshold, a rapid review process is triggered, outputting a timely release signal to ultimately confirm the test results. Simultaneously, feedback data on anomaly patterns and trend deviations is fed back to update the database to optimize the historical comparison basis. This invention effectively solves the interconnected core business problems in clinical testing, such as the difficulty in real-time fusion and judgment of multi-source anomalies, low verification efficiency when historical trends conflict with current results, and high risk of missed critical values. It achieves overall technical effects of accurate anomaly identification, dynamic trend tracking, closed-loop optimization of verification, and rapid and reliable result release, significantly improving the accuracy and timeliness of test results. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0075] Figure 1 This is a flowchart of a digital thread-driven intelligent management method for clinical laboratories according to the present invention.

[0076] Figure 2This is a schematic diagram of a digital thread-driven intelligent management method for clinical laboratories according to the present invention.

[0077] Figure 3 This is another schematic diagram of a digital thread-driven intelligent management method for clinical laboratories according to the present invention. Detailed Implementation

[0078] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0079] Please see Figures 1-3 As shown, this embodiment discloses a digital thread-driven intelligent management system for clinical laboratories. The system operates through the following methods, specifically including:

[0080] S101. By collecting current test result data, obtain patient historical records and sample data from the same batch from a pre-established database to obtain multimodal information including original images, detection curves, and trend deviations.

[0081] The system acquires the original image and detection curve data for the current sample. It extracts images and curve sequences from historical test records using patient identifiers. It obtains the detection curve data for all samples in the same batch based on their identifiers. The system compares the current detection curve with the corresponding curve sequences in historical test records point-by-point to obtain the deviation numerical sequence at each time point. It processes the deviation numerical sequence using a moving average method to obtain a smoothed trend deviation curve. If multiple consecutive segments of the trend deviation curve exceed a preset amplitude range, they are marked as abnormal trend segments. The system associates and stores the original image, detection curve, trend deviation curve, and the locations of abnormal trend segments together to form a multimodal comprehensive record of the current test.

[0082] Specifically, the system first collects the patient's current batch of test results in real time through the LIS interface. For example, a patient's serum creatinine level on a certain day in 20XX was 135.6 μmol / L and blood urea nitrogen was 8.9 mmol / L, and the sample barcode SC20XX020100478 is automatically associated. Then, using the patient's unique identifier P87654321 as the key, the system extracts the patient's historical test records for the past three years from the pre-established HBase distributed database, retrieving a total of 47 valid records. The historical creatinine values ​​show a slow upward trend, with an average of 112.4 μmol / L in December 2025 and 98.7 μmol / L in the same period of 2024. Simultaneously, the system retrieves the creatinine test results of 89 other samples within the same batch from the database, forming a batch dataset with a batch mean of 124.3 μmol / L and a standard deviation of 11.2 μmol / L. The system then invoked a pre-trained multimodal feature extraction model to read the microscope image file of the sample from the original image folder associated with LIMS. It extracted a 1024-dimensional image feature vector using ResNet-50 combined with an attention mechanism. Simultaneously, it read 3600 sampling points (1 point per second, acquisition time 1 hour) from the detection curve data curve_0478.csv exported from the fully automated biochemical analyzer and extracted 512-dimensional curve features using a one-dimensional convolutional neural network Conv1D (kernel size 5, channel number 64). The deviation percentage of the current creatinine value of 135.6 μmol / L from the patient's historical mean (weighted exponential smoothing, weight decay factor 0.92) was calculated to be 20.7%, with a Z-score of (135.6-112.4) / 11.2=2.07, indicating a significantly elevated trend. Finally, the 1024-dimensional image features, 512-dimensional curve features, and the above-mentioned numerical deviation indicators (a total of 3 dimensions) are fused at the feature level and mapped to a unified 256-dimensional multimodal representation vector through a two-layer fully connected network (1024-512-256). This vector is used for subsequent anomaly warning or disease progression prediction. The entire process is completed on a GPU cluster and takes about 1.8 seconds.

[0083] S102. Based on multimodal information, anomaly detection algorithms are used to process the original image and detection curves to determine potential anomaly patterns.

[0084] The process involves acquiring raw image data and detection curve data from multimodal information, classifying and storing them separately using a pre-established storage structure to obtain preliminary image and curve sets. For the preliminary image set, image processing tools are used for standardization to obtain a normalized image dataset. Based on the normalized image dataset, anomaly detection algorithms are applied to scan the image data layer by layer to identify potential anomaly regions. For the preliminary curve set, data smoothing tools are used to preprocess the curve data to obtain a smoothed curve dataset. If certain segments in the smoothed curve dataset deviate from a preset threshold range, they are marked as anomalous curve segments, resulting in labeled curve anomaly information. By integrating the labeled curve anomaly information and the potential anomaly regions in the images, an information matching method is used for correlation analysis to determine the correspondence between the two. Based on the results of the correlation analysis, a comprehensive anomaly pattern record is generated to determine the final anomaly detection output.

[0085] Specifically, the system automatically analyzes multimodal data by integrating an anomaly detection module. First, it extracts the original microscope image and reaction process detection curve corresponding to the sample from the LIMS system. The image is then input into a pre-trained variational autoencoder (VAE) with a latent dimension of 128 and a KL divergence weight of 0.001. The reconstruction error is calculated as a pixel-level mean square error (MSE) of 0.042, which is higher than the normal sample threshold of 0.018, indicating the possible presence of abnormal cell morphology regions in the image. Simultaneously, the Isolation Forest algorithm (with a contamination rate parameter of 0.015 and 200 trees) is applied to the detection curve. Three abnormal subsequences are detected in the 3600 sampling point sequence, with the highest anomaly score reaching 0.78 (threshold 0.65). The corresponding reaction time period is from 1240 seconds to 1480 seconds, which is characterized by a sudden drop in absorbance followed by an abnormal rebound. Next, the system concatenates the 128-dimensional latent vector obtained from VAE encoding with the anomaly score sequence output by the Isolation Forest (downsampled to 64-dimensional statistical features, including the maximum anomaly score, anomaly duration, and anomaly percentage) to form a 192-dimensional joint representation vector. This vector is then input into a single-class support vector machine (OC-SVM) with a kernel function of RBF, nu=0.02, and gamma=0.0015. The decision function output value is -0.364 (decision boundary is 0), indicating an anomaly sample deviating from the normal pattern. Further analysis of the curve statistical characteristics of other samples within the batch (the mean slope of the response curves for the 89 samples in the same batch is 0.021 / min, and the standard deviation is 0.0043 / min) reveals that the current sample's slope reaches 0.037 / min, exceeding three times the standard deviation, confirming a significant deviation in the response dynamics. The system ultimately generates an anomaly pattern label as "image cell aggregation + nonlinear mutation of response curve" and calculates a comprehensive anomaly confidence score of 0.892, which is used to trigger downstream clinical review or further molecular detection. The entire anomaly detection process is completed under a distributed computing framework, with an average processing time of approximately 2.3 seconds.

[0086] S103. If the abnormal pattern is inconsistent with the historical comparison results, the trend analysis algorithm is used to compare the patient's previous results to determine the degree of trend deviation.

[0087] The process involves acquiring raw image data and detection curve data from multimodal information and storing them separately according to a pre-established storage structure, resulting in a raw image set and a raw curve set. For the raw image set, image enhancement tools are used to adjust contrast and correct brightness, resulting in an enhanced image set. Based on the enhanced image set, the isolated forest algorithm is applied to scan the image pixel distribution, marking isolated abnormal pixel regions, resulting in isolated abnormal pixel marking results. For the raw curve set, the curves are smoothed using a moving average method, resulting in a smoothed curve set. If the slope of a curve segment in the smoothed curve set exceeds a preset slope range, it is marked as a slope abnormal segment, resulting in a slope abnormal segment marking result. By comparing the timestamps and positions of the isolated abnormal pixel marking results and the slope abnormal segment marking results, a matching operation is performed to obtain successfully matched abnormal correspondence groups. Based on the successfully matched abnormal correspondence groups and the patient's historical detection sequence, a trend analysis algorithm is used to calculate the abnormal position offset of multiple consecutive detections, determining the direction and magnitude of the trend deviation, and obtaining the final abnormal trend pattern record.

[0088] Specifically, the system automates the processing of the current sample through a multimodal anomaly detection process. First, it retrieves the fluorescence microscopy image of sample number S20XX0201-392 and its corresponding enzyme-catalyzed reaction time series data from the LIMS database. The fluorescence image is then fed into a pre-trained convolutional autoencoder (CAE) (the encoder contains 4 convolutional layers, with a latent space dimension of 96, and the reconstruction loss uses a composite loss function combining structural similarity (SSIM). The calculated pixel-level reconstruction error MSE is 0.031, and the SSIM value is 0.874, which is lower than... The normal sample SSIM threshold was 0.935, indicating localized uneven distribution of fluorescence signals in the image, potentially indicating protein aggregation or organelle damage. Simultaneously, the Local Anomaly Factor (LOF) algorithm (k-nearest neighbor count set to 35, contamination ratio preset to 0.012) was applied to the reaction curve sequence, identifying four regions with low local density across 4200 time points. The highest LOF score was 2.47 (threshold 1.80), with the abnormal interval concentrated between 1980 and 2310 seconds, characterized by a sudden plateau followed by an abrupt drop in enzyme activity. The system then extracted a 96-dimensional latent feature vector from the CAE and extracted statistical summary features (including abnormal peak values, the proportion of total duration of abnormal intervals, and the amplitude of abnormal density fluctuations, compressed to 48 dimensions) from the LOF anomaly detection results. These two features were concatenated to form a 144-dimensional fused feature vector, which was input into the Deep Isolated Forest (DIF) extended model (integrating 1000 trees, subsampling rate 0.65). The anomaly score was calculated to be 0.692 (normal decision threshold 0.58), confirming that the overall sample deviated from the baseline distribution. Further analysis of the reaction rates of 76 control samples from the same batch (batch average maximum reaction rate Vmax was 0.0182ΔA / min, standard deviation 0.0031ΔA / min) revealed that the current sample's Vmax reached 0.0276ΔA / min, exceeding the limit of mean plus 2.5 times the standard deviation, indicating an abnormal acceleration in the enzymatic reaction. The system combined the anomaly indications in the image region with deviations in the curve kinetics, calculating an anomaly confidence probability of 0.876 using weighted Bayesian updates. The automatic anomaly pattern was labeled as "local high-brightness aggregation in the fluorescence image + sudden drop in the enzyme reaction acceleration plateau." This result triggered a subsequent automatic generation of molecular targeting verification tasks. The entire analysis process was completed within a containerized microservice architecture, averaging approximately 1.9 seconds.

[0089] S104. Based on the degree of trend deviation, obtain the specific threshold for critical projects and get the key clue identification results.

[0090] From a pre-established database, historical data related to trend deviations is retrieved based on the classification of critical projects to determine the initial project classification range. For this initial range, data groups matching specific thresholds are extracted to obtain a threshold set related to critical projects. Based on this threshold set, correspondence analysis is performed, combined with dynamic changes in trend deviations, to determine the potential correlation of key clues, resulting in preliminary clue screening results. If the matching degree of key clues exceeds a preset threshold range, they are prioritized using a judgment logic to determine a high-priority clue set. Based on this high-priority clue set, and combined with the contextual information of the project classification, feature data related to the identification results is retrieved to obtain the final clue identification list. For the final clue identification list, the correlation information between key clues and critical projects is structured and stored through result recording to determine the input basis for subsequent analysis.

[0091] Specifically, the system assesses the degree of trend deviation in patient sample data through an automated analysis process. First, it extracts biochemical indicator data from the patient's last six tests from the database. Taking sample number S20XX0201-405 as an example, it obtains the sequence of specific protein concentration changes in the patient's serum. The data points are six time nodes, namely 2.3, 2.8, 3.1, 3.9, 4.5, and 5.2 mg / L. The system uses the moving average smoothing method in time series analysis algorithms (with a window size of 3) to calculate the smoothed trend values ​​as 2.4, 2.9, 3.5, and 4.2 mg / L. Then, it fits the trend slope through a linear regression algorithm, obtaining a slope of 0.58 mg / L / test. The slope threshold of 0.35 mg / L / test exceeds the normal range, indicating an abnormally accelerated increase in protein concentration. Next, based on the degree of trend deviation, the system automatically retrieves the critical item database to obtain the specific upper limit of the threshold corresponding to the protein concentration, which is 4.0 mg / L. Combining this with the historical data standard deviation of 0.4 mg / L, the system calculates that the current sample value of 5.2 mg / L exceeds the threshold plus 1.5 times the standard deviation, triggering a high-risk warning. Further, the system utilizes a key clue identification algorithm, employing the density-based clustering method DBSCAN (neighborhood radius set to 0.5, minimum sample size 2), to compare the patient's historical data with the distribution of other similar patients. It identifies the current sample's isolation level in the concentration increase rate dimension as 0.82 (normal threshold 0.60), confirming it as a key abnormal clue. The system then performs a correlation analysis with other auxiliary indicators such as inflammatory factor levels (current value 8.7 ng / mL, normal upper limit 6.5 ng / mL). The Pearson correlation coefficient yields a correlation of 0.76, suggesting a possible underlying inflammation-driven protein expression abnormality. Finally, the system automatically generates a critical project report, marks the degree of trend deviation as "high-risk acceleration", and pushes the results to the subsequent automated verification process. The entire process is completed in a cloud-based distributed computing framework and takes about 2.3 seconds.

[0092] S105. Based on the key clue identification results, a real-time data fusion method is used to integrate all verification elements and determine the completeness of the verification.

[0093] Based on the identification results of key clues, real-time data associated with the clues is retrieved from a pre-established database to determine the initial data integration scope. For this initial scope, a fusion method is used to process the real-time data in multiple dimensions, resulting in a structured set of verification elements. Based on this structured set, core information relevant to comprehensive verification is extracted. If the coverage of this core information meets a preset threshold, it is deemed to meet the verification process requirements. For the set of elements that meet the verification process requirements, contextual data related to result analysis is obtained to determine the priority ranking of the verification elements. Using this priority ranking, a logical matching method is employed to deeply mine clue associations, obtaining data mappings related to complete judgment. Based on these data mappings, the final information from comprehensive verification is integrated to determine the achievement of verification completeness.

[0094] Specifically, the system comprehensively verifies and integrates key clue identification results through real-time data fusion. Taking sample number S20XX0201-407 as an example, it first extracts the latest CA19-9 concentration sequence and related imaging quantitative indicators of the patient from a multi-source database. The CA19-9 values ​​for the last 5 times are 38.2, 45.6, 59.1, 72.8, and 91.4 U / mL. At the same time, it integrates the pancreatic lesion volume sequence calculated by AI segmentation of CT images from the same day, which are 1.8, 2.4, 3.1, 4.0, and 5.3 cm³. The Kalman filter algorithm is used to suppress noise and estimate the state of the two sets of sequences. After fusion, a smooth trend vector is obtained. Then, a multivariate vector autoregressive model (VAR) with a lag order of 2 is used to calculate the dynamic interaction between CA19-9 concentration and lesion volume. The Granger causality test p value is 0.007, confirming that volume increase has a significant driving effect on concentration increase. Next, the system invoked a pre-defined multimodal validation rule base to obtain the dynamic threshold upper limit of CA19-9 under the current trend, which is 85.0 U / mL. Combined with the standard deviation of the historical cohort of patients of the same age group, which represents a fluctuation range of 1.2 times, the system determined that the current value of 91.4 U / mL had exceeded the fusion threshold boundary. Simultaneously, the rate of increase in lesion volume was 0.68 cm³ / time, exceeding the baseline of 0.42 cm³ / time, triggering a dual-indicator synergistic high-risk signal. Further, the system introduced an abnormal pattern matching module, using the isolated forest algorithm (contamination rate parameter 0.08) to calculate the abnormal score on the fusion feature vector. The current sample's isolated score was 0.79 (warning threshold 0.65), confirming a high degree of deviation in the multidimensional space, and marking it as a core abnormal point requiring enhanced validation. Subsequently, this point was used in dynamic Bayesian network inference with the concurrently collected serum inflammation composite index (current value 12.4, reference upper limit 9.8), calculating the conditional probability P(tumor progression | elevated inflammation index ∩ abnormal biomarker) = 0.84, suggesting that the inflammatory microenvironment may accelerate lesion evolution. Ultimately, the system automatically aggregated the verification completeness score to 0.92 (out of 1.0), determining that the verification chain was sufficiently closed, generating an electronic verification conclusion containing a multi-source evidence chain, and directly transferring it to the clinical decision support stage. The entire fusion verification process was completed in a high-performance computing cluster, taking approximately 1.8 seconds.

[0095] S106. If the verification completeness is higher than the preset threshold, a timely release signal will be output through the rapid review process to obtain the final verification result confirmation.

[0096] Based on the clue matching results, corresponding real-time information is extracted from the associated resource library. For the extracted real-time information, data collection boundaries are defined and collection operations are performed. Based on the collected real-time information, multi-source synchronous processing is performed to obtain ordered verification elements. For each ordered verification element, the completeness matching degree is calculated and compared with a preset standard. If the completeness matching degree reaches the preset standard, the ordered verification elements are prioritized based on contextual association information to obtain a priority sequence. Based on the priority sequence, a sequence alignment method is used to compare clues layer by layer, uncovering implicit correspondences and forming a completeness assessment basis. Based on the completeness assessment basis, verification conclusion information is summarized, the verification completeness is determined, and the judgment result is output. If the judgment result shows that the verification completeness is higher than a preset threshold, a rapid review process is initiated, a timely release signal is generated, and the final inspection result confirmation is obtained.

[0097] Specifically, after the verification completeness exceeds a preset threshold, the system automatically initiates a rapid review process to ensure timely signal release and confirmation of the final test results. Taking sample number S20XX0315-512 as an example, the system first extracts the patient's recent serum proteomics data from the integrated database, obtaining the concentration values ​​of the specific inflammatory factor IL-6 from the past four tests as 12.5, 15.8, 19.3, and 24.7 pg / mL, respectively. Simultaneously, the system retrieves the lesion boundary clarity score sequence from the concurrent ultrasound image analysis, which is 0.72, 0.65, 0.58, and 0.51, indicating that the boundary is gradually becoming blurred. The system then applies a sliding window algorithm (window size 3) to smooth the trend of the IL-6 concentration sequence, calculating an average growth rate of 4.1 pg / mL / time. The system then analyzes the correlation between the boundary clarity score and the inflammatory factor concentration using a logistic regression model, obtaining a regression coefficient of -0.18 and a p-value of 0.012, confirming a significant negative correlation between the two, suggesting that inflammation may lead to the worsening of the lesion boundary. Next, based on a pre-defined rapid verification rule base, the system compares the current IL-6 concentration of 24.7 pg / mL with the dynamic threshold upper limit of 20.0 pg / mL. It finds that the concentration exceeds the threshold by 24.5%, and the boundary clarity score of 0.51 is lower than the warning value of 0.60, triggering a two-parameter anomaly warning. Further, the system uses a support vector machine algorithm (with a radial basis function kernel) to classify and predict the fused feature vector, calculating an anomaly probability of 0.88 (threshold 0.75), marking the sample as a high-risk category. Subsequently, the system integrates treatment response indicators from historical data (such as drug sensitivity scores, currently 3.2, reference value 5.0) through a pre-trained decision tree model, deriving a potential progression risk weight of 0.76, and combining this with the anomaly probability to generate a comprehensive risk index of 0.82 (out of 1.0). Finally, the system automatically generates a review report with a verification completeness score of 0.95, which is higher than the preset threshold of 0.90. The signal release process is completed on the distributed computing platform, taking about 1.2 seconds, and is directly pushed to the downstream early warning module, forming a closed verification chain.

[0098] S107. Based on the final test results, obtain feedback data on abnormal patterns and trend deviations, update the pre-established database, and obtain the optimized historical comparison basis.

[0099] Based on the optimized historical comparison benchmark, anomaly patterns are initially screened by retrieving relevant historical records from the database to determine the initial classification of the anomaly patterns. For this initial classification, a hierarchical comparison method is used to analyze the specific manifestations of trend deviations, obtaining a detailed description of the deviation distribution. Based on this detailed description, key points in the feedback data are extracted and matched against a pre-established rule base to determine if the deviation conforms to a known pattern. If the deviation conforms to a known pattern, corresponding solutions are obtained from historical comparisons to generate targeted adjustment strategies and determine the direction of subsequent processing. If the deviation does not conform to a known pattern, pattern analysis tools are used to deeply mine the test feedback, identify potential new patterns, and obtain new classification criteria. Based on the new classification criteria, the optimized benchmark in the database is updated, and the feedback data is reorganized using the data extraction process to determine the updated comparison basis. For the updated comparison basis, a dynamic monitoring mechanism for anomaly patterns is generated. The applicability of the monitoring mechanism is assessed through continuous data collection and result confirmation.

[0100] Specifically, after confirmation based on the final test results, the system identified the abnormal pattern of the sample as a sustained and rapid increase in inflammatory factors accompanied by a rapid expansion of the low-density area on imaging, as well as bias feedback data indicating insufficient early treatment response. This data was automatically updated to the pre-established longitudinal dynamic database of patients, forming an optimized historical comparison basis. Taking sample number S20XX0722-089 as an example, the system extracted the C-reactive protein (CRP) concentration sequences from the multimodal data lake for the last five follow-ups, which were 8.4, 11.2, 18.6, 29.1, and 41.3 mg / L. Simultaneously, the low-density lesion volumes in the corresponding CT image segmentation results were 4.8, 6.2, 9.7, 14.3, and 21.6 cm³, respectively, showing an accelerating increase in volume. The system used an exponentially weighted moving average algorithm (smoothing factor α=0.3) to process the CRP concentration sequence, obtaining a current weighted trend slope of 9.7 mg / L / time. The correlation strength between lesion volume growth rate and CRP concentration was calculated using the Pearson correlation coefficient, showing a value of 0.94 (p-value less than 0.001), confirming a strong positive correlation and suggesting that inflammation drives rapid lesion progression. Subsequently, the system invoked the embedded anomaly detection rule engine, comparing the current CRP value of 41.3 mg / L with the individualized adaptive threshold (based on the median of the previous three values ​​× 2.2) of 28.6 mg / L. The difference exceeded the threshold by 44.4%, and the lesion volume growth rate exceeded the preset warning slope of 1.8 cm³ / time, immediately activating a multi-indicator composite anomaly marker. Next, the system uses a random forest classifier (200 trees, maximum depth 12) to model multidimensional feature vectors including CRP trend slope, lesion volume change rate, and previous treatment adherence scores, outputting a malignancy probability of 0.91 (threshold set at 0.70), classifying the sample as a rapidly progressing high-risk state. Then, the system loads a gradient boosting decision tree model, integrating historical efficacy tracking data (such as the immunosuppressant sensitivity index, currently 2.1, baseline 4.5), calculating a treatment resistance contribution weight of 0.81, and weighting this with the malignancy probability to generate an overall deterioration risk score of 0.87 (out of 1.0). Finally, the system synchronously verifies the above analysis results with the optimized historical comparison base, confirming that the deviation pattern has been effectively absorbed. The update is completed in 0.9 seconds, automatically triggering the downstream personalized intervention path recommendation module, forming a continuous learning closed loop.

[0101] In some different embodiments, the present invention also provides a digital thread-driven intelligent management system for clinical laboratories, mainly comprising:

[0102] The data acquisition and processing module is used to collect current test result data and obtain patient history records and sample data from the same batch from a pre-established database to obtain multimodal information including original images, test curves and trend deviations.

[0103] The anomaly detection module is used to process the original image and detection curve based on multimodal information and anomaly detection algorithms to determine potential anomaly patterns.

[0104] The trend analysis module is used to compare the patient's previous results with those of the past few times to determine the degree of trend deviation if the abnormal pattern is inconsistent with the historical comparison results.

[0105] The key clue identification module is used to obtain specific thresholds for critical projects based on the degree of trend deviation, and to obtain key clue identification results.

[0106] The data fusion and verification module is used to integrate comprehensive verification elements based on the results of key clue identification using real-time data fusion methods to determine the completeness of verification.

[0107] The result confirmation and release module is used to output a timely release signal through a rapid review process if the verification completeness is higher than a preset threshold, so as to obtain the final test result confirmation.

[0108] The database update module is used to confirm the final test results, obtain feedback data on abnormal patterns and trend deviations, update the pre-established database, and obtain an optimized historical comparison basis.

[0109] If the technical solution of this application involves the acquisition of personal information, the product using this solution has clearly informed the user of the processing rules and obtained the user's consent before processing. If sensitive personal information is involved, the user's individual consent has been obtained and the "express consent" requirement has been met. For example, a clear sign is placed at the collection device to indicate the collection scope, and the user's voluntary entry is considered as consent; or authorization is obtained through pop-up windows, user uploads, etc. The processing rules include the processor, purpose, method, and type of information.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital thread-driven intelligent management system for clinical laboratories, characterized in that, This includes the system ontology, which operates through the following methods, specifically: By collecting current test results data, patient history records and sample data from the same batch are obtained from a pre-established database to obtain multimodal information including original images, test curves, and trend deviations; Based on multimodal information, anomaly detection algorithms are used to process the original image and detection curves to determine potential anomaly patterns; If the abnormal pattern is inconsistent with the historical comparison results, the trend analysis algorithm is used to compare the patient's previous results to determine the degree of trend deviation. Based on the degree of trend deviation, specific thresholds for critical projects are obtained to determine the key clue identification results. Based on the key clue identification results, a real-time data fusion method is used to integrate all verification elements and determine the completeness of the verification. If the verification completeness is higher than the preset threshold, a timely release signal will be output through the rapid review process to obtain the final test result confirmation; Based on the final test results, feedback data on abnormal patterns and trend deviations are obtained, the pre-established database is updated, and an optimized historical comparison basis is obtained.

2. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that: The process involves collecting current test result data and retrieving patient historical records and sample data from a pre-established database to obtain multimodal information including original images, detection curves, and trend deviations, including: Acquire the original image and detection curve data of the current sample; Extract images and curve sequences from historical test records in the database using patient identification; Obtain the detection curve data of all samples in the same batch based on the sample identifier; The current detection curve is compared point by point with the corresponding curve sequence in the historical test records to obtain the deviation value sequence at each time point; The moving average method is used to process the deviation numerical series to obtain a smoothed trend deviation curve; If multiple consecutive segments of the trend deviation curve exceed the preset range, they are marked as abnormal trend segments. The original image, detection curve, trend deviation curve, and abnormal trend segment positions are stored together to form a multimodal comprehensive record of the current test.

3. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that: The step of processing the original image and detection curve using an anomaly detection algorithm based on multimodal information to determine potential anomaly patterns includes: The raw image data and detection curve data in the multimodal information are acquired and classified and stored separately through a pre-established storage structure to obtain a preliminary image set and curve set; The initially organized image set was then standardized using image processing tools to obtain a standardized image dataset. Based on a standardized image dataset, an anomaly detection algorithm is applied to scan the image data layer by layer to identify potential anomaly regions in the image. For the initially sorted set of curves, the curve data is preprocessed using a data smoothing tool to obtain a smoothed curve dataset; If some segments in the smoothed curve dataset deviate from the preset threshold range, they are marked as abnormal curve segments, and the annotated curve anomaly information is obtained. By integrating the anomaly information of the labeled curves and the potential anomaly regions in the image, an information matching method is used to perform correlation analysis to determine the correspondence between the two. Based on the results of the correlation analysis, a comprehensive anomaly pattern record is generated to determine the final anomaly detection output.

4. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that: If the abnormal pattern is inconsistent with historical comparison results, a trend analysis algorithm is used to compare the patient's previous results to determine the degree of trend deviation, including: The original image data and detection curve data in the multimodal information are obtained and stored separately according to the pre-established storage structure to obtain the original image set and the original curve set. For the original image set, image enhancement tools are used to adjust the contrast and correct the brightness to obtain the enhanced image set; Based on the enhanced image set, the isolated forest algorithm is applied to scan the image pixel distribution, mark the isolated abnormal pixel regions, and obtain the isolated abnormal pixel marking results. For the original set of curves, the curves are smoothed using the moving average method to obtain a smoothed set of curves; If the slope of a certain curve in the set of smooth curves exceeds the preset slope range, it is marked as a slope abnormal segment, and the slope abnormal segment marking result is obtained. By comparing the timestamps and locations of isolated abnormal pixel marking results and slope abnormal segment marking results, a matching operation is performed to obtain the successfully matched abnormal corresponding groups. Based on the successfully matched abnormal corresponding groups and the patient's historical detection sequence, a trend analysis algorithm is used to calculate the abnormal position offset of multiple consecutive detections, determine the direction and magnitude of the trend deviation, and obtain the final abnormal trend pattern record.

5. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that: The process of obtaining a specific threshold for critical projects based on the degree of trend deviation, and then obtaining the key clue identification results, includes: From the pre-established database, based on the critical project classification, historical data related to trend deviations are obtained to determine the initial project classification range; Based on the initial project classification scope, the data groups that match specific thresholds are filtered out through the extraction process to obtain the threshold set related to critical projects; Based on the threshold set, correspondence analysis is used, combined with the dynamic changes in trend deviation, to determine the potential correlation of key clues and obtain preliminary clue screening results. Based on the initial clue screening results, if the matching degree of key clues exceeds the preset threshold range, they are prioritized through judgment logic to determine the set of high-priority clues. Based on the high-priority set of clues and the contextual information of the project classification, feature data related to the recognition results are obtained to obtain the final list of clues to be recognized. For the final list of identified clues, the key clues and their association with critical projects are stored in a structured manner through result recording, which serves as the input basis for subsequent analysis.

6. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that, The process of integrating comprehensive verification elements using real-time data fusion methods based on key clue identification results to determine verification completeness includes: Based on the identification results of key clues, real-time data associated with the clues are obtained from a pre-established database to determine the initial scope of data integration; Based on the initial data integration scope, a fusion method is used to process the real-time data in multiple dimensions to obtain a structured set of verification elements; Based on the structured set of verification elements, extract the core information related to comprehensive verification. If the coverage of the core information meets the preset threshold, it is determined that it meets the requirements of the verification process. For the set of elements that meet the requirements of the verification process, obtain contextual data related to the result analysis and determine the priority order of the verification elements; By prioritizing the verification elements, a logical matching method is used to deeply mine the clue associations and obtain data mappings related to complete judgment. Based on the data mapping related to the completeness judgment, integrate the final information of the comprehensive verification, and determine the achievement of the verification completeness.

7. The intelligent management system for clinical laboratories driven by digital threads according to claim 1, characterized in that, If the verification completeness is higher than a preset threshold, a timely release signal is output through a rapid review process to obtain final verification result confirmation, including: Based on the clue matching results, extract the corresponding real-time information from the associated resource library; For the extracted real-time information, define the data collection boundaries and perform the collection operation; Based on the collected real-time information, multi-source synchronous processing is performed to obtain ordered verification elements; For ordered verification elements, calculate the completeness of the match and compare it with a preset standard; If the complete matching degree reaches the preset standard, the ordered verification elements are prioritized according to the context association information to obtain a priority sequence; Based on the priority sequence, the sequence alignment method is used to compare the clues layer by layer, uncover the implicit correspondences and form the basis for integrity assessment. Based on the integrity assessment criteria, summarize the verification conclusions, determine the verification integrity, and output the determination results. If the judgment result shows that the verification completeness is higher than the preset threshold, the rapid review process is initiated, a timely release signal is generated, and the final inspection result confirmation is obtained.

8. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that, The process of confirming the final test results, obtaining feedback data on abnormal patterns and trend deviations, updating the pre-established database, and obtaining an optimized historical comparison basis includes: Based on the optimized historical comparison benchmark, anomaly patterns are initially screened, and related historical records are retrieved from the database to determine the initial classification of the anomaly patterns. For the initial classification of abnormal patterns, a hierarchical comparison method is used to analyze the specific manifestations of trend deviations and obtain a detailed description of the deviation distribution. Based on the detailed description of the deviation distribution, key points in the feedback data are extracted and matched with a pre-established rule base to determine whether the deviation conforms to a known pattern. If the deviation matches a known pattern, the corresponding solution is obtained from historical comparisons, a targeted adjustment strategy is generated, and the direction of subsequent processing is determined. If the deviation does not conform to the known pattern, the test feedback is deeply mined through pattern analysis tools to identify potential new patterns and obtain new classification criteria. Based on the new classification criteria, the optimization benchmark in the database is updated, and the feedback data is reorganized in conjunction with the data extraction process to determine the updated comparison basis. Based on the updated comparison basis, a dynamic monitoring mechanism for abnormal patterns is generated. The applicability of the monitoring mechanism is determined through continuous data collection and result confirmation.

9. The intelligent management system driven by digital threads for clinical laboratories according to claim 1, characterized in that, The system includes: The data acquisition and processing module is used to collect current test result data and obtain patient history records and sample data from the same batch from a pre-established database to obtain multimodal information including original images, test curves and trend deviations. The anomaly detection module is used to process the original image and detection curve based on multimodal information and anomaly detection algorithms to determine potential anomaly patterns. The trend analysis module is used to compare the patient's previous results with those of the past few times to determine the degree of trend deviation if the abnormal pattern is inconsistent with the historical comparison results. The key clue identification module is used to obtain specific thresholds for critical projects based on the degree of trend deviation, and to obtain key clue identification results. The data fusion and verification module is used to integrate comprehensive verification elements based on the results of key clue identification using real-time data fusion methods to determine the completeness of verification. The result confirmation and release module is used to output a timely release signal through a rapid review process if the verification completeness is higher than a preset threshold, so as to obtain the final test result confirmation. The database update module is used to confirm the final test results, obtain feedback data on abnormal patterns and trend deviations, update the pre-established database, and obtain an optimized historical comparison basis.

Images