Intelligent identification and early warning method and system for abnormal pattern of venous thrombosis test data

Abstract

This invention provides an intelligent identification and early warning method and system for abnormal patterns in venous thrombosis test data, relating to the field of data processing technology. The method includes acquiring venous thrombosis-related test data of a target patient, constructing a feature vector set based on a time-series dependency model, matching it with an abnormal pattern knowledge base to identify risks, calculating a dynamic early warning index, pushing early warning information to a medical decision-making terminal when trigger conditions are met, and adaptively adjusting the model based on clinical intervention results. This invention can improve the accuracy of early identification of venous thrombosis risk, reduce the missed diagnosis rate, and improve the timeliness of clinical intervention.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to a method and system for intelligent identification and early warning of abnormal patterns in venous thrombosis test data. Background Technology

[0002] Venous thrombosis is a common clinical condition, primarily characterized by the abnormal clotting of blood within veins, forming a thrombus. This can lead to obstructed venous return, tissue edema, and, in severe cases, life-threatening complications such as pulmonary embolism. With advancements in medical technology, diagnostic methods for venous thrombosis have become increasingly sophisticated, including multiple indicators such as D-dimer, thrombin time, and prothrombin time. These indicators can reflect the balance between coagulation and anticoagulation in a patient's body from various perspectives.

[0003] Currently, clinical assessment of venous thrombosis risk mainly relies on physicians' experience-based judgment based on patients' laboratory data or the use of traditional risk scoring scales. With the widespread adoption of electronic medical record systems, hospitals have accumulated a large amount of patient laboratory data, but this data is not yet fully utilized. Existing venous thrombosis risk assessment technologies suffer from the following problems: First, most existing assessment methods use static threshold judgments, failing to effectively capture the temporal relationships and interactions between laboratory indicators, leading to insufficient sensitivity to early, subtle changes and missed opportunities for optimal intervention. Second, existing technologies lack the ability to comprehensively analyze multidimensional laboratory indicators, typically focusing only on changes in one or a few key indicators, ignoring the complex synergistic effects between indicators, thus reducing the accuracy and comprehensiveness of risk identification. Third, existing early warning systems generally lack adaptive learning mechanisms, making it difficult to optimize the assessment model based on feedback from clinical interventions, and failing to continuously improve the accuracy of early warnings with the accumulation of medical practice, thus limiting the long-term application value of the system. Summary of the Invention

[0004] This invention provides a method and system for intelligent identification and early warning of abnormal patterns in venous thrombosis test data, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides a method for intelligent identification and early warning of abnormal patterns in venous thrombosis test data, comprising: Acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of the indicators. The feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifier corresponding to the abnormal pattern is extracted. Based on the risk level identifier and the rate of change of indicators in the feature vector set, a dynamic early warning index reflecting the evolution trend of abnormal patterns is calculated. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. When the dynamic early warning index meets the early warning triggering conditions, an early warning message containing the abnormal pattern type, the combination of related indicators, and intervention suggestions is generated and pushed to the medical decision-making terminal. Collect clinical intervention result data from medical decision-making terminals, and adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

[0006] Based on a time-series dependency model, multi-dimensional correlation analysis was performed on the venous thrombosis-related test data to construct a feature vector set reflecting the dynamic evolution of the indicators, including: The venous thrombosis-related test data are divided into multiple continuous time windows according to the time series. Time series characteristic parameters are calculated for the test data in each time window. The time series characteristic parameters are the index mean, standard deviation and coefficient of variation. Based on the time series characteristic parameters, a recurrent neural network is trained to establish a time series dependency model. The dynamic correlation coefficients between test indicators within different time windows are calculated using the time-series dependency model, the collaborative change patterns of indicator values ​​are analyzed, and a time-series correlation matrix is ​​generated. The time-series correlation matrix is ​​subjected to eigenvalue decomposition to extract the principal feature vectors, and the principal feature vectors are combined to form a feature vector set that reflects the dynamic evolution law between indicators.

[0007] The time-series dependency model is used to calculate the dynamic correlation coefficients between test indicators within different time windows, analyze the collaborative change patterns of indicator values, and generate a time-series correlation matrix, including: Based on the time-series dependency model, the values ​​of the test indicators within each time window are normalized, the correlation between the test indicators is calculated, the growth trend of the indicator values ​​is obtained, and the change characteristics reflecting the trend of indicator changes are established. A time-series analysis was performed on the changing trends of the test indicators to identify the collaborative change characteristics of the test indicators and to construct a time-series correlation matrix.

[0008] The feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifiers corresponding to the abnormal patterns are extracted, including: The feature vector set is input into the anomaly pattern recognition algorithm to calculate the similarity between the feature vector set and each record in the anomaly pattern knowledge base, generate a similarity distribution sequence, and select the anomaly pattern with the highest similarity from the similarity distribution sequence as the matching result. The matching results are used to determine the risk level. Based on the risk level identifiers recorded in the abnormal pattern knowledge base, the interval thresholds are divided, the similarity distribution sequence is mapped to the risk level intervals, and the risk level identifiers corresponding to the current test data are extracted.

[0009] Based on the risk level identifier and the rate of change of indicators in the feature vector set, a dynamic early warning index reflecting the evolution trend of abnormal patterns is calculated, including: By combining risk level identifiers with the rate of change of indicators, the changing trends of test indicators under different risk levels are calculated, and evolutionary characteristics reflecting the development process of abnormal patterns are established. Based on the evolutionary characteristics, a dynamic early warning index is calculated, which reflects the changing trend and development direction of the risk of venous thrombosis.

[0010] When the dynamic early warning index meets the early warning triggering conditions, an early warning message is generated that includes the abnormal pattern type, the relevant indicator combination, and intervention suggestions. This early warning message is then pushed to the medical decision-making terminal, including: Determine whether the dynamic early warning index meets the early warning triggering conditions. When the early warning triggering conditions are met, identify the abnormal pattern type that caused the early warning to be triggered, extract the indicator combination related to the abnormal pattern type, and generate corresponding intervention suggestions based on the abnormal pattern type and the indicator combination. The abnormal pattern type, the indicator combination, and the intervention suggestion are integrated into early warning information. A data communication link is established between the early warning information and the medical decision-making terminal, and the early warning information is pushed to the medical decision-making terminal through the data communication link.

[0011] A second aspect of the present invention provides an intelligent identification and early warning system for abnormal patterns in venous thrombosis test data, comprising: The first module is used to acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of indicators. The second module is used to match the feature vector set with a preset abnormal pattern knowledge base, identify abnormal patterns related to the risk of venous thrombosis, and extract the risk level identifier corresponding to the abnormal pattern. The third module is used to calculate a dynamic early warning index that reflects the evolution trend of abnormal patterns based on the rate of change of the indicators in the risk level identifier and the feature vector set. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. The fourth module is used to generate early warning information containing the abnormal pattern type, the combination of related indicators, and intervention suggestions when the dynamic early warning index meets the early warning triggering conditions, and to push the early warning information to the medical decision-making terminal. The fifth module is used to collect clinical intervention result data fed back from the medical decision-making terminal, and to adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

[0012] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] The beneficial effects of this application are as follows: By introducing a time-series dependency model to perform multi-dimensional correlation analysis on venous thrombosis-related test data and constructing a set of feature vectors, we can comprehensively capture the dynamic evolution patterns between indicators and overcome the limitations of traditional single-indicator evaluation methods that cannot reflect the interaction between indicators.

[0015] The calculation method based on the dynamic early warning index characterizes the urgency of risk by quantifying the rate and direction of deviation of the indicator from the benchmark value, and realizes the accurate assessment of the risk of venous thrombosis. Compared with the traditional static threshold judgment method, it can identify potential risks earlier and improve the sensitivity and specificity of early warning.

[0016] The generated early warning information includes abnormal pattern types, related indicator combinations, and intervention suggestions, providing more targeted reference for medical decision-making and helping clinicians to take timely and effective intervention measures.

[0017] By collecting clinical intervention results data feedback mechanism, the model and early warning rules are adaptively adjusted, enabling the system to continuously learn and optimize, adapt to the characteristics of different patient groups, improve the accuracy of early warning while reducing the false alarm rate. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent identification and early warning method for abnormal patterns in venous thrombosis test data according to an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the intelligent identification and early warning method for abnormal patterns in venous thrombosis test data according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of the indicators. The feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifier corresponding to the abnormal pattern is extracted. Based on the risk level identifier and the rate of change of indicators in the feature vector set, a dynamic early warning index reflecting the evolution trend of abnormal patterns is calculated. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. When the dynamic early warning index meets the early warning triggering conditions, an early warning message containing the abnormal pattern type, the combination of related indicators, and intervention suggestions is generated and pushed to the medical decision-making terminal. Collect clinical intervention result data from medical decision-making terminals, and adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

[0022] In one optional implementation, the venous thrombosis-related test data are analyzed in multiple dimensions based on a time-series dependency model to construct a feature vector set reflecting the dynamic evolution of the indicators, including: The venous thrombosis-related test data are divided into multiple continuous time windows according to the time series. Time series characteristic parameters are calculated for the test data in each time window. The time series characteristic parameters are the index mean, standard deviation and coefficient of variation. Based on the time series characteristic parameters, a recurrent neural network is trained to establish a time series dependency model. The dynamic correlation coefficients between test indicators within different time windows are calculated using the time-series dependency model, the collaborative change patterns of indicator values ​​are analyzed, and a time-series correlation matrix is ​​generated. The time-series correlation matrix is ​​subjected to eigenvalue decomposition to extract the principal feature vectors, and the principal feature vectors are combined to form a feature vector set that reflects the dynamic evolution law between indicators.

[0023] Patient data related to venous thrombosis were divided into multiple consecutive time windows based on the collection time. Taking a patient with deep vein thrombosis as an example, data on indicators such as D-dimer, platelet count, fibrinogen, and prothrombin time were collected during treatment. Data from the first 30 days after admission were divided into six 5-day time windows. Each time window contained 1-3 test results, depending on the clinical testing frequency.

[0024] When dividing venous thrombosis-related test data into multiple consecutive time windows, all test records of the target patient are extracted from the time-series database, and the records are arranged in ascending order of timestamps. The time window length is set to 24 hours, and windows are divided using a sliding method with a sliding step size of 6 hours to ensure that adjacent windows have an 18-hour overlap. Each time window is assigned a unique identifier, which is generated by concatenating the patient identifier, the window start timestamp, and the window end timestamp. When the number of test data records in a window is less than 3, the window is marked as having insufficient data and is not included in subsequent calculations. The test data in the window includes multiple indicators such as prothrombin time, activated partial thromboplastin time, fibrinogen concentration, D-dimer concentration, platelet count, white blood cell count, and hemoglobin concentration.

[0025] When calculating time-series characteristic parameters for the test data within each time window, statistical calculations are performed separately for each test indicator. The indicator mean is obtained by summing all test values ​​for that indicator within the window and dividing by the number of tests. The indicator standard deviation is obtained by calculating the square root of the sum of the squares of the differences between each test value within the window and the mean, dividing by the number of tests minus 1, and then taking the square root. The indicator coefficient of variation is obtained by dividing the standard deviation by the mean. When the absolute value of the mean is less than 0.01 times the width of the reference range, the coefficient of variation is marked as invalid. The time-series characteristic parameters are stored as a three-dimensional tensor structure. The first dimension corresponds to the number of time windows, the second dimension corresponds to the number of test indicator types, and the third dimension is fixed at 3, representing the mean, standard deviation, and coefficient of variation. When there are no test records for a certain indicator within a window, the time-series characteristic parameters corresponding to that indicator are filled with the parameter values ​​from the previous valid window.

[0026] When training a recurrent neural network based on temporal feature parameters to establish a temporal dependency model, the network structure uses long short-term memory (LSM) units as the basic computational units. The network input layer receives a tensor of temporal feature parameters, which is expanded along the time dimension, with each time step corresponding to all indicator features of one time window. The network contains three LSM layers: the first layer contains 128 hidden units, the second layer contains 64 hidden units, and the third layer contains 32 hidden units. The training process uses backpropagation to update network parameters, with mean squared error as the loss function. The optimizer uses adaptive moment estimation (IME), with a learning rate of 0.001, a batch size of 32 samples, and 100 training epochs. The training dataset contains historical patient test data sequences and corresponding thrombotic event labels. The labels are binary variables: 1 indicates that a venous thrombotic event occurred within a subsequent time window, and 0 indicates that it did not occur. The dataset is divided into training and validation sets in an 8:2 ratio. Training is terminated early when the validation set loss function fails to decrease for 10 consecutive epochs.

[0027] When calculating the dynamic correlation coefficient between test indicators within different time windows using a time-series dependency model, the time-series feature parameters of the target patient are input into a trained recurrent neural network. The hidden state vector for each time step is extracted from the network's third long short-term memory layer, with a vector dimension of 32. For any two test indicators, their mean sequences for each time window are extracted. The mean sequence is then multiplied by the hidden state vector at the corresponding time step to obtain the correlation strength sequence between the indicator and the hidden state. The Pearson correlation coefficient between the correlation strength sequences of the two indicators is calculated by dividing the covariance of the two sequences by the product of their standard deviations. The correlation coefficient ranges from -1 to +1, with a value closer to 1 indicating a stronger correlation between the indicators. When analyzing the co-variance patterns of indicator values, peaks and troughs in the correlation coefficient sequence are identified. A peak is defined as a correlation coefficient greater than 0.7 with an increase of more than 0.2 relative to the preceding and following windows, and a trough is defined as a correlation coefficient less than -0.5 with a decrease of more than 0.2 relative to the preceding and following windows.

[0028] When generating the time-series correlation matrix, the matrix dimension is set to the product of the number of test indicators and the number of time windows. The first and second dimension indices of the matrix elements correspond to the numbers of the two indicators, and the third dimension index corresponds to the time window index. The element value is the dynamic correlation coefficient of the corresponding indicator pair within the corresponding window. The time-series correlation matrix is ​​stored in a sparse storage format, storing only matrix elements with an absolute correlation coefficient greater than 0.3. After the matrix is ​​constructed, a symmetry test is performed to verify whether the element value at row index i and column index j is consistent with the element value at row index j and column index i. If they are inconsistent, the average of the two is taken as the final value. For adjacent time windows, the variation in the correlation coefficient of the same indicator pair should not exceed 0.5. If it does, the correlation coefficient sequence of the indicator pair is smoothed by median filtering, with the filtering window length set to 3 time windows.

[0029] When performing eigenvalue decomposition on the time-series correlation matrix, a two-dimensional correlation submatrix corresponding to each time window is extracted. The dimension of the submatrix is ​​the product of the number of indicators. Singular value decomposition is then performed on the submatrix, yielding a left singular vector matrix, a singular value diagonal matrix, and a right singular vector matrix. The singular values ​​are sorted in descending order, and the top few singular values ​​with a cumulative contribution rate reaching 85% are selected. The cumulative contribution rate is calculated by dividing the sum of squares of the first k singular values ​​by the sum of squares of all singular values. The left singular vector corresponding to the selected singular values ​​is extracted as the principal feature vector, and the dimension of each principal feature vector is equal to the number of indicators. Each component in the principal feature vector represents the weight of the corresponding test indicator in that feature direction; the larger the absolute value of the weight, the more significant the contribution of the indicator in that feature direction.

[0030] When combining principal feature vectors to form a feature vector set reflecting the dynamic evolution of indicators, the principal feature vectors extracted from different time windows are aligned and concatenated. The alignment process is achieved by calculating the cosine similarity between the principal feature vectors of adjacent windows, which is obtained by calculating the dot product of two vectors and dividing by the product of their magnitudes. The cosine similarity is calculated between the i-th principal feature vector of the current window and each principal feature vector of the previous window, and the vector with the highest similarity is selected as the corresponding vector. The concatenation process arranges the aligned principal feature vectors in chronological order to form a two-dimensional feature matrix. The number of rows in the matrix corresponds to the number of time windows, and the number of columns is the dimension of the principal feature vectors multiplied by the number of retained principal features. The feature matrix is ​​then normalized by subtracting the minimum value from each column and dividing by the difference between the maximum and minimum values ​​in that column, mapping the values ​​to the interval between 0 and 1.

[0031] In one optional implementation, the dynamic correlation coefficient between test indicators within different time windows is calculated using the time-series dependency model, the collaborative change pattern of indicator values ​​is analyzed, and a time-series correlation matrix is ​​generated, including: Based on the time-series dependency model, the values ​​of the test indicators within each time window are normalized, the correlation between the test indicators is calculated, the growth trend of the indicator values ​​is obtained, and the change characteristics reflecting the trend of indicator changes are established. A time-series analysis was performed on the changing trends of the test indicators to identify the collaborative change characteristics of the test indicators and to construct a time-series correlation matrix.

[0032] The dynamic correlation coefficients between test indicators within different time windows are calculated using a time-series dependency model to obtain time-series data for the test indicators. This data comes from multiple testing devices or sites and records the numerical changes of each test indicator over continuous time periods.

[0033] After obtaining the raw time-series data, preprocessing is required, including missing value handling, outlier detection and removal, and data smoothing, to ensure data quality for subsequent analysis. For example, missing values ​​can be filled using interpolation methods; outliers can be identified and replaced with reasonable values ​​by setting thresholds; and data smoothing can use methods such as moving averages to reduce the impact of noise.

[0034] When normalizing the values ​​of the test indicators within each time window based on the time-series dependency model, the first step is to determine an appropriate time window size. The selection of the time window should be based on the data characteristics and the purpose of the analysis; it can be a fixed size (such as hourly, daily, or weekly) or a sliding window. For environmental monitoring scenarios, a 24-hour window is selected to capture the daily variation patterns of the indicators.

[0035] Within each defined time window, the test index data are normalized to eliminate the influence of dimensions. Normalization methods can include maximum-minimum normalization or Z-score standardization. Taking maximum-minimum normalization as an example, for a test index X within a time window, its normalized value X' is calculated as follows: subtract the minimum value within that window from X, and then divide by the difference between the maximum and minimum values, so that all index values ​​are mapped to the interval between 0 and 1.

[0036] After normalization, the correlation coefficients between different test indicators within each time window are calculated to reflect the degree of correlation between the indicators. Methods such as Pearson correlation coefficient, Spearman's rank correlation coefficient, or Kendall's rank correlation coefficient can be used. In water quality monitoring applications, if it is necessary to analyze the relationship between pH value and dissolved oxygen content, the correlation coefficients of these two indicators within each time window are calculated to form a correlation coefficient sequence that changes over time.

[0037] To obtain the growth trend of indicator values, the first difference or growth rate is calculated on the normalized data. The first difference directly reflects the amount of change between adjacent time points, while the growth rate represents the relative degree of change. In air quality monitoring, the first difference between PM2.5 and nitrogen dioxide concentrations can be calculated to understand whether their concentration changes have similar trends.

[0038] For the calculated growth trend data, features are extracted to quantify the changing trends. These features may include trend slope, fluctuation range, and periodicity. For example, linear regression can be used to calculate the trend slope, standard deviation can be used to measure the fluctuation range, and Fourier analysis can be used to identify periodicity.

[0039] When conducting time-series analysis on the changing trends of inspection indicators, the time series data is first decomposed into trend components, seasonal components, and random components. Methods such as STL decomposition and X-12-ARIMA can be used. In industrial production quality monitoring, after decomposing the time series of various product quality indicators, it is possible to clearly identify which changes are due to long-term trends, which are due to seasonal factors, and which are caused by random fluctuations.

[0040] Based on the decomposition results, the co-variation characteristics of the test indicators are identified, including patterns such as synchronous growth, synchronous decrease, and inverse changes. The degree of co-variation is quantified by calculating the correlation between trend terms of different indicators and the phase difference of seasonal terms. In meteorological data analysis, it can be found that temperature and humidity indicators exhibit inverse variation characteristics in summer, while showing a weak correlation or no correlation in winter.

[0041] When constructing the time-series correlation matrix, the correlation coefficients of each pair of test indicators in each time window are organized into a matrix form. Each element in the matrix represents the correlation coefficient between the two indicators within a specific time window. For n test indicators, an n×n symmetric matrix is ​​formed within each time window, with the diagonal elements being 1. As the time windows slide, a series of correlation matrices are formed, reflecting the dynamic evolution of the relationship between the indicators.

[0042] To enhance the interpretability of time-series correlation matrices, matrix elements can be visualized using color coding, and heatmaps can be used to display the strength of relationships between indicators. Cluster analysis can also be used to identify groups of indicators with similar patterns of change. In environmental monitoring data analysis, visualizing time-series correlation matrices provides a clear view of the changing correlations between different pollutants, helping to identify potential pollution sources.

[0043] Time-series correlation matrix analysis results can be used in various scenarios, such as anomaly detection, change point discovery, and causal relationship inference. When the correlation between two indicators suddenly changes significantly, it signifies a shift in the system state or the emergence of external disturbances. By tracking the changing patterns of the time-series correlation matrix, the system state can be monitored in real time, potential problems can be identified early, and data support can be provided for decision-making.

[0044] In one optional implementation, the feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifiers corresponding to the abnormal patterns are extracted, including: The feature vector set is input into the anomaly pattern recognition algorithm to calculate the similarity between the feature vector set and each record in the anomaly pattern knowledge base, generate a similarity distribution sequence, and select the anomaly pattern with the highest similarity from the similarity distribution sequence as the matching result. The matching results are used to determine the risk level. Based on the risk level identifiers recorded in the abnormal pattern knowledge base, the interval thresholds are divided, the similarity distribution sequence is mapped to the risk level intervals, and the risk level identifiers corresponding to the current test data are extracted.

[0045] An abnormal pattern knowledge base is constructed, containing various abnormal pattern samples related to the risk of venous thrombosis and their corresponding risk level labels. The abnormal pattern samples can be feature vectors of typical cases extracted from a large amount of clinical data, with each sample labeled with a risk level, such as "low risk," "medium risk," "high risk," and "very high risk." Each record in the knowledge base includes a feature vector, an abnormal pattern description, and a corresponding risk level label.

[0046] The patient's feature vector set is input into an abnormal pattern recognition algorithm, which can employ a similarity calculation method based on distance metrics. Taking cosine similarity as an example, the similarity between the patient's feature vector and each abnormal pattern sample in the knowledge base is calculated: For each patient feature vector V and each abnormal pattern sample vector Mi in the knowledge base, cosine similarity is calculated. This calculation process takes into account the importance weight of each feature, assigning different weights to different clinical indicators. For example, indicators that directly reflect thrombosis risk, such as D-dimer and fibrinogen, can be assigned higher weights, while some auxiliary indicators have lower weights.

[0047] After the calculation is completed, the similarity values ​​between the patient feature vector and all abnormal pattern samples in the knowledge base are obtained, forming a similarity distribution sequence S={S1, S2, ..., Sn}, where Si represents the similarity value between the patient feature vector and the i-th abnormal pattern sample.

[0048] The system selects the anomaly pattern with the highest similarity from the similarity distribution sequence as the matching result. Let Mmax be the anomaly pattern corresponding to the highest similarity, and Smax be the similarity value. If Smax exceeds a preset matching threshold (e.g., 0.75), the match is considered successful; otherwise, the system will provide a prompt that the matching result cannot be determined and suggest that the doctor conduct further examination.

[0049] For successfully matched results, a risk level assessment is performed. Based on the risk level identifiers recorded in the abnormal pattern knowledge base, interval thresholds are defined, mapping the similarity distribution sequence to these risk level intervals. Specifically, four risk level intervals are pre-defined in the knowledge base: - Low risk: Similarity value in the range [0.75, 0.80); - Medium risk: Similarity value in the range [0.80, 0.85); - High risk: Similarity value in the range [0.85, 0.90); - Extremely high risk: Similarity value in the range of [0.90, 1.00]; By finding the range where Smax is located, the patient's current thrombosis risk level can be determined. For example, if Smax = 0.87, the patient's thrombosis risk is determined to be "high risk".

[0050] To improve matching accuracy, multiple similarity calculation methods can be used for comprehensive evaluation. In addition to cosine similarity, Euclidean distance, Mahalanobis distance, and other metrics can also be used. By weighted fusion of multiple similarity calculation results, the bias caused by a single algorithm can be reduced.

[0051] In practical applications, the patient's blood routine, coagulation function, inflammatory markers and other test data are combined into a feature vector, such as V=[10.2, 4.5, 450, 1.5, 3.8, ...], which includes indicators such as white blood cell count, red blood cell count, platelet count, D-dimer, fibrinogen and other indicators.

[0052] The similarity between the feature vector and each abnormal pattern in the knowledge base is calculated using a pre-trained abnormal pattern recognition model. Assuming there are 200 abnormal pattern samples in the knowledge base, 200 similarity values ​​are obtained after calculation, forming a similarity distribution sequence S={0.65, 0.72, 0.83, 0.79, ..., 0.91}.

[0053] The maximum value of 0.91 was found in the sequence, corresponding to the 200th abnormal pattern in the knowledge base, "hypercoagulable state with inflammatory response," which is marked as "extremely high risk" in the knowledge base. Based on the similarity value of 0.91 falling within the range of [0.90, 1.00], the patient's current risk of venous thrombosis was determined to be "extremely high risk."

[0054] To enhance the accuracy of risk assessment, time-series data analysis can be combined to perform trend analysis on multiple test data of the same patient, observing the changing trends of key indicators, such as persistently elevated D-dimer and abnormal fibrinogen. These changing trends can also serve as important references for determining the risk level.

[0055] The above-described abnormal pattern recognition and risk level assessment process can accurately evaluate a patient's risk of venous thrombosis, providing decision support for physicians to develop prevention and treatment plans. The assessment results can be further used to generate personalized treatment recommendations, such as recommending appropriate prophylactic anticoagulation therapy for patients with different risk levels.

[0056] In one optional implementation, calculating a dynamic early warning index reflecting the evolution trend of abnormal patterns based on the risk level identifier and the rate of change of indicators in the feature vector set includes: By combining risk level identifiers with the rate of change of indicators, the changing trends of test indicators under different risk levels are calculated, and evolutionary characteristics reflecting the development process of abnormal patterns are established. Based on the evolutionary characteristics, a dynamic early warning index is calculated, which reflects the changing trend and development direction of the risk of venous thrombosis.

[0057] In the process of calculating the dynamic early warning index reflecting the evolution trend of abnormal patterns based on the risk level identifier and the rate of change of indicators in the feature vector set, the risk level identifier and the rate of change of indicators are first combined and analyzed, then the changing trend of the test indicators under different risk levels is calculated, the evolution characteristics reflecting the development process of abnormal patterns are established, and finally the dynamic early warning index is calculated based on the evolution characteristics.

[0058] When combining risk level labels with the rate of change of indicators in a combined analysis, the patient's risk level label is first obtained. This risk level label can be obtained based on the aforementioned venous thrombosis risk assessment model and is typically divided into three levels: low risk, medium risk, and high risk. Simultaneously, the rate of change of each indicator in the feature vector set is calculated. The rate of change of indicators represents the trend of change of various clinical indicators of the patient within a certain time window.

[0059] For each patient's feature vector set, test index data from multiple continuously monitored time points are extracted, and the slope of the index within each time window is calculated as the rate of change. Taking D-dimer as an example, let the D-dimer value at time t be D(t). Then, the rate of change within the time window [t-Δt, t] can be expressed as [D(t)-D(t-Δt)] / Δt, where Δt is the size of the time window, usually set to 6 hours, 12 hours, or 24 hours, with an appropriate time window selected based on the clinical monitoring frequency.

[0060] By correlating the rate of change of various indicators with risk levels, an indicator-risk matrix was constructed. For each risk level, the changing trend characteristics of each test indicator were analyzed. For example, for high-risk patients, characteristics such as a sustained and rapid increase in D-dimer, a decrease in platelet count, and a shortening of thrombin time were observed; while for medium-risk patients, the rate of change of these indicators was relatively slow; and for low-risk patients, the indicators were relatively stable or fluctuated little.

[0061] By analyzing the statistical distribution of the rate of change of each indicator in patient groups with different risk levels, a characteristic pattern of indicator change for each risk level is established. Specifically, for each risk level k (k=1, 2, 3, representing low, medium, and high risk, respectively), the average value μ_i^k and standard deviation σ_i^k of the rate of change of each indicator i under that risk level are calculated to form a characteristic distribution description of that risk level.

[0062] Based on the aforementioned characteristic distributions, an evolutionary feature reflecting the development process of abnormal patterns is established. This evolutionary feature mainly considers two aspects: first, the degree of matching between the current rate of change of indicators and the characteristic distributions of each risk level; and second, the time-series trend of the rate of change of indicators. For the degree of matching between the rate of change of indicators and the characteristic distributions of risk levels, distance metrics or similarity functions can be used to calculate the closeness between the current patient's rate of change vector and the typical patterns of each risk level.

[0063] Specifically, the Mahalanobis or Euclidean distance between the rate of change vector V of each indicator in the current patient feature vector set and the feature distribution of each risk level k is calculated to obtain the matching score S_k. The lower the matching score, the closer the patient is to the feature distribution of that risk level. Simultaneously, the time series trend of the rate of change of the indicators is examined to analyze whether it is evolving towards a higher or lower risk level.

[0064] For time series trend analysis, a sliding time window approach is used to compare changes in the matching degree between the patient's feature vector and the feature distribution of each risk level within adjacent time windows. If the matching degree shifts towards a higher risk level, it indicates that the patient's condition is deteriorating; conversely, it indicates improvement. A trend factor T is defined to quantify the direction and speed of this evolution.

[0065] The Dynamic Warning Index (DWI) is calculated based on the matching score S_k and the trend factor T. The DWI comprehensively considers the current risk level, the matching degree with each risk level, and the evolution trend, and can be expressed as a weighted combination. Specifically, the calculation method combines the base score of the current risk level with a weighted combination of the matching score and the trend factor.

[0066] For example, assuming the current patient's risk level is k and the baseline score is B_k, the dynamic early warning index can be calculated as: DWI = B_k + w1·S_k + w2·T, where w1 and w2 are weighting coefficients determined through clinical data validation. A higher DWI value indicates a higher risk of venous thrombosis, and a more pronounced upward trend.

[0067] The calculation results of the dynamic early warning index are visualized, forming a trend curve that changes over time, intuitively reflecting the changing trend and development direction of venous thrombosis risk. When the DWI exceeds the preset threshold or shows a sharp upward trend, the system can trigger an early warning signal to remind medical staff to take intervention measures.

[0068] In practical applications, for continuous monitoring data of hospitalized patients, DWI calculation results are updated at fixed time intervals (such as 4 hours or 8 hours). By tracking the changing trends of DWI, medical staff can identify patients with an increased risk of venous thrombosis early and achieve proactive intervention.

[0069] Furthermore, the dynamic early warning index can be analyzed in conjunction with patients' clinical intervention records to assess the effectiveness of different interventions in reducing the risk of venous thrombosis. For example, after a patient receives anticoagulation therapy, the treatment effect can be assessed by observing changes in DWI (dense venous thrombosis weight). If the DWI value decreases significantly and tends to stabilize, it indicates that the intervention is effective; if the DWI continues to rise, the treatment plan needs to be adjusted.

[0070] In summary, by combining risk level identifiers with the rate of change of indicators, the evolutionary characteristics of abnormal patterns are established, and a dynamic early warning index is calculated based on this. This enables precise monitoring and prospective early warning of the changing trends in the risk of venous thrombosis, providing an important decision support tool for the clinical prevention and treatment of venous thrombosis.

[0071] In one optional implementation, when the dynamic early warning index meets the early warning triggering conditions, an early warning message is generated that includes the abnormal pattern type, the relevant indicator combination, and intervention suggestions, and the early warning message is pushed to the medical decision-making terminal, including: Determine whether the dynamic early warning index meets the early warning triggering conditions. When the early warning triggering conditions are met, identify the abnormal pattern type that caused the early warning to be triggered, extract the indicator combination related to the abnormal pattern type, and generate corresponding intervention suggestions based on the abnormal pattern type and the indicator combination. The abnormal pattern type, the indicator combination, and the intervention suggestion are integrated into early warning information. A data communication link is established between the early warning information and the medical decision-making terminal, and the early warning information is pushed to the medical decision-making terminal through the data communication link.

[0072] In a preferred embodiment, the patient's physiological parameters, clinical records, and medical device data are first acquired. This data includes basic physiological indicators such as heart rate, blood pressure, body temperature, blood oxygen saturation, and respiratory rate, as well as clinical information such as various laboratory test results, medication records, and treatment records. The received data undergoes data preprocessing steps, including outlier removal, missing data supplementation, and data format standardization, to ensure data quality meets the requirements of subsequent analysis.

[0073] Feature extraction was performed on the preprocessed data, selecting key indicators closely related to the patient's condition as feature variables, such as the trend of body temperature changes, the amplitude of blood pressure fluctuations, and the rate of decrease in blood oxygen saturation. Based on the extracted feature set, a dynamic early warning model was constructed. This model adopts a multi-layer neural network structure, with the input layer receiving various feature data, the hidden layer undergoing nonlinear transformation through activation functions, and the output layer generating a comprehensive early warning index. The model was trained using historical patient data, including known adverse event cases and normal cases, and the model parameters were optimized using the backpropagation algorithm.

[0074] Once the dynamic early warning model has completed training, real-time patient data will be input into the model to calculate the dynamic early warning index. The early warning index uses a quantitative scale of 0-100, where 0 represents a normal state and 100 represents an extremely high-risk state. The early warning triggering conditions include two aspects: first, the index value exceeds a preset threshold (e.g., greater than 85) in a single instance; second, the index rises rapidly within a short period of time (e.g., an increase of more than 25 within 30 minutes).

[0075] When determining whether the dynamic early warning index meets the early warning triggering conditions, a dual threshold comparison method is used. The method checks whether the latest calculated early warning index exceeds the absolute threshold (e.g., 85 points). If it does, the triggering conditions are immediately met. If it does not exceed the absolute threshold, the relative change is checked, and the difference between the current index and the index of the previous 30 minutes is calculated. If the difference exceeds the set relative threshold (e.g., 25 points), the triggering conditions are also met.

[0076] Once the warning trigger conditions are confirmed, an abnormal pattern recognition algorithm identifies the type of abnormal pattern that caused the warning. Abnormal pattern types include, but are not limited to: cardiovascular failure pattern, respiratory dysfunction pattern, infection exacerbation pattern, and organ failure pattern. The identification process uses a combination of decision trees and rule bases. By analyzing the main abnormal indicators that trigger the warning and their combined characteristics, the algorithm determines the most matching abnormal pattern type. For example, when a combination of features such as a sharp drop in blood pressure, an increase in heart rate, and a decrease in peripheral vascular resistance is detected, it is identified as a cardiovascular failure pattern.

[0077] After identifying the abnormal pattern type, the system extracts the indicator combination related to that abnormal pattern. An indicator combination refers to a set of multiple relevant clinical indicators that jointly cause or reflect the abnormal pattern. For example, for a respiratory dysfunction pattern, the relevant indicator combination includes blood oxygen saturation, respiratory rate, tidal volume, and partial pressure of carbon dioxide. The extraction process utilizes a pre-established abnormal pattern-indicator correlation matrix; based on the identified abnormal pattern type, the matrix is ​​queried to obtain the corresponding indicator combination.

[0078] Based on the identified abnormal pattern types and extracted indicator combinations, corresponding intervention recommendations are generated. This generation employs clinical pathway reasoning technology, combined with medical knowledge graphs and clinical practice guidelines. For each abnormal pattern type, a corresponding intervention decision tree is constructed. Personalized intervention recommendations are generated by traversing the decision tree based on the specific abnormalities in the indicator combinations. For example, for an infection worsening pattern, if the indicator combination shows elevated body temperature, increased white blood cell count, and elevated procalcitonin, intervention recommendations might include adjusting the antibiotic regimen, increasing the antibiotic dosage or using combination therapy, and strengthening the search for infection foci.

[0079] Abnormal pattern types, indicator combinations, and intervention recommendations are integrated into structured early warning information. This information is encapsulated in JSON format and includes fields such as early warning level, early warning time, abnormal pattern type, details of abnormal indicators (indicator name, outlier, reference range, trend, etc.), and a list of intervention recommendations. The structure of the early warning information follows medical information standards and specifications to ensure information completeness and interpretability.

[0080] To ensure timely transmission of early warning information, a data communication link is established between the early warning information and the medical decision-making terminal. This link uses an encrypted WebSocket protocol and supports real-time bidirectional communication. Terminal authentication and authorization are performed first during link establishment to verify the legitimacy and access levels of the terminal devices, ensuring information security. End-to-end encryption is implemented throughout the communication process to prevent the leakage of sensitive medical information.

[0081] Early warning information is pushed to the medical decision-making terminal via the established data communication link. The push process employs a message queue mechanism to ensure the reliability and real-time nature of message transmission. Upon receiving the early warning information, the medical decision-making terminal immediately displays a notification and shows the warning details on the medical workstation interface, automatically recording the warning event. Important early warning information is also simultaneously notified to relevant medical staff through multiple channels, including SMS and mobile application push notifications, ensuring timely processing of the warning information.

[0082] After the early warning information is pushed out, a complete log of the early warning event is recorded, including the early warning trigger time, abnormal pattern type, relevant indicator data, intervention suggestions, and the target audience, providing data support for subsequent early warning effect evaluation and model optimization. Simultaneously, an early warning tracking mechanism is activated to monitor changes in patient status and the implementation of medical interventions after the early warning, forming a complete early warning-intervention-effect closed loop.

[0083] The method further includes: The acquisition of venous thrombosis-related test data for target patients is achieved through a data interface with the hospital's laboratory information system. This interface employs an asynchronous communication mechanism based on a message queue. Once the laboratory information system completes its review of the test report, a data push event is triggered, pushing a structured message containing the patient identifier, test item code, test result value, test timestamp, and upper and lower limits of the reference range to the message queue. The data receiving module pulls messages from the message queue, parses them, and stores them in a time-series database. Each record contains a unique patient identifier field, a test indicator type field, a numerical field, a unit field, and a timestamp field. The test data includes coagulation function indicators and clinical characterization data. Coagulation function indicators cover prothrombin time, activated partial thromboplastin time, fibrinogen concentration, and D-dimer concentration. Clinical characterization data includes platelet count, white blood cell count, and hemoglobin concentration. The data receiving module performs integrity checks on the received data, checking for missing required fields and whether the values ​​are within physiologically reasonable ranges. For data that passes the checks, an index is created based on the patient identifier and the test timestamp.

[0084] When performing multi-dimensional correlation analysis on test data based on a time-series dependency model, all test records of the target patient within a specified time window are extracted from the time-series database. The time window length is set according to the clinical significance of different test indicators; a 72-hour window is used for coagulation function indicators, and a 168-hour window is used for clinical characterization data. The time-series dependency model uses a directed graph structure to represent the causal relationships between indicators. Nodes in the graph represent individual test indicators, and edges represent the time-series dependencies and influence strength between indicators. For the test data of the target patient, after arranging them in chronological order, the rate of change of each indicator relative to the previous test result is calculated. The rate of change is defined as the difference between the current value and the previous value divided by the previous value. Simultaneously, the deviation of the indicator value from the median value of the reference range is calculated. The deviation is defined as the difference between the current value and the median value of the reference range divided by the width of the reference range. For indicator pairs with time-series dependencies, the correlation strength between the rate of change of the preceding indicator and the deviation of the subsequent indicator is calculated. When constructing the feature vector set, each feature vector contains indicator trend characteristics and inter-indicator coupling characteristics. Indicator trend characteristics include the average rate of change, maximum rate of change, and consistency of change direction within the time window. The coupling and correlation characteristics between indicators include the correlation strength, correlation direction, and time delay of indicator pairs with significant temporal dependencies.

[0085] The feature vector set is matched with a pre-defined anomaly pattern knowledge base using a pattern similarity calculation method. The anomaly pattern knowledge base is stored as a rule set, where each rule defines a feature vector combination pattern related to the risk of venous thrombosis. Each rule includes feature constraints, threshold conditions, and combinational logic. The matching process iterates through each rule in the knowledge base, comparing the corresponding feature value in the feature vector set with the constraints in the rule. The rule is activated when the feature value satisfies the constraints and the combinational logic is true. Each rule is associated with a risk level identifier, represented by an enumeration of values, including low risk, medium risk, high risk, and extremely high risk. When multiple rules are activated simultaneously, the highest risk level is taken as the risk level identifier corresponding to the anomaly pattern.

[0086] The dynamic early warning index is calculated based on the rate of change of indicators in the risk level identifier and feature vector set. The rate of change of an indicator is defined as the change in the value of the indicator per unit time, and the slope value is obtained by linear regression fitting of the test data within the time window. The risk level identifier is mapped to the basic weight coefficient, with a weight coefficient of 0.2 for low risk, 0.5 for medium risk, 0.8 for high risk, and 1.0 for very high risk. The rate of change of the indicator is normalized to the interval between 0 and 1 by dividing the original rate value by the maximum rate value observed for that indicator in the historical patient population. The dynamic early warning index is calculated by a weighted combination of the basic weight coefficient and the normalized rate, with the basic weight coefficient accounting for 60% and the normalized rate accounting for 40%. When the test data contains multiple abnormal indicators, the dynamic early warning indices corresponding to each indicator are aggregated by calculating the weighted average of the early warning indices of all indicators. The weight of coagulation function indicators is set to 2.0, and the weight of clinical characterization data is set to 1.0.

[0087] A warning message is generated when the dynamic warning index meets the warning trigger conditions. These trigger conditions are set based on the threshold of the dynamic warning index, which is divided into three levels: 0.3 for low-level warning, 0.6 for medium-level warning, and 0.8 for high-level warning. The warning message includes the abnormal pattern type, the combination of involved indicators, and intervention recommendations. The abnormal pattern type is determined by the activated rule identifier, and the combination of involved indicators lists all test indicator names and current values ​​that triggered the warning. The intervention recommendations retrieve predefined clinical treatment protocols from the knowledge base based on the abnormal pattern type and risk level. The warning message is pushed to the medical decision-making terminal in a structured message format, which includes the patient identifier, warning level, generation timestamp, abnormal pattern description, indicator data list, and intervention recommendation text.

[0088] Clinical intervention outcome data collected from the medical decision-making terminal is implemented through a feedback interface. This feedback data includes the type of intervention, execution time, patient outcome, and subsequent changes in test results. Based on the clinical intervention outcome data, the calculation rules for the time-series dependency model and the dynamic early warning index are adaptively adjusted. The adjustment mechanism analyzes the correlation between early warning accuracy and intervention effectiveness, statistically analyzing the early warning accuracy rate over a certain period. When the accuracy rate falls below a set target value, the model adjustment process is triggered. The adjustment of the time-series dependency model is achieved by recalculating the correlation strength between indicators, adding recently accumulated test data and intervention outcome data to the training set, and updating the weight values ​​of the edges in the directed graph. The adjustment of the dynamic early warning index calculation rules is achieved by modifying the basic weight coefficients and rate normalization parameters, with the adjustment range controlled within ±10% of the original values.

[0089] A second aspect of the present invention provides an intelligent identification and early warning system for abnormal patterns in venous thrombosis test data, comprising: The first module is used to acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of indicators. The second module is used to match the feature vector set with a preset abnormal pattern knowledge base, identify abnormal patterns related to the risk of venous thrombosis, and extract the risk level identifier corresponding to the abnormal pattern. The third module is used to calculate a dynamic early warning index that reflects the evolution trend of abnormal patterns based on the rate of change of the indicators in the risk level identifier and the feature vector set. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. The fourth module is used to generate early warning information containing the abnormal pattern type, the combination of related indicators, and intervention suggestions when the dynamic early warning index meets the early warning triggering conditions, and to push the early warning information to the medical decision-making terminal. The fifth module is used to collect clinical intervention result data fed back from the medical decision-making terminal, and to adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

[0090] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0091] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0092] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent identification and early warning of abnormal patterns in venous thrombosis test data, characterized in that, include: Acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of the indicators. The feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifier corresponding to the abnormal pattern is extracted. Based on the risk level identifier and the rate of change of indicators in the feature vector set, a dynamic early warning index reflecting the evolution trend of abnormal patterns is calculated. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. When the dynamic early warning index meets the early warning triggering conditions, an early warning message containing the abnormal pattern type, the combination of related indicators, and intervention suggestions is generated and pushed to the medical decision-making terminal. Collect clinical intervention result data from medical decision-making terminals, and adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

2. The method according to claim 1, characterized in that, Based on a time-series dependency model, multi-dimensional correlation analysis was performed on the venous thrombosis-related test data to construct a feature vector set reflecting the dynamic evolution of the indicators, including: The venous thrombosis-related test data are divided into multiple continuous time windows according to the time series. Time series characteristic parameters are calculated for the test data in each time window. The time series characteristic parameters are the index mean, standard deviation and coefficient of variation. Based on the time series characteristic parameters, a recurrent neural network is trained to establish a time series dependency model. The dynamic correlation coefficients between test indicators within different time windows are calculated using the time-series dependency model, the collaborative change patterns of indicator values ​​are analyzed, and a time-series correlation matrix is ​​generated. The time-series correlation matrix is ​​subjected to eigenvalue decomposition to extract the principal feature vectors, and the principal feature vectors are combined to form a feature vector set that reflects the dynamic evolution law between indicators.

3. The method according to claim 2, characterized in that, The time-series dependency model is used to calculate the dynamic correlation coefficients between test indicators within different time windows, analyze the collaborative change patterns of indicator values, and generate a time-series correlation matrix, including: Based on the time-series dependency model, the values ​​of the test indicators within each time window are normalized, the correlation between the test indicators is calculated, the growth trend of the indicator values ​​is obtained, and the change characteristics reflecting the trend of indicator changes are established. A time-series analysis was performed on the changing trends of the test indicators to identify the collaborative change characteristics of the test indicators and to construct a time-series correlation matrix.

4. The method according to claim 1, characterized in that, The feature vector set is matched with a preset abnormal pattern knowledge base to identify abnormal patterns related to the risk of venous thrombosis, and the risk level identifiers corresponding to the abnormal patterns are extracted, including: The feature vector set is input into the anomaly pattern recognition algorithm to calculate the similarity between the feature vector set and each record in the anomaly pattern knowledge base, generate a similarity distribution sequence, and select the anomaly pattern with the highest similarity from the similarity distribution sequence as the matching result. The matching results are used to determine the risk level. Based on the risk level identifiers recorded in the abnormal pattern knowledge base, the interval thresholds are divided, the similarity distribution sequence is mapped to the risk level intervals, and the risk level identifiers corresponding to the current test data are extracted.

5. The method according to claim 1, characterized in that, Based on the risk level identifier and the rate of change of indicators in the feature vector set, a dynamic early warning index reflecting the evolution trend of abnormal patterns is calculated, including: By combining risk level identifiers with the rate of change of indicators, the changing trends of test indicators under different risk levels are calculated, and evolutionary characteristics reflecting the development process of abnormal patterns are established. Based on the evolutionary characteristics, a dynamic early warning index is calculated, which reflects the changing trend and development direction of the risk of venous thrombosis.

6. The method according to claim 1, characterized in that, When the dynamic early warning index meets the early warning triggering conditions, an early warning message is generated that includes the abnormal pattern type, the relevant indicator combination, and intervention suggestions. This early warning message is then pushed to the medical decision-making terminal, including: Determine whether the dynamic early warning index meets the early warning triggering conditions. When the early warning triggering conditions are met, identify the abnormal pattern type that caused the early warning to be triggered, extract the indicator combination related to the abnormal pattern type, and generate corresponding intervention suggestions based on the abnormal pattern type and the indicator combination. The abnormal pattern type, the indicator combination, and the intervention suggestion are integrated into early warning information. A data communication link is established between the early warning information and the medical decision-making terminal, and the early warning information is pushed to the medical decision-making terminal through the data communication link.

7. An intelligent identification and early warning system for abnormal patterns in venous thrombosis test data, used to implement the method described in any one of claims 1-6, characterized in that, include: The first module is used to acquire venous thrombosis-related test data of the target patient, perform multi-dimensional correlation analysis on the venous thrombosis-related test data based on the time-series dependency model, and construct a set of feature vectors that reflect the dynamic evolution of indicators. The second module is used to match the feature vector set with a preset abnormal pattern knowledge base, identify abnormal patterns related to the risk of venous thrombosis, and extract the risk level identifier corresponding to the abnormal pattern. The third module is used to calculate a dynamic early warning index that reflects the evolution trend of abnormal patterns based on the rate of change of the indicators in the risk level identifier and the feature vector set. The dynamic early warning index characterizes the urgency of the risk by quantifying the rate and direction of the indicators deviating from the benchmark value. The fourth module is used to generate early warning information containing the abnormal pattern type, the combination of related indicators, and intervention suggestions when the dynamic early warning index meets the early warning triggering conditions, and to push the early warning information to the medical decision-making terminal. The fifth module is used to collect clinical intervention result data fed back from the medical decision-making terminal, and to adaptively adjust the calculation rules of the time-series dependency model and the dynamic early warning index based on the clinical intervention result data.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images