Clinical cure prediction and management method for chronic hepatitis b based on irregular time series data

Abstract

This invention relates to the field of clinical medical technology and discloses a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data. The method includes: acquiring multimodal time-series data of chronic hepatitis B patients and interpolating the multimodal time-series data to obtain continuous time-series data; performing curve fitting on the continuous time-series data to obtain cost-bending paths and time-shift feature vectors; extracting multidimensional features and fusing and encoding the multidimensional features with the time-shift feature vectors to obtain composite feature vectors; quantitatively evaluating the composite feature vectors to obtain confidence scores and assigning corresponding quality labels to the composite feature vectors; inputting the labeled composite feature vectors into a pre-trained cure probability prediction model to obtain a clinical cure tendency index; and generating individualized management decisions based on the clinical cure tendency index and a pre-set clinical pathway rule base. This invention can improve the efficiency of predicting and managing the clinical cure of chronic hepatitis B.

Description

Technical Field

[0001] This invention relates to the field of clinical medical technology, and in particular to a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data. Background Technology

[0002] In the field of predicting and managing the clinical cure of chronic hepatitis B, existing technologies have failed to effectively address the irregularities and discontinuities of multimodal time-series data. Traditional methods lack scientific interpolation processing and deep feature mining for such data, and fail to fully consider the time offset characteristics and correlation value of multidimensional indicators in the data. This makes it difficult to accurately extract the core information reflecting the patient's disease progression, resulting in insufficient reliability and accuracy of clinical cure prediction results, and failing to provide solid data support for subsequent management.

[0003] The existing clinical management model suffers from significant problems of being too rudimentary and lacks a closed-loop precision management mechanism covering the entire disease course. Treatment plans are often formulated based on single baseline indicators or individual physician experience, failing to incorporate personalized adjustments based on dynamic monitoring data during the patient's treatment process. Furthermore, the multi-source data generated during diagnosis and treatment has not been effectively integrated and structured, and its value has not been fully explored. This results in a failure to provide precise support for real-time treatment decisions or to form data assets that drive model optimization and scientific research advancement. Ultimately, this makes it difficult to effectively improve the treatment efficiency and success rate of clinical cure. Therefore, improving the treatment efficiency of clinical cure has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data, comprising: S1. Acquire multimodal time-series data of patients with chronic hepatitis B during a continuous monitoring period, and perform interpolation processing on the multimodal time-series data to obtain continuous time-series data of patients with chronic hepatitis B. S2. Perform curve fitting on the continuous time series data to obtain the cost curve path of the chronic hepatitis B patient, and quantify the time offset of the nodes in the cost curve path to obtain the time offset feature vector of the chronic hepatitis B patient. S3. Within the sliding window, extract the multidimensional features of the continuous time series data, and fuse and encode the multidimensional features with the time offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient. S4. Quantitatively evaluate the composite feature vector to obtain a confidence score for the composite feature vector, and assign a corresponding quality label to the composite feature vector based on the confidence score. S5. Input the labeled composite feature vector into the pre-trained cure probability prediction model to obtain the clinical cure tendency index of the chronic hepatitis B patient. S6. Based on the clinical cure tendency indicators and the preset clinical pathway rule base, generate individualized management decisions for the patients with chronic hepatitis B.

[0006] In a preferred embodiment, the step of interpolating the multimodal time-series data to obtain the continuous time-series data of the chronic hepatitis B patient includes: By monitoring the laboratory test value sequences, structured clinical event records, and reporting indicators of patients with chronic hepatitis B, multimodal time-series data of these patients were obtained. Discontinuous time points in the multimodal time series data are used as nodes to be completed for the patients with chronic hepatitis B. Select the valid data at adjacent time points in the node to be completed to obtain the local reference data segment of the chronic hepatitis B patient; In the time dimension, the numerical change trend of the local reference data segment is analyzed to obtain the numerical estimate of the node to be completed; The numerical estimate is inserted into the corresponding position in the multimodal time series data to obtain the continuous time series data of the chronic hepatitis B patient.

[0007] In a preferred embodiment, the step of curve fitting the continuous time-series data to obtain the cost-bending path of the chronic hepatitis B patient, and quantifying the time offset of nodes in the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient, includes: The clinical indicator dimensions in the continuous time series data are normalized to obtain the standard clinical indicator dimensions of the continuous time series data. A two-dimensional plane is constructed with time as the horizontal axis and numerical values ​​as the vertical axis for the patients with chronic hepatitis B. By mapping the standard clinical indicator dimensions onto the two-dimensional plane, the indicator curve of the chronic hepatitis B patient is obtained. Based on the proximity of adjacent values ​​on the index curve, candidate matching pairs for the chronic hepatitis B patients are determined. Path planning is performed on the candidate matching point pairs to obtain the continuous path of the chronic hepatitis B patient; Optimal spatiotemporal curvature analysis is performed on the continuous path to obtain the overall alignment cost of the continuous path, and the continuous path with the minimum overall alignment cost is taken as the cost curvature path of the chronic hepatitis B patient. The time offset of the cost-bending path is tensor-synthesized to obtain the time offset feature vector of the chronic hepatitis B patient.

[0008] In a preferred embodiment, the step of tensor synthesis of the time offset of the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient includes: Extract the actual monitoring timestamps of the matching points on the cost bending path, and record the projection position of the actual monitoring timestamps on the two-dimensional plane; The time difference between the actual monitoring timestamp and the projected position is used as the original time offset of the chronic hepatitis B patient. The original time offsets are arranged and aggregated to obtain the time offset sequence of the chronic hepatitis B patient. Calculate the overall irregularity score of the time-offset sequence; The comprehensive irregularity score is ordered and encoded with the maximum and minimum values ​​in the time offset sequence to obtain the time offset feature vector of the chronic hepatitis B patient.

[0009] In a preferred embodiment, the formula for calculating the comprehensive irregularity score is: ; in, This represents the comprehensive irregularity score. This represents the total number of values ​​in the time offset sequence. Indicates the first The original time offset. This represents the preset weighting coefficient. This represents the average value of the time offset sequence. This represents the preset positive decimal stabilization parameter. It represents the absolute value.

[0010] In a preferred embodiment, the step of extracting multidimensional features from the continuous time-series data within a sliding window, and fusing and encoding the multidimensional features with the time-offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient, includes: The continuous time-series data is divided into time-window segments to obtain time-series data segments of the continuous time-series data; A distribution center trend analysis is performed on the time series data segment to obtain the first set of characteristics of the time series data segment; The degree of fluctuation dispersion of the time series data segment is used as the second set of features of the time series data segment; The time series data segment is differentially compared to obtain the direction and magnitude of change of the time series data segment, and the direction and magnitude of change are used as the third set of features of the time series data segment. By integrating the first set of features, the second set of features, and the third set of features, multidimensional features of the time series data segment are obtained. By associating and coupling the multidimensional features and the time-off feature vector, a preliminary fusion vector for the chronic hepatitis B patient is obtained. The preliminary fusion vector is subjected to principal component transformation to obtain the composite feature vector of the chronic hepatitis B patient.

[0011] In a preferred embodiment, the step of quantifying and evaluating the composite feature vector to obtain a confidence score for the composite feature vector, and assigning a corresponding quality label to the composite feature vector based on the confidence score, includes: Based on the clinical indicator types of the patients with chronic hepatitis B, cluster analysis is performed on the composite feature vector to obtain a feature subset of the composite feature vector. The feature subset is evaluated in multiple dimensions to obtain the subset confidence score. The confidence scores of the subsets are aggregated to obtain the global confidence score of the composite feature vector; The global confidence score is encoded as a quality label for the composite feature vector, and the quality label is assigned to the composite feature vector.

[0012] In a preferred embodiment, the step of inputting the labeled composite feature vector into a pre-trained cure probability prediction model to obtain the clinical cure tendency index of the chronic hepatitis B patient includes: The weight adjustment factor of the composite feature vector is determined based on the quality label carried by the composite feature vector. Based on the weight adjustment factor, the composite feature vector is weighted to obtain the weighted feature vector of the chronic hepatitis B patient; The weighted feature vector is input into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the patient with chronic hepatitis B. The original prediction of cure probability was rationally corrected to obtain the clinical cure tendency index for the patients with chronic hepatitis B.

[0013] In a preferred embodiment, the step of inputting the weighted feature vector into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the chronic hepatitis B patient includes: Baseline information, multimodal time-series monitoring records, and clinical cure outcome labels of patients with a history of chronic hepatitis B were collected to obtain historical treatment cycle data of these patients. The historical treatment cycle data is deconstructed to obtain the historical feature vector of the historical chronic hepatitis B patients; The historical feature vectors are associated and matched with the clinical cure outcome labels to obtain the model training samples of the historical chronic hepatitis B patients; Identify statistical correlation patterns and quantitative relationships among features in the training samples of the model; The statistical correlation pattern and the quantitative relationship are encapsulated in a structured manner to obtain the predictive architecture for the historical chronic hepatitis B patients; Based on the training samples of the model, the prediction architecture is trained to obtain the cure probability prediction model for the historical chronic hepatitis B patients. Based on the cure probability prediction model, the weighted feature vector is mapped and transformed to obtain the original cure probability prediction for the patient with chronic hepatitis B.

[0014] In a preferred embodiment, generating individualized management decisions for patients with chronic hepatitis B based on the clinical cure propensity index and a pre-defined clinical pathway rule base includes: The clinical cure tendency indicators are categorized to obtain the prognostic stratification category of the patients with chronic hepatitis B; Based on the prognostic stratification categories, a pre-defined clinical pathway rule base is traversed and searched to obtain candidate management strategies for patients with chronic hepatitis B. Obtain the latest treatment stage identifier and historical management decision execution records for the patients with chronic hepatitis B; The candidate management strategies and the latest treatment stage identifier are subjected to compatibility verification to obtain the current feasible strategies for the patients with chronic hepatitis B. Based on the historical management decision execution records, a comprehensive assessment of the current feasible strategies is conducted to determine the conflict between the execution continuity risk and the expected results of the current feasible strategies. Based on the prognostic stratification categories, the conflict between the risk of continuity of treatment and the expected results, the currently feasible strategies are optimized and screened to obtain individualized management decisions for patients with chronic hepatitis B.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention significantly improves the accuracy of composite feature vectors by interpolating and completing multimodal irregular time-series data of patients with chronic hepatitis B, combining curve fitting to extract time-shift features, and then using sliding window multidimensional feature fusion encoding and confidence quantification assessment. This makes the prediction of clinical cure propensity indicators more reliable and efficient. Based on this prediction indicator and combined with a pre-set clinical pathway rule base, highly adaptable individualized management decisions can be generated for patients, effectively improving the overall accuracy and practicality of clinical cure prediction and management for chronic hepatitis B.

[0016] 2. This invention integrates baseline and dynamic monitoring data through a multimodal data fusion platform, uses an artificial intelligence screening model to accurately identify advantageous populations, and combines a treatment response prediction model to dynamically output the clinical cure probability. It automatically generates individualized treatment decision suggestions that align with guidelines to guide treatment strategies, thus constructing a closed-loop management process throughout the entire disease course. This significantly improves the targeting and success rate of clinical cure treatment. Through standardized processing and structured integration of diagnostic and treatment data, it automatically constructs a high-quality real-world research dataset, providing data support for model iteration and optimization, realizing the asset value of diagnostic and treatment data, and further strengthening the sustainability and optimization capabilities of clinical cure management. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data, provided in an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data, according to an embodiment of the present invention. In this embodiment, the method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data includes: S1. Acquire multimodal time-series data of patients with chronic hepatitis B during a continuous monitoring period, and perform interpolation processing on the multimodal time-series data to obtain continuous time-series data of patients with chronic hepatitis B. In this embodiment of the invention, the step of interpolating the multimodal time-series data to obtain the continuous time-series data of the chronic hepatitis B patient includes: By monitoring the laboratory test value sequences, structured clinical event records, and reporting indicators of patients with chronic hepatitis B, multimodal time-series data of these patients were obtained. Discontinuous time points in the multimodal time series data are used as nodes to be completed for the patients with chronic hepatitis B. Select the valid data at adjacent time points in the node to be completed to obtain the local reference data segment of the chronic hepatitis B patient; In the time dimension, the numerical change trend of the local reference data segment is analyzed to obtain the numerical estimate of the node to be completed; The numerical estimate is inserted into the corresponding position in the multimodal time series data to obtain the continuous time series data of the chronic hepatitis B patient.

[0021] This system monitors laboratory test result sequences, structured clinical event records, and reporting indicators for patients with chronic hepatitis B. The laboratory test result sequences include past test results for core laboratory tests related to chronic hepatitis B, such as hepatitis B surface antigen quantification, hepatitis B virus deoxyribonucleic acid (HBeAg), alanine aminotransferase (ALT), and aspartate aminotransferase (AST). The structured clinical event records cover key clinical events such as treatment initiation time, medication adjustments, details of adverse reactions, and medical records. The reporting indicators include summative indicators such as liver function classification and liver fibrosis assessment results. By systematically collecting these different types of data and organizing them chronologically, multimodal time-series data for patients with chronic hepatitis B is generated.

[0022] Based on the established monitoring cycle for patients with chronic hepatitis B, each time point in the multimodal time series data is checked one by one. If monitoring data that should exist according to the monitoring cycle is missing in a certain time period, the time point corresponding to the missing data is a discontinuous time point. All such discontinuous time points are clearly defined as nodes to be supplemented for patients with chronic hepatitis B.

[0023] For each node to be completed, find the time point on the time axis that immediately precedes and follows the node with complete monitoring data. Extract all recorded laboratory test values, structured clinical event-related data, and reporting indicators from these two adjacent time points. After these effective data are screened and confirmed to be without missing or errors, they are integrated to form a local reference data segment for patients with chronic hepatitis B.

[0024] In terms of time, the numerical changes of each indicator in the local reference data segment are analyzed one by one. If the value of a certain indicator shows a gradual upward trend from the previous adjacent time point to the next adjacent time point, the value of the indicator to be filled in the node is estimated by taking into account the time span between the two time points in a uniformly increasing manner. If the value of the indicator shows a gradual downward trend, it is estimated by taking a uniformly decreasing manner. If the value of the indicator is basically consistent between two adjacent time points, the value of the indicator at the adjacent time points is directly determined as the value of the node to be filled in. In this way, the numerical estimation of all indicators is completed, and the numerical estimate of the node to be filled in is obtained.

[0025] The numerical estimates of the nodes to be completed are precisely mapped to the time position of the nodes in the multimodal time series data. According to the original data format and field requirements of the multimodal time series data, the numerical estimates of each indicator are filled into the corresponding fields to fill the data gaps at that time point. After supplementing the numerical values ​​of all nodes to be completed, a complete and coherent continuous time series data of chronic hepatitis B patients is formed on the time axis.

[0026] The beneficial effects are that the above interpolation processing steps transform the originally missing multimodal time series data into complete and coherent continuous time series data, fully preserving various key information related to the patient's condition and its change trajectory. This provides a high-quality and usable data foundation for subsequent cost curve path construction, multidimensional feature extraction, composite feature vector generation, and the calculation of cure probability prediction models. It effectively solves the problem that irregular time series data is difficult to use directly for accurate analysis, ensures the smooth implementation of subsequent links in the clinical cure prediction and management of chronic hepatitis B, and improves the reliability and accuracy of the overall process.

[0027] S2. Perform curve fitting on the continuous time series data to obtain the cost curve path of the chronic hepatitis B patient, and quantify the time offset of the nodes in the cost curve path to obtain the time offset feature vector of the chronic hepatitis B patient. In this embodiment of the invention, the step of curve fitting the continuous time-series data to obtain the cost-bending path of the chronic hepatitis B patient, and quantifying the time offset of nodes in the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient, includes: The clinical indicator dimensions in the continuous time series data are normalized to obtain the standard clinical indicator dimensions of the continuous time series data. A two-dimensional plane is constructed with time as the horizontal axis and numerical values ​​as the vertical axis for the patients with chronic hepatitis B. By mapping the standard clinical indicator dimensions onto the two-dimensional plane, the indicator curve of the chronic hepatitis B patient is obtained. Based on the proximity of adjacent values ​​on the index curve, candidate matching pairs for the chronic hepatitis B patients are determined. Path planning is performed on the candidate matching point pairs to obtain the continuous path of the chronic hepatitis B patient; Optimal spatiotemporal curvature analysis is performed on the continuous path to obtain the overall alignment cost of the continuous path, and the continuous path with the minimum overall alignment cost is taken as the cost curvature path of the chronic hepatitis B patient. The time offset of the cost-bending path is tensor-synthesized to obtain the time offset feature vector of the chronic hepatitis B patient.

[0028] The step of tensor synthesis of the time offset of the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient includes: Extract the actual monitoring timestamps of the matching points on the cost bending path, and record the projection position of the actual monitoring timestamps on the two-dimensional plane; The time difference between the actual monitoring timestamp and the projected position is used as the original time offset of the chronic hepatitis B patient. The original time offsets are arranged and aggregated to obtain the time offset sequence of the chronic hepatitis B patient. Calculate the overall irregularity score of the time-offset sequence; The comprehensive irregularity score is ordered and encoded with the maximum and minimum values ​​in the time offset sequence to obtain the time offset feature vector of the chronic hepatitis B patient.

[0029] The formula for calculating the comprehensive irregularity score is as follows: ; in, This represents the comprehensive irregularity score. This represents the total number of values ​​in the time offset sequence. Indicates the first The original time offset. This represents the preset weighting coefficient. This represents the average value of the time offset sequence. This represents the preset positive decimal stabilization parameter. It represents the absolute value.

[0030] The clinical indicator dimensions in continuous time series data are normalized. For all clinical indicators included in the continuous time series data, such as hepatitis B surface antigen quantification, hepatitis B virus deoxyribonucleic acid, and alanine aminotransferase, all values ​​of each indicator are adjusted to a fixed range of 0 to 1. Specifically, the minimum value of the indicator in all data is subtracted from each actual value of the indicator, and then the result is divided by the difference between the maximum and minimum values ​​of the indicator. After this processing, all clinical indicator-related data are obtained as the standard clinical indicator dimensions of the continuous time series data.

[0031] Using time as the horizontal axis and numerical values ​​as the vertical axis, a two-dimensional plane is constructed for patients with chronic hepatitis B. Each mark on the horizontal axis corresponds to a specific monitoring time point, arranged sequentially according to the monitoring cycle. Each mark on the vertical axis corresponds to a normalized clinical indicator value, covering the entire range from 0 to 1. Through this setting of the horizontal and vertical axes, a two-dimensional plane is built specifically to present the temporal changes of clinical indicators in patients with chronic hepatitis B.

[0032] By mapping the standard clinical indicators onto a two-dimensional plane, the indicator curves for patients with chronic hepatitis B are obtained. For each standard clinical indicator, the value corresponding to each monitoring time point is extracted in chronological order. Each time point and its corresponding value are used as coordinate points on the two-dimensional plane. Then, smooth lines are used to connect all the coordinate points of the same indicator in chronological order to form the change curve of the indicator. The curves corresponding to all clinical indicators together constitute the indicator curves for patients with chronic hepatitis B.

[0033] Based on the proximity of adjacent values ​​on the indicator curve, candidate matching point pairs for chronic hepatitis B patients are determined. The values ​​corresponding to two adjacent time points on each indicator curve are analyzed one by one, and the difference between the two values ​​is calculated. If the difference is less than a preset fixed threshold, it means that the values ​​corresponding to the two time points are close enough. The coordinate points corresponding to these two time points on the two-dimensional plane are determined as a set of candidate matching point pairs. The adjacent time points of all indicator curves are traversed, and all point pairs that meet the conditions are collected to form a set of candidate matching point pairs.

[0034] Path planning is performed on candidate matching point pairs to obtain the continuous path of chronic hepatitis B patients. All points in the candidate matching point pairs are arranged in an orderly manner according to the time progression, ensuring that the path direction is consistent with the timeline sequence, while ensuring that the connection lines between adjacent point pairs are smooth and uninterrupted. Through this orderly and smooth connection method, a complete path is formed, which is the continuous path of chronic hepatitis B patients.

[0035] Optimal spatiotemporal curvature analysis is performed on continuous paths to obtain the overall alignment cost. The continuous path with the minimum overall alignment cost is selected as the cost curvature path for patients with chronic hepatitis B. The curvature of each continuous path in the time and numerical dimensions is analyzed, and the consistency between the temporal distribution and numerical changes of each point on the path is considered. The degree of fit between each path and the ideal temporal trajectory is calculated. The quantitative result of this degree of fit is the overall alignment cost. By comparing the overall alignment costs of all continuous paths, the continuous path with the minimum value is selected and determined as the cost curvature path for patients with chronic hepatitis B.

[0036] Extract the actual monitoring timestamps of the matching points on the cost bending path and record the projection position of the actual monitoring timestamps on the two-dimensional plane. Filter out each independent point among all candidate matching point pairs on the cost bending path and accurately record the actual monitoring time information corresponding to each point. This information is the actual monitoring timestamp. At the same time, according to the coordinate rules of the two-dimensional plane, determine the specific coordinate position of each actual monitoring timestamp on the plane. This coordinate position is the projection position.

[0037] The time difference between the actual monitoring timestamp and the projected position is used as the original time offset for patients with chronic hepatitis B. The horizontal time scale corresponding to the projected position on the two-dimensional plane is used as a reference benchmark. The specific time value corresponding to the benchmark is extracted. The difference between the time value corresponding to the actual monitoring timestamp of the matching point and the benchmark time value is calculated as the original time offset of the matching point.

[0038] The original time offsets are arranged and aggregated to obtain the time offset sequence of patients with chronic hepatitis B. According to the order of the actual monitoring timestamps corresponding to all matching points on the cost curve path, the original time offsets corresponding to each matching point are arranged in sequence to form an ordered set of values. This ordered set is the time offset sequence of patients with chronic hepatitis B.

[0039] To calculate the overall irregularity score of a time-shifted sequence, first, count the total number of original time shifts in the sequence and assign a fixed weight to each. The weights are pre-set based on the importance of the corresponding clinical indicator. Calculate the absolute value of each original time shift and multiply it by its corresponding weight. Sum all weighted results to obtain the total. Next, calculate the average of all original time shifts in the sequence. Subtract the average from each original time shift and square the result. Sum all squared results and divide by the total number to obtain the variance. Take the square root of the variance to obtain the standard deviation. Add a fixed, small positive value to the standard deviation to ensure computational stability. Finally, divide the sum of the weighted absolute values ​​obtained earlier by the stability-processed standard deviation to obtain the overall irregularity score.

[0040] The total number of values ​​in the time offset sequence comes from the number of values ​​contained in the time offset sequence obtained by arranging and aggregating the original time offsets.

[0041] The original time offset is derived from the actual monitoring timestamp of the matching point on the cost curved path. The projection position of the actual monitoring timestamp on the two-dimensional plane is recorded, and the time difference between the two is the original time offset.

[0042] The preset weighting coefficients are pre-set values ​​used to adjust the degree of influence of different original time offsets in the calculation.

[0043] The average value of the time offset sequence is obtained by summing all the original time offsets in the time offset sequence and then dividing the sum by the total number of values ​​in the time offset sequence.

[0044] The preset positive decimal stabilization parameter is a pre-set positive decimal value used to ensure the stability of the denominator value during the calculation process.

[0045] The calculation process first takes the absolute value of each original time offset, multiplies each absolute value by the corresponding preset weight coefficient, and sums up all the results to form the numerator.

[0046] Next, calculate the difference between each original time offset and the average value of the time offset sequence, square each difference, sum all the squared results, divide the sum by the total number of values ​​in the time offset sequence, add the result to the preset positive decimal stabilization parameter, and take the square root of the sum to form the denominator.

[0047] The result of dividing the numerator by the denominator is the overall irregularity score. This score can quantify the degree of irregularity of the time offset sequence and accurately reflect the overall irregularity characteristics of the time offset on the cost curve path.

[0048] When the absolute value of each original time offset in the time offset sequence is larger and the corresponding preset weight coefficient is larger, the value of the numerator will increase accordingly. With the value of the denominator remaining unchanged, the overall irregularity score will increase accordingly.

[0049] The smaller the difference between each original time offset and the average value in the time offset sequence, the smaller the sum of the squares of these differences divided by the total number, the smaller the value after adding the preset positive decimal stabilization parameter, and the smaller the denominator value after taking the square root. With the numerator value unchanged, the overall irregularity score will increase accordingly.

[0050] When the preset positive decimal stabilization parameter increases, the value of the denominator will increase accordingly, and the overall irregularity fraction will decrease while the value of the numerator remains unchanged.

[0051] The time-shift feature vector of chronic hepatitis B patients is obtained by sequentially encoding the comprehensive irregularity score with the maximum and minimum values ​​in the time-shift sequence. First, the maximum and minimum values ​​are selected from all values ​​in the time-shift sequence. Then, these three key values ​​are converted into vector data in a uniform format according to the fixed order of "comprehensive irregularity score - maximum value - minimum value". The accuracy of the values ​​and the fixed order are maintained during the conversion process. The final vector is the time-shift feature vector of chronic hepatitis B patients.

[0052] The beneficial effects are as follows: through a series of processing steps such as normalization, curve fitting, path selection, and tensor synthesis of time offsets for continuous time-series data, the temporal variation patterns and time offset characteristics of clinical indicators in patients with chronic hepatitis B are accurately captured. The generated time offset feature vector comprehensively and accurately integrates the temporal fluctuation information and time deviation characteristics of the patient's clinical data, providing a high-quality temporal foundation for the subsequent extraction and fusion of multidimensional features within the sliding window. This effectively enhances the temporal representation capability of composite feature vectors, thereby providing key support for the accurate calculation of clinical cure tendency indicators and the scientific generation of individualized management decisions, significantly improving the accuracy and reliability of clinical cure prediction and management for chronic hepatitis B.

[0053] S3. Within the sliding window, extract the multidimensional features of the continuous time series data, and fuse and encode the multidimensional features with the time offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient. In this embodiment of the invention, the step of extracting multidimensional features from the continuous time-series data within a sliding window, and fusing and encoding the multidimensional features with the time-offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient, includes: The continuous time-series data is divided into time-window segments to obtain time-series data segments of the continuous time-series data; A distribution center trend analysis is performed on the time series data segment to obtain the first set of characteristics of the time series data segment; The degree of fluctuation dispersion of the time series data segment is used as the second set of features of the time series data segment; The time series data segment is differentially compared to obtain the direction and magnitude of change of the time series data segment, and the direction and magnitude of change are used as the third set of features of the time series data segment. By integrating the first set of features, the second set of features, and the third set of features, multidimensional features of the time series data segment are obtained. By associating and coupling the multidimensional features and the time-off feature vector, a preliminary fusion vector for the chronic hepatitis B patient is obtained. The preliminary fusion vector is subjected to principal component transformation to obtain the composite feature vector of the chronic hepatitis B patient.

[0054] The continuous time series data is divided into time window segments to obtain time series data segments. A fixed time window duration and movement step size are preset. The time window duration is determined according to the routine monitoring cycle of patients with chronic hepatitis B. The movement step size is half of the window duration. Starting from the beginning time point of the continuous time series data, the continuous time series data within the window duration corresponding to that time point is divided into the first time series data segment. Then, the window is moved forward sequentially according to the set movement step size. After each movement, the continuous time series data within the current window is divided into a new time series data segment, until the window moves to the end time point of the continuous time series data, completing the division of all time series data segments.

[0055] Central tendency analysis was performed on the time series data segments to obtain the first set of characteristics of the time series data segments. For each clinical indicator in each time series data segment, the sum of all values ​​of the indicator in the time series data segment was calculated. The sum was divided by the total number of values ​​to calculate the mean of each indicator. All values ​​of each indicator were arranged in ascending order. If the total number of values ​​was odd, the middle value after the arrangement was taken as the median of the indicator. If the total number of values ​​was even, the average of the two middle values ​​after the arrangement was taken as the median. The means and medians of all clinical indicators were sorted by indicator type to form the first set of characteristics of the time series data segment.

[0056] The degree of fluctuation and dispersion of time series data segments is used as the second set of features for time series data segments. For each clinical indicator in each time series data segment, the maximum and minimum values ​​of the indicator within the time series data segment are found. The range of the indicator is obtained by subtracting the minimum value from the maximum value. The difference between all values ​​of each indicator and the mean value of the indicator is calculated. Each difference is squared and summed. The sum is divided by the total number of values, and the square root of the result is taken to obtain the dispersion value of the indicator. The range and dispersion values ​​of all clinical indicators are collected to form the second set of features of the time series data segment.

[0057] Differential comparisons are performed on time-series data segments to obtain the direction and magnitude of change in the data segments. These directions and magnitudes are then used as the third set of features for the time-series data segments. The values ​​corresponding to two adjacent time points for each clinical indicator in the time-series data segment are selected sequentially. The value of the latter time point is subtracted from the value of the former time point. If the result is positive, it indicates that the indicator is changing upwards between these two time points, and the absolute value of the result is the magnitude of change. If the result is negative, the direction of change is downwards, and the absolute value is the magnitude of change. If the result is zero, the direction of change is stable, and the magnitude of change is zero. The direction of change and corresponding magnitude of change for all adjacent time points for each indicator are recorded, forming the third set of features for the time-series data segment.

[0058] By integrating the first set of features, the second set of features, and the third set of features, we obtain the multidimensional features of the time series data segment. Following the fixed order of "first set of features - second set of features - third set of features", we arrange all the feature values ​​corresponding to each clinical indicator in the three sets of features in sequence to ensure that the relevant feature values ​​of each indicator are continuously distributed, forming a set containing all feature information. This set is the multidimensional feature of the time series data segment.

[0059] By associating and coupling multidimensional features and time-shifted feature vectors, a preliminary fusion vector for patients with chronic hepatitis B is obtained. First, the multidimensional features are transformed into a one-dimensional numerical sequence, keeping the original order of each feature value unchanged. Then, all the values ​​of the time-shifted feature vector are appended to the end of the one-dimensional numerical sequence after the transformation of the multidimensional features in the original order. This allows the clinical indicator change information contained in the multidimensional features to be correlated and fully fused with the temporal shift information contained in the time-shifted feature vector. The new vector formed is the preliminary fusion vector.

[0060] Principal component transformation is performed on the preliminary fusion vector to obtain the composite feature vector of patients with chronic hepatitis B. The correlation of all feature values ​​in the preliminary fusion vector is analyzed to identify key components that reflect the core information of the data and are independent of each other. These key components are linear combinations of the original features in the preliminary fusion vector. The key components are sorted from high to low according to their degree of interpretation of the data information. The top few key components whose cumulative degree of interpretation reaches a preset proportion are retained. These key components are combined according to the sorting results to form a vector, which is the composite feature vector of patients with chronic hepatitis B.

[0061] The beneficial effects are that by segmenting the data through time windows, key segments of continuous time-series data can be accurately extracted, and multi-dimensional features such as central trend, degree of fluctuation dispersion, and direction and amplitude of change can be comprehensively extracted. At the same time, the temporal offset information of the time offset feature vector is deeply integrated. After principal component transformation to remove redundant information and focus on core features, the generated composite feature vector not only fully preserves the temporal change pattern of clinical indicators of chronic hepatitis B patients, but also integrates the temporal offset characteristics, which greatly improves the representation ability and effectiveness of the features. This provides a high-quality and high-value feature foundation for the subsequent quantitative evaluation of composite feature vectors, prediction of cure probability, and generation of individualized management decisions, further ensuring the accuracy and scientific nature of clinical cure prediction and management of chronic hepatitis B.

[0062] S4. Quantitatively evaluate the composite feature vector to obtain a confidence score for the composite feature vector, and assign a corresponding quality label to the composite feature vector based on the confidence score. In this embodiment of the invention, the step of quantifying and evaluating the composite feature vector to obtain a confidence score for the composite feature vector, and assigning a corresponding quality label to the composite feature vector based on the confidence score, includes: Based on the clinical indicator types of the patients with chronic hepatitis B, cluster analysis is performed on the composite feature vector to obtain a feature subset of the composite feature vector. The feature subset is evaluated in multiple dimensions to obtain the subset confidence score. The confidence scores of the subsets are aggregated to obtain the global confidence score of the composite feature vector; The global confidence score is encoded as a quality label for the composite feature vector, and the quality label is assigned to the composite feature vector.

[0063] Based on the clinical indicator types of patients with chronic hepatitis B, cluster analysis was performed on the composite feature vector to obtain the feature subsets of the composite feature vector. The clinical indicator types of patients with chronic hepatitis B are divided into three categories according to their clinical significance: virus-related indicators, liver function-related indicators, and clinical event-related indicators. Among them, virus-related indicators include quantitative hepatitis B surface antigen and hepatitis B virus deoxyribonucleic acid, etc.; liver function-related indicators include alanine aminotransferase and aspartate aminotransferase, etc.; and clinical event-related indicators include treatment initiation time and adverse reaction occurrence, etc. According to this classification standard, the clinical indicator type corresponding to each feature in the composite feature vector was identified one by one, and all features under the same type were grouped together. Each group forms an independent set, and these sets are the feature subsets of the composite feature vector.

[0064] The feature subset is evaluated in three dimensions: data completeness, data consistency, and data relevance. Data completeness assessment involves calculating the percentage of valid values ​​for each feature in the feature subset. A perfect score is awarded if the percentage of valid values ​​for all features is 100%, and 10 points are deducted for each feature whose percentage of valid values ​​decreases by 10%. Data consistency assessment verifies whether the values ​​of each feature in the feature subset are within the clinically accepted reasonable range. A perfect score is awarded if all feature values ​​are within the reasonable range, and 20 points are deducted for each feature whose value exceeds the reasonable range. Data relevance assessment, based on clinical consensus, determines the degree of correlation between the features in the feature subset and the clinical cure outcome of chronic hepatitis B. A very high correlation earns a perfect score, a high correlation earns 80 points, a moderate correlation earns 60 points, a low correlation earns 40 points, and no correlation earns 0 points. The scores of the three dimensions are weighted and summed in a 4:3:3 ratio to obtain the subset confidence score.

[0065] The confidence scores of subsets are aggregated to obtain the global confidence score of the composite feature vector. Based on the importance of clinical indicator types in predicting the clinical cure of chronic hepatitis B, corresponding weights are assigned to each feature subset. The weight of the virus-related indicator subset is set to 0.4, the weight of the liver function-related indicator subset is set to 0.3, and the weight of the clinical event-related indicator subset is set to 0.3. The confidence score of each feature subset is multiplied by its corresponding weight, and all the product results are added together. The sum is the global confidence score of the composite feature vector.

[0066] The global confidence score is encoded into a quality label of a composite feature vector, and the quality label is assigned to the composite feature vector. Four quality label levels and corresponding global confidence score ranges are pre-defined: a global confidence score of 90 or above corresponds to the "excellent" label, 70 to 89 corresponds to the "good" label, 50 to 69 corresponds to the "average" label, and below 50 corresponds to the "poor" label. The score range to which the global confidence score belongs is determined based on the calculated global confidence score, and then the corresponding quality label is matched. The quality label is then associated and bound with the corresponding composite feature vector to complete the assignment of the quality label.

[0067] The beneficial effects are that by clustering and splitting features according to clinical indicator types, quantitatively assessing subset quality in multiple dimensions, weighted aggregating global scores, and using hierarchical coding labels, the reliability of composite feature vectors is comprehensively and accurately controlled. The generated quality labels can clearly distinguish the quality levels of different feature vectors, helping subsequent cure probability prediction models to prioritize the use of high-quality feature vectors for calculation, effectively avoiding the interference of low-quality data on prediction results, significantly improving the calculation accuracy of clinical cure tendency indicators, providing high-quality data support for the scientific generation of subsequent individualized management decisions, and further ensuring the rigor and effectiveness of the clinical cure prediction and management process for chronic hepatitis B.

[0068] S5. Input the labeled composite feature vector into the pre-trained cure probability prediction model to obtain the clinical cure tendency index of the chronic hepatitis B patient. In this embodiment of the invention, the step of inputting a labeled composite feature vector into a pre-trained cure probability prediction model to obtain the clinical cure tendency index of the chronic hepatitis B patient includes: The weight adjustment factor of the composite feature vector is determined based on the quality label carried by the composite feature vector. Based on the weight adjustment factor, the composite feature vector is weighted to obtain the weighted feature vector of the chronic hepatitis B patient; The weighted feature vector is input into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the patient with chronic hepatitis B. The original prediction of cure probability was rationally corrected to obtain the clinical cure tendency index for the patients with chronic hepatitis B.

[0069] The step of inputting the weighted feature vector into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the chronic hepatitis B patient includes: Baseline information, multimodal time-series monitoring records, and clinical cure outcome labels of patients with a history of chronic hepatitis B were collected to obtain historical treatment cycle data of these patients. The historical treatment cycle data is deconstructed to obtain the historical feature vector of the historical chronic hepatitis B patients; The historical feature vectors are associated and matched with the clinical cure outcome labels to obtain the model training samples of the historical chronic hepatitis B patients; Identify statistical correlation patterns and quantitative relationships among features in the training samples of the model; The statistical correlation pattern and the quantitative relationship are encapsulated in a structured manner to obtain the predictive architecture for the historical chronic hepatitis B patients; Based on the training samples of the model, the prediction architecture is trained to obtain the cure probability prediction model for the historical chronic hepatitis B patients. Based on the cure probability prediction model, the weighted feature vector is mapped and transformed to obtain the original cure probability prediction for the patient with chronic hepatitis B.

[0070] Based on the quality label carried by the composite feature vector, the weight adjustment factor of the composite feature vector is determined. A fixed correspondence between the quality label and the weight adjustment factor is preset, where the "high quality" label corresponds to a weight adjustment factor of 1.2, the "good" label corresponds to a weight adjustment factor of 1.0, the "average" label corresponds to a weight adjustment factor of 0.8, and the "poor quality" label corresponds to a weight adjustment factor of 0.5. By identifying the specific quality label carried by the composite feature vector, the corresponding fixed value is directly matched, and this value is the weight adjustment factor of the composite feature vector.

[0071] Based on the weight adjustment factor, the composite feature vector is weighted to obtain the weighted feature vector of patients with chronic hepatitis B. The original order of all feature values ​​in the composite feature vector is kept unchanged. Each feature value is multiplied by the determined weight adjustment factor. The new feature values ​​obtained after the calculation are arranged in the original order, and the new vector is the weighted feature vector of patients with chronic hepatitis B.

[0072] Baseline information, multimodal time-series monitoring records, and clinical cure outcome labels were collected from patients with a history of chronic hepatitis B to obtain historical treatment cycle data for these patients. Baseline information included the patient's age, gender, date of diagnosis of chronic hepatitis B, and whether they had cirrhosis, among other basic health information. Multimodal time-series monitoring records covered the patient's laboratory test value sequences, structured clinical event records, and reporting indicators throughout the treatment period. The clinical cure outcome label was a clear result indicating whether the patient had achieved hepatitis B surface antigen clearance after treatment. All this information from the same patient was integrated to form the patient's historical treatment cycle data.

[0073] The historical treatment cycle data is deconstructed to obtain the historical feature vector of historical chronic hepatitis B patients. Key features related to the clinical cure of chronic hepatitis B are extracted from the historical treatment cycle data, including the central trend characteristics, fluctuation dispersion characteristics, and change direction and amplitude characteristics of various clinical indicators. All extracted feature values ​​are arranged in a fixed order of "central trend characteristics - fluctuation dispersion characteristics - change characteristics", and the resulting ordered feature set is the historical feature vector of historical chronic hepatitis B patients.

[0074] By associating and matching historical feature vectors with clinical cure outcome labels, model training samples of historical chronic hepatitis B patients are obtained. The historical feature vector of each historical patient is used as a set of input data, and the corresponding clinical cure outcome label of the patient is used as the corresponding output result. A one-to-one correspondence between input data and output results is established. Each corresponding "input data - output result" pair is a model training sample. The corresponding data of all historical patients are collected to form a model training sample set.

[0075] Identify the statistical association patterns and quantitative relationships among features in the model training samples. Analyze the association between each feature value of each historical feature vector in the model training sample set and the corresponding clinical cure outcome label. Find the patterns between feature value changes and clinical cure results, such as the trend of increasing clinical cure probability when a certain feature value increases. At the same time, clarify the strength of this association, such as the probability of a corresponding change in clinical cure result when a certain feature changes by a certain amount. These patterns and the description of the strength of the association are the statistical association patterns and quantitative relationships.

[0076] By structurally encapsulating statistical association patterns and quantitative relationships, a predictive architecture for historical chronic hepatitis B patients is obtained. Following the logical structure of "feature input - association calculation - result output", the identified statistical association patterns and quantitative relationships are integrated into the corresponding layers. The feature input layer is used to receive the feature vector to be predicted, the association calculation layer is used to perform calculations based on the statistical association patterns and quantitative relationships, and the result output layer is used to output the predicted clinical cure probability. This fixed hierarchical structure and calculation logic constitute the predictive architecture.

[0077] Based on the model training samples, the prediction architecture is trained to obtain a prediction model for the cure probability of historical chronic hepatitis B patients. The input data from the model training sample set is sequentially input into the prediction architecture. The prediction results output by the architecture are compared with the actual clinical cure outcome labels corresponding to the samples. The relevant parameters of the association calculation in the architecture are adjusted according to the difference between the two, such as the influence weight of features and the critical value of association judgment. The adjustment is iterated repeatedly until the error between the prediction results and the actual outcome labels reaches the preset minimum range. The prediction architecture at this time is the cure probability prediction model.

[0078] Based on the cure probability prediction model, the weighted feature vector is mapped and transformed to obtain the original cure probability prediction for patients with chronic hepatitis B. The weighted feature vector is input into the feature input layer of the cure probability prediction model. The model processes the feature values ​​in the weighted feature vector through the correlation calculation layer according to the internal fixed statistical correlation pattern and quantification relationship, and converts the feature information into the corresponding probability value. The probability value output by the result output layer is the original cure probability prediction for patients with chronic hepatitis B.

[0079] The original cure probability prediction is rationally corrected to obtain a clinical cure tendency index for patients with chronic hepatitis B. Referring to the general rules of clinical treatment of chronic hepatitis B and a large number of clinical case data, the reasonable range of clinical cure probability is set to 0 to 1. If the original cure probability prediction exceeds this range, it is adjusted to the corresponding boundary value. If it is within the reasonable range, the value is slightly modified based on the specific characteristics of the patient's condition, such as whether there is cirrhosis and the duration of treatment. The modified probability value is the clinical cure tendency index for patients with chronic hepatitis B.

[0080] The beneficial effects are that by matching the weight adjustment factor with quality labels, the influence of high-quality feature vectors is strengthened and the interference of low-quality data is weakened. The cure probability prediction model trained based on real historical treatment data has reliable predictive ability. The rationality correction combined with clinical patterns further ensures the accuracy of the results. The generated clinical cure tendency index accurately reflects the cure potential of patients with chronic hepatitis B, providing a core basis for the formulation of subsequent individualized management decisions. It effectively improves the credibility and scientificity of clinical cure prediction for chronic hepatitis B and helps clinicians achieve more precise patient management.

[0081] S6. Based on the clinical cure tendency indicators and the preset clinical pathway rule base, generate individualized management decisions for the patients with chronic hepatitis B.

[0082] In this embodiment of the invention, generating individualized management decisions for patients with chronic hepatitis B based on the clinical cure tendency indicators and a preset clinical pathway rule base includes: The clinical cure tendency indicators are categorized to obtain the prognostic stratification category of the patients with chronic hepatitis B; Based on the prognostic stratification categories, a pre-defined clinical pathway rule base is traversed and searched to obtain candidate management strategies for patients with chronic hepatitis B. Obtain the latest treatment stage identifier and historical management decision execution records for the patients with chronic hepatitis B; The candidate management strategies and the latest treatment stage identifier are subjected to compatibility verification to obtain the current feasible strategies for the patients with chronic hepatitis B. Based on the historical management decision execution records, a comprehensive assessment of the current feasible strategies is conducted to determine the conflict between the execution continuity risk and the expected results of the current feasible strategies. Based on the prognostic stratification categories, the conflict between the risk of continuity of treatment and the expected results, the currently feasible strategies are optimized and screened to obtain individualized management decisions for patients with chronic hepatitis B.

[0083] The clinical cure propensity index is categorized to determine the prognostic stratification of patients with chronic hepatitis B. The clinical cure propensity index is a probability value between 0 and 1. Three prognostic stratification categories and corresponding index intervals are preset. A clinical cure propensity index greater than or equal to 0.7 is the "high cure potential layer", an index between 0.3 and 0.7 is the "medium cure potential layer", and an index less than 0.3 is the "low cure potential layer". The calculated clinical cure propensity index of the patient is compared with the preset interval to determine the range to which the index belongs. The corresponding interval category is the prognostic stratification category of the patient with chronic hepatitis B.

[0084] Based on prognostic stratification, a pre-defined clinical pathway rule base is traversed and retrieved to obtain candidate management strategies for patients with chronic hepatitis B. The pre-defined clinical pathway rule base stores corresponding management strategies according to prognostic stratification. Each stratification contains multiple related strategies for treatment adjustments, follow-up frequencies, and examination items for that type of patient. According to the patient's prognostic stratification, the corresponding category is found in the clinical pathway rule base, and all management strategies stored in that category are extracted one by one. These strategies together constitute the candidate management strategies for patients with chronic hepatitis B.

[0085] We obtain the latest treatment stage markers and historical management decision execution records for patients with chronic hepatitis B. The latest treatment stage markers clearly define the patient's current treatment progress, including four categories: initial treatment stage, intensive treatment stage, maintenance treatment stage, and consolidation treatment stage. These are determined by querying the treatment progress records in the patient's electronic medical records. The historical management decision execution records cover all treatment plans previously received by the patient, medication adjustments, follow-up execution, completion of examination items, and corresponding efficacy feedback. This information is fully extracted from the patient's medical database to form the historical management decision execution records.

[0086] The candidate management strategies and the latest treatment stage identifiers are subjected to a fit test to obtain the current feasible strategies for patients with chronic hepatitis B. The applicable treatment stage range of each candidate management strategy is analyzed one by one to determine whether the strategy matches the patient's latest treatment stage identifier. If the applicable stage of the candidate management strategy includes the current latest treatment stage, the strategy passes the fit test; otherwise, it is eliminated. All candidate management strategies that pass the test are aggregated to form the current feasible strategies for patients with chronic hepatitis B.

[0087] Based on historical management decision execution records, a comprehensive assessment of current feasible strategies is conducted to identify the continuity risks and expected outcome conflicts of these strategies. The analysis examines the connection between current feasible strategies and past strategies recorded in historical management decision execution records. Significant differences between current and past strategies in areas such as medication types and treatment intensity may lead to treatment interruptions or fluctuations in the patient's condition, constituting a continuity risk. Simultaneously, the expected treatment outcomes of current feasible strategies are compared with the actual therapeutic feedback of past strategies. If the expected outcome of the current strategy contradicts the actual outcome of similar strategies in the past, or if the patient's past treatment responses may affect the effectiveness of the current strategy, this constitutes an expected outcome conflict. Detailed records of the risk points and conflict points corresponding to each current feasible strategy are maintained.

[0088] Based on prognostic stratification, the risk of continuity of execution, and the conflict between expected outcomes, current feasible strategies are optimized and screened to obtain individualized management decisions for patients with chronic hepatitis B. Following the principle of "prioritizing the retention of intensive treatment-related strategies for patients with high cure potential, prioritizing adjustment and optimization strategies for patients with medium cure potential, and prioritizing strategies that change treatment plans for patients with low cure potential," and considering the level of continuity of execution risk and the degree of conflict between expected outcomes and current feasible strategies, strategies with low risk, no significant effect conflict, and that meet the corresponding stratification treatment needs are prioritized. Strategies with minor risks or conflicts are retained after adjustment and optimization if they meet the patient's long-term treatment goals. The final one or two optimal strategies are the individualized management decisions for patients with chronic hepatitis B.

[0089] The beneficial effects include: prognostic stratification based on clinical cure tendency indicators to accurately locate patients' cure potential; a candidate management strategy based on stratified retrieval to ensure targeting; and effective avoidance of treatment transition risks and effect conflicts by combining the latest treatment stage with the verification and evaluation of historical execution records. The final optimized individualized management decisions not only fit the characteristics of the patient's current condition but also take into account the historical treatment situation and long-term cure goals, realizing personalized and precise treatment management for patients with chronic hepatitis B. This significantly improves the scientificity and feasibility of management decisions and provides a strong guarantee for the clinical cure of patients.

[0090] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0091] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data, characterized in that, The method includes: S1. Acquire multimodal time-series data of patients with chronic hepatitis B during a continuous monitoring period, and perform interpolation processing on the multimodal time-series data to obtain continuous time-series data of patients with chronic hepatitis B. S2. Perform curve fitting on the continuous time series data to obtain the cost curve path of the chronic hepatitis B patient, and quantify the time offset of the nodes in the cost curve path to obtain the time offset feature vector of the chronic hepatitis B patient. S3. Within the sliding window, extract the multidimensional features of the continuous time series data, and fuse and encode the multidimensional features with the time offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient. S4. Quantitatively evaluate the composite feature vector to obtain a confidence score for the composite feature vector, and assign a corresponding quality label to the composite feature vector based on the confidence score. S5. Input the labeled composite feature vector into the pre-trained cure probability prediction model to obtain the clinical cure tendency index of the chronic hepatitis B patient. S6. Based on the clinical cure tendency indicators and the preset clinical pathway rule base, generate individualized management decisions for the patients with chronic hepatitis B.

2. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, The interpolation process performed on the multimodal time-series data to obtain the continuous time-series data of the chronic hepatitis B patient includes: By monitoring the laboratory test value sequences, structured clinical event records, and reporting indicators of patients with chronic hepatitis B, multimodal time-series data of these patients were obtained. Discontinuous time points in the multimodal time series data are used as nodes to be completed for the patients with chronic hepatitis B. Select the valid data at adjacent time points in the node to be completed to obtain the local reference data segment of the chronic hepatitis B patient; In the time dimension, the numerical change trend of the local reference data segment is analyzed to obtain the numerical estimate of the node to be completed; The numerical estimate is inserted into the corresponding position in the multimodal time series data to obtain the continuous time series data of the chronic hepatitis B patient.

3. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, The process involves curve fitting of the continuous time-series data to obtain the cost-bending path of the chronic hepatitis B patient, and quantifying the time offset of nodes in the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient, including: The clinical indicator dimensions in the continuous time series data are normalized to obtain the standard clinical indicator dimensions of the continuous time series data. A two-dimensional plane is constructed with time as the horizontal axis and numerical values ​​as the vertical axis for the patients with chronic hepatitis B. By mapping the standard clinical indicator dimensions onto the two-dimensional plane, the indicator curve of the chronic hepatitis B patient is obtained. Based on the proximity of adjacent values ​​on the index curve, candidate matching pairs for the chronic hepatitis B patients are determined. Path planning is performed on the candidate matching point pairs to obtain the continuous path of the chronic hepatitis B patient; Optimal spatiotemporal curvature analysis is performed on the continuous path to obtain the overall alignment cost of the continuous path, and the continuous path with the minimum overall alignment cost is taken as the cost curvature path of the chronic hepatitis B patient. The time offset of the cost-bending path is tensor-synthesized to obtain the time offset feature vector of the chronic hepatitis B patient.

4. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 3, characterized in that, The step of tensor synthesis of the time offset of the cost-bending path to obtain the time offset feature vector of the chronic hepatitis B patient includes: Extract the actual monitoring timestamps of the matching points on the cost bending path, and record the projection position of the actual monitoring timestamps on the two-dimensional plane; The time difference between the actual monitoring timestamp and the projected position is used as the original time offset of the chronic hepatitis B patient. The original time offsets are arranged and aggregated to obtain the time offset sequence of the chronic hepatitis B patient. Calculate the overall irregularity score of the time-offset sequence; The comprehensive irregularity score is ordered and encoded with the maximum and minimum values ​​in the time offset sequence to obtain the time offset feature vector of the chronic hepatitis B patient.

5. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 4, characterized in that, The formula for calculating the comprehensive irregularity score is as follows: ； in, This represents the comprehensive irregularity score. This represents the total number of values ​​in the time offset sequence. Indicates the first The original time offset. This represents the preset weighting coefficient. This represents the average value of the time offset sequence. This represents the preset positive decimal stabilization parameter. It represents the absolute value.

6. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, Within the sliding window, multidimensional features of the continuous time-series data are extracted, and these multidimensional features are fused and encoded with the time-offset feature vector to obtain the composite feature vector of the chronic hepatitis B patient, including: The continuous time-series data is divided into time-window segments to obtain time-series data segments of the continuous time-series data; A distribution center trend analysis is performed on the time series data segment to obtain the first set of characteristics of the time series data segment; The degree of fluctuation dispersion of the time series data segment is used as the second set of features of the time series data segment; The time series data segment is differentially compared to obtain the direction and magnitude of change of the time series data segment, and the direction and magnitude of change are used as the third set of features of the time series data segment. By integrating the first set of features, the second set of features, and the third set of features, multidimensional features of the time series data segment are obtained. By associating and coupling the multidimensional features and the time-off feature vector, a preliminary fusion vector for the chronic hepatitis B patient is obtained. The preliminary fusion vector is subjected to principal component transformation to obtain the composite feature vector of the chronic hepatitis B patient.

7. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, The process of quantifying and evaluating the composite feature vector to obtain a confidence score for the composite feature vector, and assigning a corresponding quality label to the composite feature vector based on the confidence score, includes: Based on the clinical indicator types of the patients with chronic hepatitis B, cluster analysis is performed on the composite feature vector to obtain a feature subset of the composite feature vector. The feature subset is evaluated in multiple dimensions to obtain the subset confidence score. The confidence scores of the subsets are aggregated to obtain the global confidence score of the composite feature vector; The global confidence score is encoded as a quality label for the composite feature vector, and the quality label is assigned to the composite feature vector.

8. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, The step of inputting a labeled composite feature vector into a pre-trained cure probability prediction model to obtain clinical cure tendency indicators for the patients with chronic hepatitis B includes: The weight adjustment factor of the composite feature vector is determined based on the quality label carried by the composite feature vector. Based on the weight adjustment factor, the composite feature vector is weighted to obtain the weighted feature vector of the chronic hepatitis B patient; The weighted feature vector is input into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the patient with chronic hepatitis B. The original prediction of cure probability was rationally corrected to obtain the clinical cure tendency index for the patients with chronic hepatitis B.

9. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 8, characterized in that, The step of inputting the weighted feature vector into a pre-trained cure probability prediction model to obtain the original cure probability prediction for the chronic hepatitis B patient includes: Baseline information, multimodal time-series monitoring records, and clinical cure outcome labels of patients with a history of chronic hepatitis B were collected to obtain historical treatment cycle data of these patients. The historical treatment cycle data is deconstructed to obtain the historical feature vector of the historical chronic hepatitis B patients; The historical feature vectors are associated and matched with the clinical cure outcome labels to obtain the model training samples of the historical chronic hepatitis B patients; Identify statistical correlation patterns and quantitative relationships among features in the training samples of the model; The statistical correlation pattern and the quantitative relationship are encapsulated in a structured manner to obtain the predictive architecture for the historical chronic hepatitis B patients; Based on the training samples of the model, the prediction architecture is trained to obtain the cure probability prediction model for the historical chronic hepatitis B patients. Based on the cure probability prediction model, the weighted feature vector is mapped and transformed to obtain the original cure probability prediction for the patient with chronic hepatitis B.

10. The method for predicting and managing the clinical cure of chronic hepatitis B based on irregular time-series data as described in claim 1, characterized in that, The process of generating individualized management decisions for patients with chronic hepatitis B based on the clinical cure tendency indicators and a pre-defined clinical pathway rule base includes: The clinical cure tendency indicators are categorized to obtain the prognostic stratification category of the patients with chronic hepatitis B; Based on the prognostic stratification categories, a pre-defined clinical pathway rule base is traversed and searched to obtain candidate management strategies for patients with chronic hepatitis B. Obtain the latest treatment stage identifier and historical management decision execution records for the patients with chronic hepatitis B; The candidate management strategies and the latest treatment stage identifier are subjected to compatibility verification to obtain the current feasible strategies for the patients with chronic hepatitis B. Based on the historical management decision execution records, a comprehensive assessment of the current feasible strategies is conducted to determine the conflict between the execution continuity risk and the expected results of the current feasible strategies. Based on the prognostic stratification categories, the conflict between the risk of continuity of treatment and the expected results, the currently feasible strategies are optimized and screened to obtain individualized management decisions for patients with chronic hepatitis B.

Images