Feedback training system for multimodal dosing analog data sampling

Abstract

The application discloses a multi-modal drug administration simulation data sampling feedback training system, and relates to the technical field of data processing.The system comprises the following steps: constructing a multi-modal drug administration simulation environment by integrating multi-modal drug administration parameters, generating an initial simulation parameter set for multi-modal drug administration simulation, obtaining a multi-modal drug administration simulation data set, performing feature extraction, obtaining a multi-dimensional feature vector for anomaly detection, and generating a drug administration quality evaluation report; based on the drug administration quality evaluation report, parameters are optimized in the multi-modal drug administration simulation environment, an optimized simulation parameter set is generated for drug administration verification, and the multi-modal drug administration simulation environment is iteratively trained in a closed loop based on the verification result.The application solves the technical problems of lack of comprehensive modeling, data feature extraction and quality evaluation of the multi-modal drug administration process in the prior art, and inability to realize parameter optimization and closed loop training, and achieves the technical effects of improving the authenticity and reliability of the drug administration simulation data, and realizing intelligent optimization and stability improvement of the drug administration process.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a feedback training system for sampling multimodal drug administration simulation data. Background Technology

[0002] Research and application of drug administration processes involve multiple dimensions of factors, including individual physiological characteristics, drug metabolism characteristics, and drug administration operation characteristics, which are complexly coupled. Traditional methods often focus on single-dimensional modeling and analysis, lacking a unified construction and collaborative simulation of multimodal parameters, making it difficult to fully reflect the real drug administration process. In data processing, the lack of systematic feature extraction and anomaly identification methods makes it difficult to promptly detect biases and problems in the data. Furthermore, the absence of a feature-feedback-based training mechanism prevents effective parameter optimization, resulting in a lack of closed-loop iterative capabilities in the simulation environment, hindering continuous improvement and stable enhancement of the drug administration process. Summary of the Invention

[0003] This application provides a feedback training system for sampling multimodal drug administration simulation data, which is used to address the technical problems in the prior art that lack comprehensive modeling, data feature extraction and quality assessment of multimodal drug administration processes, and cannot achieve parameter optimization and closed-loop training.

[0004] In view of the above problems, this application provides a feedback training system for sampling multimodal drug delivery simulation data.

[0005] This application provides a feedback training system for sampling multimodal drug delivery simulation data, the system comprising:

[0006] The simulation module integrates multimodal drug administration parameters to construct a multimodal drug administration simulation environment, generates an initial simulation parameter set for multimodal drug administration simulation, and obtains a multimodal drug administration simulation dataset. The feature extraction module extracts features from the multimodal drug administration simulation dataset to obtain multi-dimensional feature vectors for anomaly detection and generates a drug administration quality assessment report. The parameter optimization module optimizes parameters based on the drug administration quality assessment report fed back to the multimodal drug administration simulation environment, generates an optimized simulation parameter set for drug administration verification, and performs closed-loop iterative training on the multimodal drug administration simulation environment based on the verification results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] This application integrates multimodal drug administration parameters to construct a multimodal drug administration simulation environment, generates an initial simulation parameter set for multimodal drug administration simulation, and obtains a multimodal drug administration simulation dataset. Feature extraction is performed on the multimodal drug administration simulation dataset to obtain multi-dimensional feature vectors for anomaly detection, generating a drug administration quality assessment report. Based on the drug administration quality assessment report, the multimodal drug administration simulation environment is fed back for parameter optimization, generating an optimized simulation parameter set for drug administration verification. Based on the verification results, the multimodal drug administration simulation environment undergoes closed-loop iterative training. This invention addresses the technical problems in existing technologies that lack comprehensive modeling, data feature extraction, and quality assessment of the multimodal drug administration process, making parameter optimization and closed-loop training impossible. By constructing a multimodal drug administration simulation environment, performing multi-dimensional feature extraction and anomaly detection, implementing parameter optimization based on quality assessment results, and conducting verification iterations, the invention improves the realism and reliability of drug administration simulation data, achieving intelligent optimization and stability enhancement of the drug administration process. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of the feedback training system structure for sampling multimodal drug delivery simulation data provided in this application embodiment;

[0011] Figure 2 This is a schematic diagram of the simulation module in the feedback training system for sampling multimodal drug delivery simulation data provided in the embodiments of this application.

[0012] Figure labeling: Simulation module 11, Feature extraction module 12, Parameter optimization module 13. Detailed Implementation

[0013] This application addresses the technical problems in existing technologies that lack comprehensive modeling, data feature extraction, and quality assessment of multimodal drug administration processes, thus hindering parameter optimization and closed-loop training. By constructing a multimodal drug administration simulation environment, performing multi-dimensional feature extraction and anomaly detection, and implementing parameter optimization and verification iteration based on quality assessment results, this application aims to improve the realism and reliability of drug administration simulation data, thereby achieving intelligent optimization and stability enhancement of the drug administration process.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown in the embodiments of this application, a feedback training system for sampling multimodal drug delivery simulation data is provided. The system includes:

[0017] Simulation module 11 is used to integrate multimodal drug administration parameters to construct a multimodal drug administration simulation environment, generate an initial simulation parameter set for multimodal drug administration simulation, and obtain a multimodal drug administration simulation dataset.

[0018] In this embodiment, the simulation module 11 constructs a multimodal drug delivery simulation environment by integrating multimodal drug delivery parameters, generates an initial simulation parameter set in the multimodal drug delivery simulation environment, performs multimodal drug delivery simulation, and obtains a multimodal drug delivery simulation dataset.

[0019] Specifically, firstly, individual physiological characteristic parameters, drug metabolism characteristic parameters, and drug administration operation characteristic parameters are extracted based on historical drug administration data logs. Then, physiological drug administration simulation, pharmacokinetic drug administration simulation, and controlled drug administration simulation are performed to generate corresponding first, second, and third drug administration parameters. Finally, a multimodal drug administration simulation environment is constructed through co-simulation.

[0020] Next, in a multimodal drug delivery simulation environment, the initial simulation parameter set was determined through uniform sampling and feasibility verification, and a synchronous drug delivery simulation was performed to obtain a drug delivery response dataset. Finally, based on the drug delivery response dataset, drug physiological distribution, drug metabolism time, and drug operation status were calculated. The obtained drug delivery distribution concentration simulation data, drug delivery physiological metabolism simulation data, and drug delivery operation simulation data were then integrated to form a multimodal drug delivery simulation dataset.

[0021] Furthermore, such as Figure 2 As shown, in the system provided in the application embodiment, the simulation module 11 further includes:

[0022] The analysis unit is used to collect and analyze historical drug administration data logs, and extract individual physiological characteristic parameters, drug metabolism characteristic parameters, and drug administration operation characteristic parameters; the first simulation unit is used to perform physiological drug administration simulation based on the individual physiological characteristic parameters to generate first drug administration parameters; the second simulation unit is used to perform pharmacokinetic drug administration simulation based on the drug metabolism characteristic parameters to generate second drug administration parameters; the third simulation unit is used to perform controlled drug administration simulation based on the drug administration operation characteristic parameters to generate third drug administration parameters; and the integration unit is used to integrate the first drug administration parameters, the second drug administration parameters, and the third drug administration parameters for collaborative simulation to construct the multimodal drug administration simulation environment.

[0023] In this embodiment, the analysis unit first extracts and analyzes historical drug administration data logs from the historical database. Specifically, through data cleaning and processing steps, redundant and incomplete information in the historical drug administration data logs is removed, and the data is formatted to obtain individual physiological characteristic parameters, drug metabolism characteristic parameters, and drug administration operation characteristic parameters. Individual physiological characteristic parameters include weight, blood flow, and organ function indicators. Drug metabolism characteristic parameters include clearance rate, volume of distribution, and absorption rate. Drug administration operation characteristic parameters include route of administration, dosage, and dosing interval.

[0024] Next, the first simulation unit performs physiological drug administration simulation based on individual physiological characteristic parameters. It uses a preset physiological model to simulate the distribution process of drugs in blood circulation and in different tissues and organs, thereby generating the first drug administration parameters.

[0025] The second simulation unit performs pharmacokinetic dosing simulation based on drug metabolism characteristic parameters. It uses a preset pharmacokinetic model to simulate the absorption, distribution, and metabolism of drugs in the body, thereby generating second dosing parameters.

[0026] The third simulation unit performs controlled drug delivery simulation based on drug delivery operation characteristic parameters. By simulating different drug delivery methods and operation schemes, it simulates the impact of operation modes such as continuous infusion and intermittent infusion on the drug delivery process, thereby generating the third drug delivery parameters.

[0027] Finally, the integration unit merges the first drug delivery parameters obtained from physiological drug delivery simulation, the second drug delivery parameters obtained from pharmacokinetic drug delivery simulation, and the third drug delivery parameters obtained from controlled drug delivery simulation, and performs collaborative simulation on a unified platform to construct a multimodal drug delivery simulation environment.

[0028] Furthermore, in the system provided in the application embodiment, the simulation module 11 further includes:

[0029] The system includes the following components: a uniform sampling unit for uniformly sampling across the multimodal drug administration simulation environment to obtain multiple simulation environment sample data; a feasibility verification unit for setting parameter value ranges to verify the feasibility of the multiple simulation environment sample data and determining an initial simulation parameter set; a synchronous drug administration simulation unit for performing synchronous drug administration simulation in the multimodal drug administration simulation environment according to the initial simulation parameter set to obtain a drug administration response dataset; a distribution calculation unit for calculating drug physiological distribution based on the drug administration response dataset to obtain drug administration distribution concentration simulation data; a metabolic time calculation unit for calculating drug metabolism time based on the drug administration response dataset to obtain drug administration physiological metabolism simulation data; an operational status calculation unit for calculating drug operational status based on the drug administration response data to obtain drug administration operation simulation data; and an integration unit for integrating the drug administration distribution concentration simulation data, the drug administration physiological metabolism simulation data, and the drug administration operation simulation data to construct the multimodal drug administration simulation dataset.

[0030] In this embodiment, the uniform sampling unit sets a parameter space in the multimodal drug delivery simulation environment, traverses all possible parameter combinations, and uses a uniform sampling method to sequentially take values ​​within the parameter range at a fixed step size. For example, when the dosage parameter is set between 0 and 100 mg, the uniform sampling unit generates sampling points at a step size of 10 mg, combines other parameters one by one to form a simulation sample, thereby covering the entire drug delivery protocol range and finally obtaining multiple simulation environment sample data.

[0031] After obtaining sample data from multiple simulated environments, the feasibility verification unit verifies the parameter range for each sample. This is done by comparing the sampling results with preset parameter value ranges; for example, the drug distribution volume must be between 0.5 and 1 liter. Samples exceeding this range are directly discarded. This method ensures that each sample data meets basic physical and physiological constraints, ultimately determining the initial simulation parameter set.

[0032] Subsequently, the simultaneous drug delivery simulation unit conducts simultaneous drug delivery simulations in a multimodal drug delivery simulation environment based on the initial simulation parameter set, performing parallel simulations on all parameter combinations under a unified time axis. For example, by setting the time step to 1 minute, the unit simultaneously calculates the concentration and metabolic processes over time for different combinations of doses and dosing frequencies, thereby collecting response data in the same time series and ultimately obtaining a complete drug delivery response dataset.

[0033] The distribution computing unit then uses the drug administration response dataset to calculate the physiological distribution of the drug, employing a mass conservation method based on a multi-compartment model to simulate the diffusion and distribution process of the drug within the body. For example, the body is divided into multiple compartments such as blood, liver, kidneys, and muscles. By calculating the inter-compartment drug exchange rate, the concentration values ​​in each tissue over time are obtained, ultimately yielding simulated drug administration distribution concentration data.

[0034] Next, the metabolic time calculation unit calculates the drug metabolism time based on the drug administration response dataset, and analyzes the concentration-time curve using the curve integral method. For example, it calculates the area under the curve to reflect the overall drug exposure, and then estimates parameters such as the drug's half-life and mean residence time, ultimately obtaining physiological metabolic simulation data of drug administration.

[0035] The operational status calculation unit calculates the drug operational status based on the drug administration response dataset and reproduces the entire drug administration process through state space playback. For example, in the case of continuous infusion, the operational status calculation unit can record the infusion rate curve and the changes in control signals at each time point, thereby completely restoring the drug's operational mode during the infusion process and ultimately obtaining simulation data of drug administration operation.

[0036] Finally, the integration unit integrates the simulated data of drug distribution concentration, simulated data of drug physiological metabolism, and simulated data of drug administration operation to obtain a multimodal drug administration simulation dataset.

[0037] The feature extraction module 12 is used to extract features from the multimodal drug administration simulation dataset, obtain multi-dimensional feature vectors for anomaly detection, and generate a drug administration quality assessment report.

[0038] In this embodiment, when the feature extraction module 12 extracts features from the multimodal drug administration simulation dataset, it first aligns and normalizes the drug distribution concentration simulation data, drug administration physiological metabolism simulation data, and drug administration operation simulation data according to timestamps to construct a unified standardized drug administration simulation matrix. Then, it performs multi-scale feature extraction and feature fusion on the standardized drug administration simulation matrix to obtain a multi-dimensional feature vector. Next, based on the multi-dimensional feature vector, it performs global outlier detection and local outlier detection to identify global and local outlier parameters, respectively. Finally, it combines the two types of anomaly detection results to perform density clustering, identify anomaly clusters, complete data quality assessment, and generate a drug administration quality assessment report.

[0039] Furthermore, in the system provided in the application embodiment, the feature extraction module 12 further includes:

[0040] The system comprises the following components: an alignment unit for aligning the simulated drug distribution concentration data, the simulated drug physiological metabolism data, and the simulated drug administration operation data according to timestamps to construct a time coordinate system; a normalization unit for normalizing the simulated drug distribution concentration data, the simulated drug physiological metabolism data, and the simulated drug administration operation data according to the time coordinate system to construct a standardized drug administration simulation matrix; a multi-scale feature extraction unit for traversing the standardized drug administration simulation matrix to extract multi-scale features, fusing multi-dimensional features to generate a multi-dimensional feature vector; a global isolation detection unit for performing global isolation detection based on the multi-dimensional feature vector to determine global outlier parameters; a local outlier detection unit for performing local outlier detection based on the multi-dimensional feature vector to determine local outlier parameters; and a data quality assessment unit for performing density clustering based on the global outlier parameters and the local outlier parameters to identify outlier clusters, assessing the data quality of the multimodal drug administration simulation dataset, and constructing the drug administration quality assessment report.

[0041] In this embodiment, the alignment unit first unifies the start and end times and sampling intervals of the drug distribution concentration simulation data, drug physiological metabolism simulation data, and drug operation simulation data according to timestamps, and performs linear interpolation or adjacent value filling for missing time points so that the three types of data correspond one-to-one at the same time point, thereby constructing a time coordinate system and obtaining aligned multi-source time series data.

[0042] The normalization unit then uses the Z-score normalization method to calculate the mean and standard deviation for each class of aligned data. Then, it subtracts the mean from each data point and divides by the standard deviation to obtain a standardized result that eliminates dimensional and numerical differences. Finally, it combines the data in a way that uses time as the row and variables as the column to construct a standardized drug administration simulation matrix.

[0043] Next, the multi-scale feature extraction unit traverses the standardized drug delivery simulation matrix to extract multi-scale features. During this traversal, time-domain analysis is performed sequentially to obtain time-domain features such as mean and variance; frequency-domain analysis is performed to obtain frequency-domain features such as periodicity and volatility; analysis is performed according to the time dimension to obtain the temporal evolution features of data over time; and analysis is performed according to the spatial dimension to obtain the spatial distribution features between different variables. Finally, the time-domain features, frequency-domain features, temporal evolution features, and spatial distribution features are fused to generate a multi-dimensional feature vector.

[0044] The global isolated detection unit then uses the global threshold method to analyze the overall distribution of multi-dimensional feature vectors. First, it calculates the overall baseline and dispersion of each feature, and then compares the value of each sample with the baseline. When the deviation exceeds the preset threshold, the sample is identified as a global outlier, and the time and location of the outlier and the features involved are recorded. Finally, the global outlier parameters are obtained.

[0045] The local outlier detection unit uses the nearest neighbor average distance method to calculate the average distance between each sample and several neighboring samples. Then, it compares this distance with the benchmark level of the local area. If the distance is greater than the normal value in the neighborhood, the sample is identified as a local outlier and its time location and corresponding features are marked, thereby obtaining the local outlier parameters.

[0046] Finally, the data quality assessment unit performs density clustering based on global and local outlier parameters. In this process, firstly, high-density outlier regions in the multimodal drug administration simulation dataset are identified through density analysis. Then, based on the outlier distribution characteristics, the target number of clusters is set and clustering is completed to obtain outlier clusters. Next, root cause analysis is performed on the outlier clusters to determine multiple root causes and calculate their influence coefficients. Then, the multimodal drug administration simulation dataset is quality assessed according to these influence coefficients to generate multiple drug administration quality scores. Finally, the multiple drug administration quality scores are sorted in descending order by multiple indicators to generate a drug administration quality assessment report.

[0047] Furthermore, in the system provided in the application embodiments, the multi-scale feature extraction unit further includes:

[0048] The system includes a time-domain analysis subunit for traversing the standardized drug administration simulation matrix to perform time-domain analysis and obtain time-domain features; a frequency-domain analysis subunit for traversing the standardized drug administration simulation matrix to perform frequency-domain analysis and obtain frequency-domain features; a time-dimensional analysis subunit for traversing the standardized drug administration simulation matrix to perform analysis according to the time dimension and obtain time evolution features; a spatial-dimensional analysis subunit for traversing the standardized drug administration simulation matrix to perform analysis according to the spatial dimension and obtain spatial distribution features; and a multi-dimensional feature determination subunit for adding the time-domain features, the frequency-domain features, the time evolution features, and the spatial distribution features to the multi-dimensional features.

[0049] In this embodiment, the time-domain analysis subunit adopts the sliding window statistical method. First, the window length and step size are set, and the time series of each column of the standardized drug administration simulation matrix is ​​traversed segment by segment according to the window. For each window, the mean, variance and the average of the first difference are calculated as time-series statistics, and these statistics are collected in time order to obtain the time-domain features.

[0050] The frequency domain analysis subunit uses Fast Fourier Transform (FFT) to perform FFT on each column of the time series in the standardized drug administration simulation matrix to obtain the amplitude spectrum. Frequency components such as the dominant frequency, corresponding amplitude, and total spectral energy are extracted from the amplitude spectrum and organized according to variable dimensions and time positions to obtain the frequency domain features.

[0051] The time dimension analysis subunit employs a linear regression trend fitting method to fit a linear model to each time series over the entire time range, with time as the independent variable and the standardized sequence as the dependent variable, and extracts the slope and intercept. The slope is used to characterize the upward or downward trend over time, and the intercept is used to characterize the baseline level. The two are combined to obtain the time evolution characteristics.

[0052] The spatial dimension analysis subunit uses the Pearson correlation coefficient matrix method to calculate the Pearson correlation coefficients between each pair of variables in the standardized drug administration simulation matrix at the same time coordinate, resulting in a correlation coefficient matrix. Then, for each variable, quantitative indicators such as the average correlation and maximum correlation with other variables are summarized to form the spatial distribution characteristics reflecting the interrelationships of multiple variables.

[0053] Finally, the multi-dimensional feature determination sub-unit adopts the feature splicing and fusion method. Under the premise of time alignment, the time domain features, frequency domain features, time evolution features and spatial distribution features are concatenated and spliced ​​according to the feature dimensions. If necessary, the spliced ​​feature vector is standardized to eliminate the difference in dimensions, and finally a multi-dimensional feature vector is generated.

[0054] Furthermore, in the system provided in the application embodiments, the data quality assessment unit further includes:

[0055] The system comprises the following subunits: a density analysis subunit, used to perform density analysis based on the global and local anomaly parameters to identify high-density anomaly regions; an anomaly distribution detection subunit, used to perform anomaly distribution detection based on a target clustering number set for the high-density anomaly regions, and to cluster the anomalies based on their distribution characteristics to obtain anomaly clusters; an impact calculation subunit, used to traverse the anomaly clusters to perform root cause analysis, identify multiple anomaly root causes, and perform impact calculation on the multimodal drug administration simulation dataset based on these multiple anomaly root causes to obtain impact coefficients; a quality assessment subunit, used to perform quality assessment on the multimodal drug administration simulation dataset according to the impact coefficients, and generate multiple drug administration quality scores; and a multi-indicator descending order sorting subunit, used to sort the multimodal drug administration simulation dataset in descending order based on the multiple drug administration quality scores, and generate the drug administration quality assessment report.

[0056] In this embodiment, when the density analysis subunit performs density analysis based on global and local outlier parameters, it adopts a fixed-radius neighborhood counting method. It sets the neighborhood radius and counting threshold in the multidimensional features and time space, calculates the density of each outlier, counts the number of outliers in its neighborhood, and maps the results to density values ​​to form a density distribution map. When the count of a certain area exceeds the threshold, the area is identified as a high-density outlier area.

[0057] Subsequently, the anomaly distribution detection subunit performs anomaly distribution detection based on the target number of clusters set for high-density anomaly regions. It adopts the K-means clustering method, which determines the target number of clusters based on the number or area ratio of high-density anomaly regions. After initializing the cluster centers, it performs iterative calculations in the set of anomaly points, including point allocation, center update, and convergence determination, until the results are stable or the set conditions are met. Finally, it performs clustering based on the distribution characteristics, obtains anomaly clusters, and outputs their center positions.

[0058] Next, the influence calculation sub-unit traverses the anomaly cluster to perform root cause analysis. The standardized regression coefficient method is used to extract candidate causal variables from the anomaly cluster, construct a design matrix and select response quantities related to the anomaly intensity. After standardizing the independent variables and response quantities, the least squares method is used to calculate the regression coefficients. The absolute values ​​of the coefficients are normalized to obtain the influence coefficients of multiple anomaly root causes on the multimodal drug administration simulation dataset.

[0059] The quality assessment subunit then performs a quality assessment on the multimodal drug administration simulation dataset according to the aforementioned influence coefficients. In this process, for each data set in the multimodal drug administration simulation dataset, the proportion of outliers is statistically analyzed across three dimensions: simulated drug distribution concentration data, simulated drug administration physiological metabolism data, and simulated drug administration operation data. This proportion is converted into a base score, which is then multiplied by the influence coefficient corresponding to that dimension to obtain a weighted sub-score. The three weighted sub-scores are then summed to obtain the drug administration quality score for that data set, while retaining the three weighted sub-scores as individual quality scores. This process is repeated across the multimodal drug administration simulation dataset to ultimately generate multiple drug administration quality scores.

[0060] Finally, the multi-index descending sorting subunit sorts the multimodal drug administration simulation dataset in descending order based on multiple drug administration quality scores. In this process, the calculated drug administration quality scores are sorted. Specifically, the drug administration quality score and the three sub-item drug administration quality scores for each data set are extracted and used as a multi-index score set. Then, the scores are sorted in descending order of priority. If the drug administration quality scores are the same, the scores of the drug distribution concentration simulation data, drug administration physiological metabolism simulation data, and drug administration operation simulation data are compared sequentially, continuing the descending sorting process. If a tie still exists when all the above scores are the same, the time order is used as the final distinguishing criterion. In this way, multiple drug administration quality scores are globally ordered, ultimately generating a drug administration quality assessment report.

[0061] The parameter optimization module 13 is used to optimize parameters based on the drug administration quality assessment report fed back to the multimodal drug administration simulation environment, generate an optimized simulation parameter set for drug administration verification, and perform closed-loop iterative training on the multimodal drug administration simulation environment based on the verification results.

[0062] In this embodiment, when the parameter optimization module 13 optimizes parameters based on the drug administration quality assessment report fed back to the multimodal drug administration simulation environment, it first performs parameter anomaly analysis based on multiple abnormal root causes to identify abnormal patterns. It then obtains key indicators by parsing the drug administration quality assessment report. Next, it performs parameter optimization analysis based on the correspondence between abnormal patterns and key indicators, identifies sensitive parameter sets, and determines the direction of parameter optimization. Finally, it performs parameter optimization in the multimodal drug administration simulation environment to generate an optimized simulation parameter set. Subsequently, it uses the optimized simulation parameter set for drug administration verification, repeatedly tests the multimodal drug administration simulation dataset, generates data test results, and then performs drug administration quality assessment to calculate the test quality score. This score is compared with the initial drug administration quality score to obtain the quality improvement parameter. Based on the quality improvement parameter, it conducts anomaly change analysis to obtain anomaly distribution parameters, and verifies the drug administration stability of the multimodal drug administration simulation environment. Training terminates when the anomaly distribution parameters meet the optimization objective; otherwise, it continues to feed back and perform closed-loop iterative training to gradually improve the stability and rationality of the multimodal drug administration simulation environment.

[0063] Furthermore, in the system provided in the application embodiment, the parameter optimization module 13 further includes:

[0064] The system includes the following components: a parameter anomaly analysis unit for analyzing parameters based on the multiple root causes of anomalies, identifying multiple anomaly patterns, and ensuring a correspondence between these anomaly patterns and the root causes; a parsing unit for parsing the drug administration quality assessment report to obtain multiple key indicators; a parameter optimization analysis unit for performing parameter optimization analysis based on the multiple anomaly patterns and key indicators to determine parameter optimization requirements; a data mapping analysis unit for performing data mapping analysis between the multimodal drug administration simulation dataset and the drug administration quality assessment report, and identifying sensitive parameter sets based on the analysis results; a parameter optimization direction determination unit for performing parameter optimization analysis on the sensitive parameter sets according to the parameter optimization requirements to determine the parameter optimization direction; and an optimization unit for feeding back the drug administration quality assessment report to the multimodal drug administration simulation environment to optimize parameters according to the parameter optimization direction, generating the optimized simulation parameter set.

[0065] In this embodiment, the parameter anomaly analysis unit employs cluster analysis when analyzing multiple abnormal root causes. Global and local anomaly parameters are categorized based on similarity. After classification, the corresponding abnormal root cause is extracted for each category of anomalies, and the anomaly pattern is determined by combining its specific manifestation. For example, when multiple abnormal root causes are concentrated in a slow rate of decrease in drug concentration, a metabolic delay pattern is formed. When multiple abnormal root causes are concentrated in a large fluctuation in blood drug concentration, a concentration instability pattern is formed, thus establishing a one-to-one correspondence between multiple abnormal patterns and multiple abnormal root causes.

[0066] The parsing unit employs an index extraction method when analyzing the drug administration quality assessment report. It structures the multi-dimensional data from the report and calculates key performance indicators (KPIs) item by item, including drug distribution concentration deviation, metabolic half-life error, and drug administration stability fluctuation. During the calculation, drug distribution concentration deviation is obtained by comparing the simulated concentration curve with a reference concentration curve; metabolic half-life error is obtained by comparing the simulated half-life with a standard half-life; and drug administration stability fluctuation is obtained by statistically analyzing the fluctuation range of control parameters during the drug administration process. These multiple KPIs serve as references for subsequent optimization.

[0067] The parameter optimization analysis unit employs a mapping method when analyzing multiple abnormal patterns and key indicators. It matches each abnormal pattern with a key indicator, compares the abnormal behavior with the indicator deviation, and clarifies the parameter optimization requirements. For example, when the abnormal pattern is a metabolic delay pattern and the key indicator shows a half-life error greater than a threshold, it proposes a need to optimize the drug metabolism constant. When the abnormal pattern is a concentration instability pattern and the key indicator shows large fluctuations in blood drug concentration, it proposes a need to optimize the dosing interval or dosing rate.

[0068] The data mapping analysis unit employs sensitivity analysis when mapping multimodal drug administration simulation datasets to drug administration quality assessment reports. It calculates the correlation between various parameters in the simulation dataset and key indicators in the report. During the calculation process, regression fitting analysis is used to analyze the relationship between parameter changes and indicator deviations, identifying a set of sensitive parameters that significantly impact the assessment results. For example, when changes in the clearance constant have the greatest impact on metabolic half-life error, the clearance constant is identified as a sensitive parameter.

[0069] The parameter optimization direction determination unit employs a single-parameter iterative method when analyzing the sensitive parameter set. Each sensitive parameter is gradually adjusted within a feasible range, and the changes in key indicators are calculated after each adjustment. By comparing the improvement effects, the optimal parameter adjustment direction is determined. For example, the clearance rate constant is gradually increased, and the change in metabolic half-life error is observed. When the error approaches the standard value, the optimization direction of the clearance rate constant is determined.

[0070] Finally, the optimization unit feeds back the drug administration quality assessment report to the multimodal drug administration simulation environment for parameter optimization according to the parameter optimization direction. In this process, firstly, multiple key indicators are fed back to the multimodal drug administration simulation environment and matched with the multimodal drug administration simulation dataset to identify key defect information. Then, a targeted search is performed according to the parameter optimization direction to obtain a candidate parameter set. Subsequently, multi-objective optimization is performed on the candidate parameter set to form multiple optimization schemes. Data interaction sensitivity analysis is conducted by combining the optimization with key indicators to determine the parameter optimization step size. Finally, the multiple optimization schemes are adjusted according to the optimization step size to generate an optimized simulation parameter set.

[0071] Furthermore, in the system provided in the application embodiments, the optimization unit further includes:

[0072] The system comprises the following subunits: a matching subunit, used to match the multimodal drug delivery simulation dataset with the multimodal drug delivery simulation environment based on the feedback from the multiple key indicators, and to determine key defect information; a directional search subunit, used to perform directional search according to the parameter optimization direction, and to determine a candidate parameter set; a multi-objective optimization subunit, used to perform multi-objective optimization based on the candidate parameter set, and to determine multiple sets of optimization schemes; a data interaction sensitivity analysis subunit, used to perform data interaction sensitivity analysis on the multiple sets of optimization schemes according to the multiple key indicators, and to determine the parameter optimization step size; and a scheme adjustment subunit, used to adjust the multiple sets of optimization schemes according to the parameter optimization step size, and to generate the optimized simulation parameter set.

[0073] In this embodiment, when the matching subunit feeds back multiple key indicators to the multimodal drug delivery simulation environment, it first uses a difference comparison method to compare the corresponding data in the multimodal drug delivery simulation dataset item by item. Specifically, it calculates the difference between each key indicator and the simulation data, and when a difference exceeds a set threshold, it is identified as key defect information. For example, if the peak blood drug concentration in the drug delivery distribution concentration simulation data differs from the reference value by more than 20%, then this part is determined to be key defect information of concentration deviation.

[0074] Subsequently, after identifying key defect information, the targeted search subunit adjusts parameters using a step-by-step search method according to the parameter optimization direction. Specifically, it selects the parameters most relevant to the key defect information, sets their range and step size for change, gradually increases or decreases the parameter values, and observes whether the difference between the adjusted simulated data and the key indicators narrows, thereby screening out the candidate parameters that best meet the optimization objectives. For example, if prolonged metabolic half-life is a key defect, a targeted search is performed starting with the metabolic rate constant, and the parameters are gradually adjusted and optimized.

[0075] Next, after obtaining the candidate parameter set, the multi-objective optimization subunit performs multi-objective optimization using a step-by-step calculation method. Specifically, it iteratively calculates different combinations of candidate parameters, recording the changes in multiple key indicators under each parameter combination, such as drug distribution concentration deviation, metabolic half-life error, and operational stability fluctuations. Based on these results, it determines which parameter combinations can simultaneously improve multiple objective indicators, ultimately forming multiple sets of optimization schemes. For example, if a certain set of parameter combinations can effectively reduce blood drug concentration deviation and shorten metabolic half-life error, then that set of parameters is retained as an effective optimization scheme.

[0076] The data interaction sensitivity analysis subunit then employs single-factor sensitivity analysis to progressively adjust each candidate parameter in multiple optimization schemes. Specifically, while keeping other parameters constant, the value of one parameter is adjusted individually, and its impact on key indicators is observed. If adjusting a parameter causes a significant change in the key indicator, it indicates that the parameter is highly sensitive to the optimization results. This analysis identifies the parameter that most significantly affects the optimization results, thus determining the adjustment range of the parameter optimization step size. To determine a suitable parameter optimization step size, the degree of influence of the parameter on changes in key indicators is observed during the initial adjustment process. If the change in indicators is large after parameter adjustment, a smaller step size is selected for fine-tuning. If the change is small, the step size is increased for a larger optimization range. Through multiple trial adjustments, a suitable parameter optimization step size is finally determined, thereby ensuring the efficiency and accuracy of the optimization process.

[0077] The final adjustment subunit employs an iterative correction method to adjust the parameters in multiple optimization schemes. Specifically, each sensitive parameter is adjusted step-by-step according to the optimization step size. After each correction, the key indicators are recalculated, and the results are compared with the target values. When the key indicators gradually approach the target values, it indicates that the optimization direction is correct. After multiple iterations, the optimized parameter combination is finally obtained, and an optimization simulation parameter set is generated.

[0078] Furthermore, in the system provided in the application embodiment, the parameter optimization module 13 further includes:

[0079] The system includes the following components: a repeated testing unit for repeatedly testing the multimodal drug administration simulation dataset based on the optimized simulation parameter set, and obtaining data test results; a comparison unit for evaluating drug administration quality based on the data test results, calculating multiple test quality scores, comparing the multiple test quality scores with the multiple drug administration quality scores, and calculating a quality improvement parameter; an anomaly analysis unit for performing anomaly analysis based on the quality improvement parameter, obtaining anomaly distribution parameters, and verifying drug administration stability based on the anomaly distribution parameters; a stop training unit for stopping training the multimodal drug administration simulation environment when the anomaly distribution parameters meet the optimization objective, indicating that the drug administration stability verification has passed; and a closed-loop iterative training unit for performing closed-loop iterative training on the multimodal drug administration simulation environment when the anomaly distribution parameters do not meet the optimization objective, indicating that the drug administration stability verification has failed.

[0080] In this embodiment, when the repeated testing unit performs tests based on the optimized simulation parameter set, it first applies the optimized simulation parameter set to repeatedly test the multimodal drug delivery simulation dataset. Each test is performed in a multimodal drug delivery simulation environment, simulating the absorption, distribution, metabolism, and excretion processes of the drug in different individuals, generating multiple test data results. These test data results include corresponding simulated drug distribution concentration data, simulated drug delivery physiological metabolism data, and simulated drug delivery operation data.

[0081] Subsequently, the comparison unit performs a drug administration quality assessment based on the aforementioned test results. During the assessment, multiple test quality scores are first calculated, following the same process as calculating the drug administration quality score. These scores include a comprehensive score based on simulated drug distribution concentration data, simulated drug administration physiological metabolism data, and simulated drug administration operation data, reflecting the overall quality of the drug administration process. Then, the calculated multiple test quality scores are compared with the drug administration quality score to calculate a quality improvement parameter. The quality improvement parameter measures the difference between the optimized test results and the initial results, thereby evaluating the optimization effect. A positive value for the improvement parameter indicates that the optimized test quality has improved. A negative value indicates that the optimization has not effectively improved the quality.

[0082] Next, the anomaly analysis unit performs anomaly analysis based on the quality improvement rate parameter. By analyzing changes in the quality improvement rate, it determines whether the anomalies have been effectively mitigated. If the quality improvement rate parameter decreases, it indicates that the optimization measures have reduced the anomalies; for example, a reduction in drug concentration fluctuations or metabolic time errors suggests effective optimization. If the quality improvement rate increases, it indicates that the optimization has failed to improve the anomalies, requiring further adjustment of the simulation parameters. Based on this, the anomaly distribution parameters are analyzed, and drug administration stability is verified to check the stability and rationality of the drug administration process, ensuring that the optimization objectives are met.

[0083] Next, the training unit is stopped. When the outlier distribution parameters meet the optimization objective, the drug administration stability verification is passed, meaning the simulation environment has reached the required stability standard. At this point, the training process ends, indicating that the drug administration process has stabilized and meets the expected requirements, and the simulation environment no longer needs further optimization.

[0084] The closed-loop iterative training unit is used when the abnormal distribution parameters do not meet the optimization objective, resulting in a failed drug administration stability verification, indicating that the optimization objective has not been achieved. In this case, closed-loop iterative training continues. Based on feedback from quality assessment and abnormal change analysis, the optimization parameters are adjusted, and through multiple iterations, the stability of the drug administration process gradually reaches the required level.

[0085] Through this repeated closed-loop iterative training process, the optimized parameter set will gradually improve the stability and rationality of the simulation environment, and ultimately achieve the ideal drug delivery process, ensuring that the absorption, distribution, metabolism and excretion of drugs meet expectations.

[0086] Furthermore, in the system provided in the application embodiments, the closed-loop iterative training unit further includes:

[0087] The system includes a deep analysis subunit, used to perform deep analysis on the abnormal distribution parameters when they do not meet the optimization objective, and to identify multiple non-compliant abnormal parameters; a priority reallocation subunit, used to reallocate priorities based on the multiple non-compliant abnormal parameters according to the influence coefficient, and to construct an indicator priority sequence; a secondary verification test subunit, used to set optimization constraints based on drug administration stability, and to perform secondary verification tests on drug administration stability according to the optimization constraints and the indicator priority sequence, and to collect verification data; and a sampling evaluation subunit, used to perform closed-loop iterative training on the multimodal drug administration simulation dataset under the multimodal drug administration simulation environment based on the verification data, construct an iterative training log for sampling evaluation, and generate a training process report.

[0088] In this embodiment, the deep analysis subunit is used to first perform in-depth analysis on the abnormal parameters when the abnormal distribution parameters fail to meet the optimization target. Cluster analysis is used to group the abnormal parameters, grouping parameters exhibiting similar abnormal characteristics into the same group. By calculating the similarity between parameters, it identifies which parameters consistently deviate from the predetermined range in multiple tests. This process identifies multiple non-compliant abnormal parameters.

[0089] After identifying multiple non-compliant outlier parameters, the priority reallocation subunit prioritizes these parameters using impact analysis. Each parameter is evaluated based on its impact on the stability and accuracy of the drug administration process. For example, if a parameter has a significant impact on fluctuations in drug concentration or metabolic time, it will be assigned a higher priority. This step generates a priority-ranked list of parameters.

[0090] The secondary validation test subunit then conducts secondary validation tests based on the set optimization constraints. These constraints include stability requirements such as the drug concentration fluctuation range and metabolic half-life error. In this step, simulation calculations are used to verify whether the optimized parameters can meet these constraints. For example, a drug concentration fluctuation range is set, and it is verified whether it is controlled within a predetermined standard. Through these tests, validation data is finally obtained.

[0091] Finally, the sampling and evaluation subunit further optimizes the drug administration process using a closed-loop iterative training method. In each iteration, the drug administration parameters are adjusted based on validation data, and simulation tests are repeated to observe whether the optimized parameters meet the predetermined goals. After each training iteration, new validation data is collected and compared to evaluate the optimization effect. Ultimately, through repeated training and feedback, an iterative training log is generated to record the adjustments and effects in each training round.

[0092] In summary, the embodiments of this application have at least the following technical effects:

[0093] This application integrates multimodal drug administration parameters to construct a multimodal drug administration simulation environment, generates an initial simulation parameter set for multimodal drug administration simulation, and obtains a multimodal drug administration simulation dataset. Feature extraction is performed on the multimodal drug administration simulation dataset to obtain multi-dimensional feature vectors for anomaly detection, generating a drug administration quality assessment report. Based on the drug administration quality assessment report, the multimodal drug administration simulation environment is fed back for parameter optimization, generating an optimized simulation parameter set for drug administration verification. Based on the verification results, the multimodal drug administration simulation environment undergoes closed-loop iterative training. This invention addresses the technical problems in existing technologies that lack comprehensive modeling, data feature extraction, and quality assessment of the multimodal drug administration process, making parameter optimization and closed-loop training impossible. By constructing a multimodal drug administration simulation environment, performing multi-dimensional feature extraction and anomaly detection, implementing parameter optimization based on quality assessment results, and conducting verification iterations, the invention improves the realism and reliability of drug administration simulation data, achieving intelligent optimization and stability enhancement of the drug administration process.

[0094] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0095] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0096] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A feedback training system for sampling multimodal drug delivery simulation data, characterized in that, The system includes: The simulation module is used to integrate multimodal drug delivery parameters to construct a multimodal drug delivery simulation environment, generate an initial simulation parameter set for multimodal drug delivery simulation, and obtain a multimodal drug delivery simulation dataset. The feature extraction module is used to extract features from the multimodal drug administration simulation dataset, obtain multi-dimensional feature vectors for anomaly detection, and generate a drug administration quality assessment report. The parameter optimization module is used to optimize parameters based on the drug administration quality assessment report fed back to the multimodal drug administration simulation environment, generate an optimized simulation parameter set for drug administration verification, and perform closed-loop iterative training on the multimodal drug administration simulation environment based on the verification results. The feature extraction module includes: Alignment units are used to align drug distribution concentration simulation data, drug physiological metabolism simulation data, and drug administration operation simulation data according to timestamps to construct a time coordinate system; The normalization processing unit is used to normalize the drug distribution concentration simulation data, the drug physiological metabolism simulation data, and the drug operation simulation data according to the time coordinate system, and construct a standardized drug administration simulation matrix. The multi-scale feature extraction unit is used to traverse the standardized drug administration simulation matrix to perform multi-scale feature extraction, obtain multi-dimensional features, fuse them, and generate a multi-dimensional feature vector. A global isolation detection unit is used to perform global isolation detection based on the multi-dimensional feature vector and determine global outlier parameters. A local outlier detection unit is used to perform local outlier detection based on the multi-dimensional feature vector and determine the parameters of local outliers. The data quality assessment unit is used to perform density clustering based on the global outlier parameters and the local outlier parameters, identify outlier clusters, assess the data quality of the multimodal drug administration simulation dataset, and construct the drug administration quality assessment report. The data quality assessment unit includes: The density analysis subunit is used to perform density analysis based on the global anomaly point parameters and the local anomaly point parameters to identify high-density anomaly regions. An anomaly distribution detection subunit is used to perform anomaly distribution detection based on a target cluster number set according to the high-density anomaly region, and to cluster the anomaly clusters based on the distribution characteristics. The influence calculation subunit is used to traverse the abnormal cluster to perform root cause analysis, identify multiple abnormal root causes, and perform influence calculation on the multimodal drug administration simulation dataset based on the multiple abnormal root causes to obtain influence coefficients. The quality assessment subunit is used to assess the quality of the multimodal drug administration simulation dataset according to the influence coefficient, and generate multiple drug administration quality scores. The multi-index descending sorting subunit is used to sort the multimodal drug administration simulation dataset in descending order based on the multiple drug administration quality scores, and generate the drug administration quality assessment report. The parameter optimization module includes: The parameter anomaly analysis unit is used to perform parameter anomaly analysis based on the multiple anomaly root causes, determine multiple anomaly patterns, and the multiple anomaly patterns correspond to the multiple anomaly root causes. The analysis unit is used to analyze the drug administration quality assessment report to obtain multiple key indicators. The parameter optimization analysis unit is used to perform parameter optimization analysis based on the multiple abnormal patterns and the multiple key indicators to determine the parameter optimization requirements. The data mapping analysis unit is used to perform data mapping analysis between the multimodal drug administration simulation dataset and the drug administration quality assessment report, and to identify a set of sensitive parameters based on the analysis results. The parameter optimization direction determination unit is used to perform parameter optimization analysis on the sensitive parameter set according to the parameter optimization requirements and determine the parameter optimization direction. The optimization unit is used to feed back the drug administration quality assessment report to the multimodal drug administration simulation environment to optimize the parameters according to the parameter optimization direction and generate the optimized simulation parameter set.

2. The feedback training system for multimodal drug delivery simulation data sampling as described in claim 1, characterized in that, The simulation module includes: The analysis unit is used to collect and analyze historical drug administration data logs, and extract individual physiological characteristic parameters, drug metabolism characteristic parameters, and drug administration operation characteristic parameters. The first simulation unit is used to perform physiological drug administration simulation based on the individual physiological characteristic parameters and generate the first drug administration parameters. The second simulation unit is used to perform pharmacokinetic dosing simulation based on the drug metabolism characteristic parameters and generate second dosing parameters. The third simulation unit is used to perform controlled drug administration simulation based on the drug administration operation characteristic parameters and generate the third drug administration parameters. An integration unit is used to integrate the first drug administration parameter, the second drug administration parameter, and the third drug administration parameter for collaborative simulation to construct the multimodal drug administration simulation environment.

3. The feedback training system for multimodal drug administration simulation data sampling as described in claim 1, characterized in that, The simulation module includes: A uniform sampling unit is used to traverse the multimodal drug delivery simulation environment to perform uniform sampling and obtain multiple simulation environment sample data. The feasibility verification unit is used to set the parameter value range to verify the feasibility of the multiple simulated environment sample data and determine the initial simulation parameter set. The synchronous drug delivery simulation unit is used to perform synchronous drug delivery simulation in the multimodal drug delivery simulation environment according to the initial simulation parameter set, and obtain a drug delivery response dataset; The distribution calculation unit is used to perform drug physiological distribution calculation based on the drug administration response dataset to obtain simulated drug administration distribution concentration data. The metabolic time calculation unit is used to calculate the drug metabolism time based on the drug administration response dataset to obtain drug administration physiological metabolism simulation data. The operation status calculation unit is used to calculate the drug operation status based on the drug administration response data to obtain drug administration operation simulation data. The integration unit is used to integrate the drug distribution concentration simulation data, the drug administration physiological metabolism simulation data, and the drug administration operation simulation data to construct the multimodal drug administration simulation dataset.

4. The feedback training system for multimodal drug administration simulation data sampling as described in claim 1, characterized in that, The multi-scale feature extraction unit includes: The time-domain analysis subunit is used to traverse the standardized drug administration simulation matrix to perform time-domain analysis and obtain time-domain features. The frequency domain analysis subunit is used to traverse the standardized drug administration simulation matrix to perform frequency domain analysis and obtain frequency domain features. The time dimension analysis subunit is used to traverse the standardized drug administration simulation matrix and analyze it according to the time dimension to obtain the time evolution characteristics. The spatial dimension analysis subunit is used to traverse the standardized drug administration simulation matrix and analyze it according to the spatial dimension to obtain spatial distribution characteristics. A multi-dimensional feature determination subunit is used to add the time-domain features, the frequency-domain features, the time evolution features, and the spatial distribution features to the multi-dimensional features.

5. The feedback training system for multimodal drug delivery simulation data sampling as described in claim 4, characterized in that, The optimization unit includes: The matching subunit is used to match the multimodal drug delivery simulation dataset based on the multiple key indicators fed back to the multimodal drug delivery simulation environment to determine key defect information; The directional search subunit is used to perform a directional search according to the parameter optimization direction to determine the candidate parameter set; A multi-objective optimization subunit is used to perform multi-objective optimization based on the candidate parameter set and determine multiple sets of optimization schemes; The data interaction sensitivity analysis subunit is used to perform data interaction sensitivity analysis on the multiple sets of optimization schemes according to the multiple key indicators, and determine the parameter optimization step size. The scheme adjustment subunit is used to adjust the multiple sets of optimization schemes according to the parameter optimization step size, and generate the optimization simulation parameter set.

6. The feedback training system for multimodal drug delivery simulation data sampling as described in claim 1, characterized in that, The parameter optimization module includes: The repeated testing unit is used to repeatedly test the multimodal drug delivery simulation dataset based on the optimized simulation parameter set to obtain data test results; The comparison unit is used to evaluate the quality of drug administration based on the data test results, calculate multiple test quality scores, compare the multiple test quality scores with the multiple drug administration quality scores, and calculate the quality improvement parameter. An abnormal change analysis unit is used to perform abnormal change analysis based on the quality improvement parameter, obtain abnormal distribution parameters, and perform drug administration stability verification based on the abnormal distribution parameters. The training stop unit is used to stop training the multimodal drug delivery simulation environment when the abnormal distribution parameters meet the optimization objective, indicating that the drug delivery stability verification has passed. A closed-loop iterative training unit is used to perform closed-loop iterative training on the multimodal drug delivery simulation environment when the abnormal distribution parameters do not meet the optimization objective, thus failing the drug delivery stability verification.

7. The feedback training system for multimodal drug delivery simulation data sampling as described in claim 6, characterized in that, The closed-loop iterative training unit includes: The deep analysis subunit is used to perform deep analysis on the abnormal distribution parameters when the abnormal distribution parameters do not meet the optimization target, and to identify multiple non-compliant abnormal parameters. The priority reallocation subunit is used to reallocate priorities based on the multiple non-compliant abnormal parameters according to the influence coefficient, and to construct an indicator priority sequence. The secondary verification test subunit is used to set optimized constraints based on drug administration stability, perform secondary verification tests on drug administration stability according to the optimized constraints and the priority sequence of indicators, and collect verification data. The sampling and evaluation subunit is used to perform closed-loop iterative training on the multimodal drug administration simulation dataset in the multimodal drug administration simulation environment based on the verification data, construct an iterative training log for sampling and evaluation, and generate a training process report.

Images