Method and system for detecting quality of medicine for treating rhinitis

Abstract

The invention relates to the technical field of medicine quality detection, and discloses a method and a system for detecting the quality of a medicine for treating rhinitis. The method comprises establishing a quality characteristic network comprising a drug matrix, an active ingredient distribution and a physical property presentation layer. Acquiring spectrum, chromatography and particle size distribution data of the medicine to be detected, and mapping the data to the network to generate initial quality characterization. And quantitatively calculating the association strength between the feature points to form a dynamic association link, and identifying a conduction path and a key node of the quality attribute according to the dynamic association link. And triggering adjustment of a network topology structure according to the path and the node state, and redistributing the weights of the feature points. And executing quality judgment based on the updated network, and outputting a structured quality report. According to the method, systematic quantitative analysis of dynamic association between drug quality attributes is realized, judgment logic can be adaptively optimized according to sample features, and the depth and accuracy of quality evaluation are improved.

Description

Technical Field

[0001] This invention relates to the field of drug quality testing technology, specifically to a drug quality testing method and system for treating rhinitis. Background Technology

[0002] Current quality testing of rhinitis medications relies on independent measurements of indicators such as spectral density, chromatographic properties, and particle size distribution, compared with standard ranges. Conventional methods involve isolated comparisons of various test data with preset thresholds, or simple multivariate statistical processing of the data. These techniques treat the drug's matrix composition, active ingredients, and physical properties as separate testing items, only able to determine the compliance of a single parameter.

[0003] Existing technical solutions cannot analyze the intrinsic relationships and dynamic interaction mechanisms between different quality attributes. This results in insufficient ability to identify complex quality defects. At the same time, quality judgment systems based on fixed rules or static models lack adaptability, and their accuracy and reliability are limited when the sample under test exhibits atypical abnormal patterns or novel defects.

[0004] Existing detection technologies are insufficient for in-depth, system-level assessment of drug quality status. A detection method is needed that can quantify the dynamic correlations between quality attributes and adaptively adjust judgment logic based on specific sample characteristics to reveal systemic quality risks and enhance the depth and intelligence of quality control. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for testing the quality of drugs used to treat rhinitis, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for testing the quality of a drug used to treat rhinitis, the method comprising: A quality feature network is established, which consists of multiple interconnected feature layers to characterize the multi-dimensional quality attributes of rhinitis drugs. The quality feature network includes a drug matrix layer, an active ingredient distribution layer, and a physical property expression layer, each of which consists of several quality feature points. The original sample data of the rhinitis drug to be tested is obtained, and the original sample data is mapped to the corresponding feature points of the quality feature network to generate an initial quality characterization map; the original sample data includes spectral data, chromatographic data and particle size distribution data; Within the quality feature network, the correlation strength between feature points in the initial quality characterization map is quantified and calculated to form dynamic correlation links; based on the changes in the strength of the dynamic correlation links, the transmission paths and key nodes of quality attributes in the network are identified. Based on the transmission path and the state of key nodes, the topology of the quality feature network is adjusted to generate an updated quality feature network; in the updated quality feature network, the weights of feature points are redistributed. Based on the quality feature network after weight redistribution, the quality judgment process for the nasal medication to be tested is performed, and a structured quality report is output.

[0007] Preferably, the process of establishing the quality feature network includes: extracting reference sample data from a standard rhinitis drug sample library; performing cluster analysis on the quality attributes in the reference sample data to identify three core attribute clusters: drug matrix, active ingredient, and physical properties; defining each core attribute cluster as a feature layer; and configuring independent feature encoding rules and feature point association rules for each feature layer; the feature encoding rules are used to discretize continuous quality data into state values ​​of feature points, and the feature point association rules are used to define the potential connection relationships between feature points within the same feature layer and between different feature layers.

[0008] Preferably, the step of acquiring the original sample data of the rhinitis drug to be tested and mapping the original sample data to the corresponding feature points of the quality feature network to generate an initial quality characterization map specifically involves: calling the feature encoding rules corresponding to the drug matrix layer, active ingredient distribution layer, and physical property expression layer, respectively, to process the spectral data, chromatographic data, and particle size distribution data in parallel; mapping the processed spectral data to feature point state values ​​in the drug matrix layer regarding excipient composition and uniformity; mapping the processed chromatographic data to feature point state values ​​in the active ingredient distribution layer regarding main component content and impurity spectrum; and mapping the processed particle size distribution data to feature point state values ​​in the physical property expression layer regarding particle morphology and particle size range; and integrating all the above feature point state values ​​and their position information in their respective feature layers to form the initial quality characterization map.

[0009] Preferably, the step of quantifying the correlation strength between feature points in the initial quality characterization map within the quality feature network to form a dynamic correlation link includes: traversing all feature point pairs in the initial quality characterization map according to the feature point correlation rules; for each feature point pair, calculating the cooperative change index and distance decay coefficient of its state value; multiplying the cooperative change index and the distance decay coefficient to obtain the initial correlation strength of the feature point pair; continuously acquiring multiple batches of the original sample data within a preset time window, and repeating the mapping process to generate a series of time-series quality characterization maps; calculating the variance of the correlation strength of the same feature point pair within the time window, defining the reciprocal of the variance as the correlation stability factor; and multiplying the initial correlation strength by the correlation stability factor to obtain the final correlation strength of the dynamic correlation link.

[0010] Preferably, the step of identifying the transmission paths and key nodes of quality attributes in the network based on the strength changes of the dynamic association links specifically involves: setting an association strength threshold, filtering out links in the dynamic association links whose final association strength exceeds the association strength threshold, and marking them as valid association links; finding the shortest path between all feature point pairs in the sub-network composed of all feature points and valid association links; counting the number of shortest paths passing through each feature point, and defining the number of shortest paths as the path centrality of the feature point; identifying feature points whose path centrality ranks higher than a preset proportion as key nodes; and defining the network structure composed of key nodes and the valid association links between them as the main transmission paths.

[0011] Preferably, the step of triggering topology adjustment of the quality feature network based on the state of the transmission path and key nodes includes: monitoring the fluctuation frequency and amplitude of the state values ​​of the key nodes on the transmission path; when the fluctuation frequency of the state values ​​of a specific key node exceeds a frequency threshold and the fluctuation amplitude exceeds an amplitude threshold, determining that the local network where the specific key node is located is in an unstable state, and triggering a re-evaluation of all feature point association links directly connected to the specific key node; the re-evaluation process includes: temporarily freezing the state values ​​of the key node, recalculating the cooperative change index and distance decay coefficient between it and all adjacent feature points based on the latest batch of original sample data, and generating a new association strength; if the difference between the new association strength and the original final association strength exceeds an adjustment threshold, updating the weights of the feature point association links and recording a topology change event.

[0012] Preferably, the updated quality feature network is generated; in the updated quality feature network, the weights of feature points are redistributed, including: counting the total number and distribution of topology change events triggered by all key nodes within a preset period; calculating the number of times each feature point is affected by topology change events based on the distribution of topology change events; weighting and fusing the path centrality of each feature point with the number of times the feature point is affected to obtain the network influence score of the feature point; and normalizing the initial weights of all feature points in the quality feature network based on the network influence score to complete the weight redistribution, thereby generating the updated quality feature network.

[0013] Preferably, the quality determination process for the rhinitis drug to be tested based on the weighted reassignment quality feature network includes: inputting the original sample data of the latest batch of the rhinitis drug to be tested into the updated quality feature network to obtain the corresponding latest quality characterization map; multiplying the state value of each feature point in the latest quality characterization map by the weighted reassignment of the feature point to obtain a weighted state value; and calculating the sum of the weighted state values ​​of the drug matrix layer, the active ingredient distribution layer, and the physical property expression layer respectively as the comprehensive score of each feature layer.

[0014] Preferably, the process of outputting the structured quality report includes: pre-setting the acceptable range interval for the comprehensive score of each feature layer; comparing the calculated comprehensive score of the drug matrix layer, the comprehensive score of the active ingredient distribution layer, and the comprehensive score of the physical property expression layer with the corresponding acceptable range intervals; generating a structured report containing the hierarchical judgment results, wherein the structured report clearly records the comprehensive score value of each feature layer, the corresponding acceptable range interval, the judgment conclusion of whether it is acceptable, and when it is judged as unacceptable, the specific feature points whose weighted state values ​​showed significant abnormalities.

[0015] Preferably, the present invention also includes a drug quality testing system for treating rhinitis, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the drug quality testing method for treating rhinitis as described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By constructing a quality feature network comprising a drug matrix layer, an active ingredient distribution layer, and a physical property representation layer, and mapping raw data such as spectra, chromatography, and particle size distribution to feature points in the network, this method achieves structured characterization of multiple quality attributes. Based on this, the correlation strength between feature points in the initial quality characterization map is quantitatively calculated to form dynamic correlation links. The transmission paths and key nodes of quality attributes in the network are identified based on changes in link strength. This makes the interactions between attributes hidden behind isolated data explicit and quantifiable, enabling the tracing of how a physical property anomaly triggers key changes in the distribution of active ingredients, thereby discovering complex quality defects caused by multi-factor coupling that may be missed by traditional isolated detection methods.

[0017] Based on the identified transmission paths and the states of key nodes, the topology of the quality feature network is adjusted, and the weights of feature points are redistributed in the updated network. This makes the model used for quality judgment no longer static; its internal structure and judgment logic can be optimized and focused in real time on the specific quality correlation patterns exhibited by the current test samples. Finally, quality judgment is performed based on this adaptively adjusted network, allowing the discrimination criteria to dynamically emphasize the most significant quality features and potential risk paths of the current batch, improving the ability to identify unconventional quality fluctuation patterns and the specificity of the discrimination results. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the drug quality testing method for treating rhinitis as described in this invention. Figure 2 A flowchart for generating the initial quality characterization map; Figure 3 A flowchart for forming dynamic association links; Figure 4 A statistical chart showing the distribution and frequency of topology change event types for each feature point; Figure 5 This is a batch heatmap showing the original state values ​​of the physical properties of nasal medications. Detailed Implementation

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

[0020] Please see Figure 1This invention provides a method for quality testing of drugs used to treat rhinitis. The method includes: establishing a quality feature network composed of multiple interconnected feature layers, which is used to characterize the multi-dimensional quality attributes of rhinitis drugs. The network includes a drug matrix layer, an active ingredient distribution layer, and a physical property layer. Each layer consists of several quality feature points. The method involves acquiring original sample data of the rhinitis drug to be tested, including spectral data, chromatographic data, and particle size distribution data. These data are mapped to corresponding feature points in the quality feature network to generate an initial quality characterization map. Within the quality feature network, the correlation strength between feature points in the initial quality characterization map is quantified to form dynamic correlation links. Based on the changes in the strength of the dynamic correlation links, the transmission paths and key nodes of quality attributes in the network are identified. Based on the identified transmission paths and key node states, the topology of the quality feature network is adjusted to generate an updated quality feature network. The weights of the feature points are redistributed in the updated quality feature network. Based on the weight-redistributed quality feature network, a quality judgment process for the rhinitis drug to be tested is performed, and a structured quality report is output.

[0021] In one embodiment of the present invention, the process of establishing a quality feature network includes extracting reference sample data from a standard rhinitis drug sample library, performing cluster analysis on the quality attributes in the reference sample data, identifying three core attribute clusters: drug matrix, active ingredient, and physical properties, defining each core attribute cluster as a feature layer, configuring independent feature encoding rules and feature point association rules for each feature layer, using the feature encoding rules to discretize continuous quality data into the state values ​​of feature points, and using the feature point association rules to define the potential connection relationships between feature points within the same feature layer and between different feature layers.

[0022] In practice, the process of establishing a quality characteristic network relies on a complete standard nasal medication sample library. This library stores a large amount of complete quality data for qualified nasal medication samples that have been verified by authoritative bodies. Reference sample data is extracted from this library. This data typically includes various types of data, such as spectral scanning data, high-performance liquid chromatography (HPLC) spectra, laser particle size distribution data, and microscopic morphology images, forming a multi-dimensional set of raw quality attribute data. Cluster analysis is then performed on the quality attributes in the reference sample data. This cluster analysis uses a distance-based hierarchical clustering algorithm, treating each sample's complete quality attribute vector as a point in a high-dimensional space. In practice, when calculating the distance between any two sample quality attribute vectors, a weighted Euclidean distance metric is used to distinguish the differences in importance between different quality attributes. The weighted Euclidean distance formula can be expressed as: in: Indicates sample With sample The weighted distance between them This represents the total number of dimensions of the quality attribute. and Representing samples respectively and samples In the Standardized values ​​for each quality attribute It is to give the first The non-negative weight coefficients of each quality attribute are summed to 1. Based on the calculated distance matrix, the clustering analysis algorithm gradually merges the closest sample clusters until three stable sample clusters with small internal differences but large differences between each other are formed, thus identifying three core attribute clusters: drug matrix, active ingredient, and physical properties.

[0023] Each core attribute cluster is defined as a feature layer, and independent feature encoding rules and feature point association rules are configured for each feature layer. In some embodiments, for the drug matrix feature layer, the feature encoding rules specify that the absorbance ratio of a specific band in the spectral data is mapped to discrete hierarchical state values ​​to characterize the proportional relationship of excipient composition; for the active ingredient distribution feature layer, the feature encoding rules specify that the peak area ratio of the main peak to the adjacent impurity peak in the chromatogram is divided into several continuous intervals, each interval corresponding to a feature point state value. Feature point association rules are used to define connection relationships. For example, within the same feature layer, feature points that are physically adjacent or have similar chemical properties are pre-defined to have a connection relationship; between different feature layers, feature points with known causal relationships or process linkages are also pre-defined to have a connection relationship. For example, there is a pre-defined connection between the feature point characterizing the content of the main component in the active ingredient distribution feature layer and the feature point characterizing the solubility of particles in the physical property feature layer. The feature encoding rules are responsible for converting continuous reference sample data into discrete feature point state values, while the feature point association rules construct the initial, potential connection network framework between feature points.

[0024] In one embodiment of the present invention, see [reference] Figure 2The process involves acquiring raw sample data of the rhinitis drug to be tested, mapping the raw sample data to corresponding feature points in a quality feature network, and generating an initial quality characterization map. Specifically, this process calls the feature encoding rules corresponding to the drug matrix layer, active ingredient distribution layer, and physical property expression layer, respectively, and processes the spectral data, chromatographic data, and particle size distribution data in parallel. The processed spectral data is mapped to feature point state values ​​in the drug matrix layer regarding excipient composition and uniformity; the processed chromatographic data is mapped to feature point state values ​​in the active ingredient distribution layer regarding main component content and impurity spectrum; and the processed particle size distribution data is mapped to feature point state values ​​in the physical property expression layer regarding particle morphology and particle size range. All feature point state values ​​and their position information in their respective feature layers are integrated to form the initial quality characterization map.

[0025] In practice, the process of acquiring raw sample data of the nasal medication to be tested and generating an initial quality characterization map is based on a pre-constructed quality feature network comprising a drug matrix layer, an active ingredient distribution layer, and a physical property representation layer. The process begins with acquiring raw sample data of the nasal medication to be tested. This raw sample data includes continuous spectral data acquired by a spectrometer, chromatographic data acquired by a chromatograph, and particle size distribution data acquired by a particle size analyzer. Feature encoding rules corresponding to the drug matrix layer, active ingredient distribution layer, and physical property representation layer are invoked to process the spectral data, chromatographic data, and particle size distribution data in parallel. In practice, the feature encoding rules for the drug matrix layer specify a set of characteristic wavelengths or bands, and the absorbance or reflectance values ​​of the input spectral data at these wavelength positions are read and calculated.

[0026] In some embodiments, the mapped spectral data represents the state values ​​of characteristic points in the drug matrix layer regarding the composition and uniformity of excipients. The feature encoding rule may specify that by calculating the absorbance ratio of two specific wavelengths and matching this ratio with a preset discretization interval, a discrete value representing the excipient proportion level is output; this discrete value is the state value of the corresponding characteristic point. Optionally, the feature encoding rule for the active ingredient distribution layer analyzes peak information in the chromatographic data. The mapping process after chromatographic data processing compares the retention time, peak area, and peak area ratio of adjacent impurity peaks of the main component peak in the chromatogram with multiple threshold intervals defined in the feature encoding rule. It can be understood that the feature encoding rule defines a unique characteristic point identifier and corresponding state value calculation logic for each component or impurity spectral pattern of interest. The processed chromatographic data is mapped to the state values ​​of characteristic points in the active ingredient distribution layer regarding the content of the main component and the impurity spectrum. For example, the state value of the main component content characteristic point may be directly obtained from the normalized peak area value through linear transformation and rounding.

[0027] In practical implementation, the feature encoding rules for the physical property representation layer focus on the statistical characteristics of particle size distribution data, mapping the processed particle size distribution data to feature point state values ​​related to particle morphology and particle size range in the physical property representation layer. In some embodiments, the mapping process may involve feature extraction from the particle size distribution curve, such as calculating the particle size values ​​corresponding to 10%, 50%, and 90% of the cumulative distribution, i.e., D10, D50, and D90, and substituting these values ​​along with the distribution width index into the function defined by the feature encoding rules. Optionally, an example of a mapping function for calculating the particle morphology feature point state values ​​is as follows: in: The output state value represents the feature point of particle morphology. and These represent the particle sizes corresponding to cumulative particle size distributions of 10% and 90%, respectively. This indicates the particle size corresponding to a cumulative particle size distribution of 50%. and These are predefined weighting coefficients in the feature coding rules. This represents the floor function. This formula can be understood as a specific implementation of converting continuous granularity parameters into discrete state values ​​in feature encoding rules. After parallel processing and mapping of all original sample data, the state values ​​of all feature points and their position information in their respective feature layers are integrated. The position information includes the feature layer identifier to which the feature point belongs and its coordinate index within the feature layer. The integrated structured data set containing complete mapping relationships is the initial quality characterization map.

[0028] In one embodiment of the present invention, see [reference] Figure 3Within the quality feature network, the correlation strength between feature points in the initial quality characterization map is quantified to form a dynamic correlation link. This process traverses all feature point pairs in the initial quality characterization map according to the feature point correlation rules. For each feature point pair, the cooperative change index and distance decay coefficient of its state value are calculated. The initial correlation strength of the feature point pair is obtained by multiplying the cooperative change index and the distance decay coefficient. Multiple batches of original sample data are continuously acquired within a preset time window, and the mapping process is repeated to generate a series of temporal quality characterization maps. The variance of the correlation strength of the same feature point pair within the time window is calculated, and the reciprocal of the variance is defined as the correlation stability factor. The final correlation strength of the dynamic correlation link is obtained by multiplying the initial correlation strength by the correlation stability factor. Based on the strength changes of dynamic association links, the transmission paths and key nodes of quality attributes in the network are identified. Specifically, an association strength threshold is set and links whose final association strength exceeds the threshold are selected and marked as valid association links. In the sub-network composed of all feature points and valid association links, the shortest path between all feature point pairs is found. The number of shortest paths passing through each feature point is counted and defined as the path centrality of the feature point. Feature points whose path centrality ranks above a preset proportion are identified as key nodes. The network structure composed of key nodes and the valid association links between them is defined as the main transmission path.

[0029] In practical implementation, the process of quantifying the correlation strength between feature points in the initial quality characterization map and identifying transmission paths and key nodes within the quality feature network is initiated after the initial quality characterization map is generated. All feature point pairs in the initial quality characterization map are traversed according to the feature point association rules. For each feature point pair, the cooperative change index and distance decay coefficient of its state values ​​are calculated. In some embodiments, the cooperative change index is obtained by calculating the Pearson correlation coefficient of the state values ​​of two feature points in historical batch data or multiple parallel sample data in the current batch. The distance decay coefficient is defined by a decreasing function based on the relative positional distance or semantic distance of the feature points in the network topology. Multiplying the cooperative change index and the distance decay coefficient yields the initial correlation strength of the feature point pair.

[0030] Within a preset time window, multiple batches of raw sample data are continuously acquired, and the mapping process is repeated to generate a series of time-series quality characterization maps. The length of the time window is set according to the batch frequency of production or testing, and is used to capture the dynamic correlations between quality attributes. The variance of the correlation strength of the same feature point pair within the time window is calculated, and the reciprocal of the variance is defined as the correlation stability factor. Optionally, an example formula for calculating the final correlation strength of dynamic correlation links is as follows: in: This represents the final association strength of the dynamic association link between feature point A and feature point B. This represents the initial correlation strength between feature point A and feature point B. The variance of the association strength between feature points A and B is calculated within a preset time window. The association stability factor is expressed by the following expression: The calculation yields the final association strength of the dynamic association link by multiplying the initial association strength by the association stability factor.

[0031] Based on the strength changes of dynamic association links, the transmission paths and key nodes of quality attributes in the network are identified. Specifically, an association strength threshold is set, and links whose final association strength exceeds the threshold are selected and marked as valid association links. In specific implementations, the association strength threshold can be determined based on the statistical quantiles of historical valid link strengths. In the sub-network composed of all feature points and valid association links, the shortest path between all feature point pairs is found. It can be understood that the calculation of the shortest path is based on the final association strength of the valid association links as the edge weight, using the classic graph theory shortest path algorithm. The number of shortest paths passing through each feature point is counted, and the number of shortest paths is defined as the path centrality of the feature point. Feature points whose path centrality ranks above a preset proportion are identified as key nodes, such as the top 10% or top 5%. The network structure composed of key nodes and the valid association links between them is defined as the main transmission path. In some embodiments, the identification of transmission paths reveals the possible diffusion and influence routes of quality attribute fluctuations in the network.

[0032] In one embodiment of the present invention, the topology adjustment of the quality feature network is triggered based on the state of the transmission path and key nodes. This process monitors the fluctuation frequency and amplitude of the state value of key nodes on the transmission path. When the fluctuation frequency of the state value of a specific key node exceeds the frequency threshold and the fluctuation amplitude exceeds the amplitude threshold, it is determined that the local network where the specific key node is located is in an unstable state, triggering a re-evaluation of the association links of all feature points directly connected to the specific key node. The re-evaluation process includes temporarily freezing the state value of the key node, recalculating the cooperative change index and distance decay coefficient between it and all adjacent feature points based on the latest batch of original sample data to generate a new association strength. If the difference between the new association strength and the original final association strength exceeds the adjustment threshold, the weight of the feature point association links is updated and a topology change event is recorded. An updated quality feature network is generated, and the weights of feature points are redistributed within it. This process involves counting the total number and distribution of topology change events triggered by all key nodes within a preset period. Based on the distribution of topology change events, the number of times each feature point is affected by topology change events is calculated. The path centrality of each feature point is weighted and fused with the number of times the feature point is affected to obtain the network influence score of the feature point. Based on the network influence score, the initial weights of all feature points in the quality feature network are normalized to complete the weight redistribution and generate the updated quality feature network.

[0033] In practice, the process involves adjusting the topology of the quality feature network and generating an updated network based on the transmission path and the state of key nodes, using the identified transmission path and set of key nodes as input. The frequency and amplitude of state value fluctuations of key nodes along the transmission path are monitored. The frequency of state value fluctuations is counted as the number of times the state value of a key node changes per unit time, and the amplitude of state value fluctuations is calculated as the absolute difference or relative proportion of the state value of a key node deviating from its historical baseline value. When the frequency of state value fluctuations of a specific key node exceeds a preset frequency threshold and the amplitude of state value fluctuations simultaneously exceeds a preset amplitude threshold, the local network where the specific key node is located is determined to be in an unstable state, triggering a re-evaluation of all feature point links directly connected to the specific key node.

[0034] In some embodiments, the re-evaluation process includes temporarily freezing the state values ​​of key nodes, generating a new quality characterization map based on the latest batch of original sample data, and recalculating the co-change index and distance decay coefficient between the key node and all adjacent feature points to generate a new association strength. It is understood that the recalculation process follows the same logic as the initial association strength calculation, but the state value data used is from the latest batch. If the difference between the new association strength and the original final association strength exceeds a preset adjustment threshold, the weights of the feature point association links are updated, and this change in link weights is recorded as a topology change event. The adjustment threshold can be set as an absolute value or a percentage relative to the original final association strength.

[0035] The system statistically analyzes the total number and distribution of topology change events triggered by all key nodes within a preset period, such as a week or a month, covering all detection batches. Based on the distribution of topology change events, it calculates the number of times each feature point is affected by a topology change event. In practice, each update to the link weight between a feature point and a key node is counted as being affected by a topology change event. The path centrality of each feature point is weighted and fused with the number of times it is affected to obtain the network influence score of the feature point. Optionally, an example formula for calculating the network influence score is as follows: in: The network influence score represents the feature point v. The path centrality of feature point v is represented. This indicates the number of times a feature point v is affected by a topology change event within a statistical period. and These are preset fusion weighting coefficients used to balance the contribution ratios of path centrality and influence frequency. This represents the natural logarithm function.

[0036] Based on the network influence score, the initial weights of all feature points in the quality feature network are normalized, resulting in a weight redistribution and an updated quality feature network. Normalization typically involves dividing the network influence score of each feature point by the sum of the network influence scores of all feature points, ensuring that the sum of all weights equals 1. It can be understood that feature points with higher network influence scores will be assigned higher weights in the updated quality feature network, giving them greater influence in subsequent quality assessments. Table 1 shows the path centrality, number of times affected, and calculated network influence scores of some feature points over a simplified statistical period.

[0037] Table 1: Calculation Table of Feature Point Topology Change Impact and Network Influence Score In some embodiments, based on the data in the table above, set , According to the formula Calculate the network influence score. After normalization, the feature point FP-0032 will receive a higher weight than the feature point FP-0074. Although the path centrality of the feature point FP-0074 is not low, it is less affected by topology changes during the statistical period, indicating that its association state is relatively stable. Therefore, its gain in weight redistribution is limited.

[0038] See Figure 4 In the topology adjustment phase of the rhinitis drug quality feature network, the distribution and frequency statistics of topology change event types corresponding to different feature points (FP-0032, FP-0105, etc.) are presented. The figure uses a stacked bar chart to classify topology change events into four categories: link weight updates, node state freezes, association rule redefinitions, and link re-evaluations, distinguished by different colors. The height of the bar corresponding to each feature point represents the total frequency of topology change events for that feature point (e.g., FP-0105 has a total of 19 events), and the internal layering of the bars reflects the specific distribution of each type of event at that feature point. In the specific analysis, the distribution of topology change events for a feature point is directly related to its role in the quality feature network: for example, FP-0105 has the highest total event frequency, and the proportion of link re-evaluation events is significant. Combined with the project background, it can be seen that the path centrality of this feature point is likely to be affected by topology changes more frequently, and its associated links are less stable, requiring frequent re-evaluations. In contrast, FP-0074 has a total frequency of only 3 events, indicating that its association state is relatively stable, and the corresponding network influence score gain is limited. These statistical results provide data support for the weight redistribution of quality feature networks. By quantifying the distribution of topology change events for each feature point, unstable nodes in the network can be accurately identified, thereby optimizing the network influence score calculation and weight allocation logic for feature points.

[0039] In one embodiment of the present invention, a quality judgment process for the nasal medication under test is performed based on a quality feature network with reassigned weights. This process inputs the latest batch of original sample data of the nasal medication under test into the updated quality feature network to obtain the corresponding latest quality characterization map. The state value of each feature point in the latest quality characterization map is multiplied by the reassigned weight of that feature point to obtain a weighted state value. The sum of the weighted state values ​​for the drug matrix layer, active ingredient distribution layer, and physical property performance layer is calculated as the comprehensive score for each feature layer. The process of outputting a structured quality report includes pre-setting the acceptable range interval for the comprehensive score of each feature layer. The calculated comprehensive scores for the drug matrix layer, active ingredient distribution layer, and physical property performance layer are compared with their corresponding acceptable range intervals to generate a structured report containing the hierarchical judgment results. The structured report clearly records the comprehensive score value of each feature layer, the corresponding acceptable range interval, the judgment conclusion of whether it is acceptable, and, when the judgment is unacceptable, which specific feature points' weighted state values ​​showed significant anomalies.

[0040] In specific implementation, the process of performing quality judgment and outputting a structured quality report based on the weighted quality feature network is executed after the updated quality feature network is generated. In the updated quality feature network, the original sample data of the latest batch of the nasal medication to be tested is input. The original sample data undergoes the same mapping process as the initial quality characterization map to obtain the corresponding latest quality characterization map. The latest quality characterization map contains the state values ​​reflected by all feature points of the current batch of drugs. The state value of each feature point in the latest quality characterization map is multiplied by the weighted feature point after the weighting to obtain the weighted state value. It can be understood that the weighted feature point after weighting comes from the normalization result based on the network influence score. The sum of the weighted state values ​​of the drug matrix layer, the active ingredient distribution layer, and the physical property performance layer is calculated as the comprehensive score for each feature layer. The calculation process involves summing the weighted state values ​​of all feature points belonging to the same feature layer. In some embodiments, an example formula for calculating the comprehensive score of the physical property performance layer is as follows: in: This represents the overall score of the physical properties performance layer. This represents the set of all feature points belonging to the physical property representation layer. This represents the state value of feature point i in the latest quality characterization graph. This represents the weight of feature point i after reallocation in the updated quality feature network.

[0041] The process of outputting a structured quality report includes pre-setting the acceptable range intervals for the comprehensive scores of each feature layer. These intervals are determined based on the statistical distribution of historical acceptable sample data across the comprehensive scores of each feature layer; for example, they can be set to a range of plus or minus three standard deviations of the mean. The calculated comprehensive scores for the drug matrix layer, active ingredient distribution layer, and physical property performance layer are compared with their corresponding acceptable range intervals. In practice, the comparison operation determines whether the value of each comprehensive score falls within its corresponding acceptable range interval. A structured report containing the hierarchical judgment results is generated, presented in tabular or document format with fixed fields. Essentially, the structured report clearly records the comprehensive score value of each feature layer, its corresponding acceptable range interval, and the judgment of whether it is acceptable. When a judgment of non-acceptance is made, the structured report also needs to record which specific feature points' weighted state values ​​showed significant anomalies. The criterion for significant anomalies could be that the weighted state value of that feature point deviates from its historical normal range by exceeding a preset anomaly threshold. In some embodiments, the structured report may also visually compare the quality characterization chart of the current batch with the standard reference quality characterization chart, and include the comparison chart as an appendix to the report. Optionally, the report may be automatically stored in a database after generation and trigger corresponding quality alerts or notification processes.

[0042] See Figure 5 In the physical property performance layer analysis of this nasal medication quality testing method, a heatmap presented the original state value distribution of four core feature points (average particle size, particle size distribution coefficient, particle sphericity, and bulk density) for 10 production batches. Specifically, the figure uses the production batch as the horizontal axis and the physical property feature points as the vertical axis, and the color gradient (from dark green to dark red) corresponds to the difference in state values ​​(80.0-100.0): the higher the state value, the darker the color; the lower the value, the darker the color. For example, the "average particle size" state value (83.7) of batch 9 is relatively low in the figure, corresponding to a dark red color; while the "particle size distribution coefficient" (92.8) of batch 5 and the "bulk density" (93.4) ​​of batch 8 have higher state values, corresponding to a darker green color. Heatmaps visually present the original state fluctuations of physical property feature points under different production batches. They serve as the basic data carrier for subsequent mapping to generate initial quality characterization maps and calculating the correlation strength of feature points. They can also help identify batch-to-batch variation characteristics of feature points within the physical property expression layer.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for testing the quality of a drug used to treat rhinitis, characterized in that, The process includes the following: A quality feature network is established, which consists of multiple interconnected feature layers to characterize the multi-dimensional quality attributes of rhinitis drugs. The quality feature network includes a drug matrix layer, an active ingredient distribution layer, and a physical property expression layer, each of which consists of several quality feature points. The original sample data of the rhinitis drug to be tested is obtained, and the original sample data is mapped to the corresponding feature points of the quality feature network to generate an initial quality characterization map; the original sample data includes spectral data, chromatographic data and particle size distribution data; Within the quality feature network, the correlation strength between feature points in the initial quality characterization map is quantified and calculated to form dynamic correlation links; based on the changes in the strength of the dynamic correlation links, the transmission paths and key nodes of quality attributes in the network are identified. Based on the transmission path and the state of key nodes, the topology of the quality feature network is adjusted to generate an updated quality feature network. In the updated quality feature network, the weights of the feature points are redistributed; Based on the quality feature network after weight redistribution, the quality judgment process for the nasal medication to be tested is performed, and a structured quality report is output.

2. The method for quality testing of a drug for treating rhinitis according to claim 1, characterized in that, The process of establishing the quality feature network includes: extracting reference sample data from a standard rhinitis drug sample library; performing cluster analysis on the quality attributes in the reference sample data to identify three core attribute clusters: drug matrix, active ingredient, and physical properties; defining each core attribute cluster as a feature layer; and configuring independent feature encoding rules and feature point association rules for each feature layer; the feature encoding rules are used to discretize continuous quality data into the state values ​​of feature points, and the feature point association rules are used to define the potential connection relationships between feature points within the same feature layer and between different feature layers.

3. The method for quality testing of a drug for treating rhinitis according to claim 2, characterized in that, The process of acquiring the original sample data of the rhinitis drug to be tested and mapping the original sample data to the corresponding feature points of the quality feature network to generate an initial quality characterization map involves: calling the feature encoding rules corresponding to the drug matrix layer, active ingredient distribution layer, and physical property expression layer, respectively, to process the spectral data, chromatographic data, and particle size distribution data in parallel; mapping the processed spectral data to feature point state values ​​in the drug matrix layer regarding excipient composition and uniformity; mapping the processed chromatographic data to feature point state values ​​in the active ingredient distribution layer regarding main component content and impurity spectrum; and mapping the processed particle size distribution data to feature point state values ​​in the physical property expression layer regarding particle morphology and particle size range; and integrating all the above feature point state values ​​and their position information in their respective feature layers to form the initial quality characterization map.

4. The method for quality testing of a drug for treating rhinitis according to claim 3, characterized in that, The process of quantifying the correlation strength between feature points in the initial quality characterization map within the quality feature network to form a dynamic correlation link includes: traversing all feature point pairs in the initial quality characterization map according to the feature point correlation rules; for each feature point pair, calculating the cooperative change index and distance decay coefficient of its state value; multiplying the cooperative change index and the distance decay coefficient to obtain the initial correlation strength of the feature point pair; continuously acquiring multiple batches of the original sample data within a preset time window and repeating the mapping process to generate a series of time-series quality characterization maps; calculating the variance of the correlation strength of the same feature point pair within the time window, defining the reciprocal of the variance as the correlation stability factor; and multiplying the initial correlation strength by the correlation stability factor to obtain the final correlation strength of the dynamic correlation link.

5. The method for quality testing of a drug for treating rhinitis according to claim 4, characterized in that, The step of identifying the transmission paths and key nodes of quality attributes in the network based on the strength changes of the dynamic association links specifically involves: setting an association strength threshold, filtering out links in the dynamic association links whose final association strength exceeds the association strength threshold, and marking them as valid association links; finding the shortest path between all feature point pairs in the sub-network composed of all feature points and valid association links; counting the number of shortest paths passing through each feature point, and defining the number of shortest paths as the path centrality of the feature point; identifying feature points whose path centrality ranks higher than a preset proportion as key nodes; and defining the network structure composed of key nodes and the valid association links between them as the main transmission paths.

6. The method for quality testing of a drug for treating rhinitis according to claim 5, characterized in that, The step of triggering topology adjustments in the quality feature network based on the state of the transmission path and key nodes includes: monitoring the frequency and amplitude of state value fluctuations of the key nodes on the transmission path; when the frequency of state value fluctuations of a specific key node exceeds a frequency threshold and the amplitude exceeds an amplitude threshold, determining that the local network where the specific key node is located is in an unstable state, and triggering a re-evaluation of all feature point association links directly connected to the specific key node; the re-evaluation process includes: temporarily freezing the state value of the key node, recalculating the cooperative change index and distance decay coefficient between it and all adjacent feature points based on the latest batch of original sample data, and generating a new association strength; if the difference between the new association strength and the original final association strength exceeds an adjustment threshold, updating the weights of the feature point association links and recording a topology change event.

7. The method for quality testing of a drug for treating rhinitis according to claim 6, characterized in that, The updated quality feature network is generated; In the updated quality feature network, the weights of feature points are redistributed, including: counting the total number and distribution of topology change events triggered by all key nodes within a preset period; calculating the number of times each feature point is affected by a topology change event based on the distribution of the topology change events; weighting and fusing the path centrality of each feature point with the number of times the feature point is affected to obtain the network influence score of the feature point; and normalizing the initial weights of all feature points in the quality feature network based on the network influence score to complete the weight redistribution, thereby generating the updated quality feature network.

8. The method for quality testing of a drug for treating rhinitis according to claim 1, characterized in that, The quality feature network based on weight reallocation performs a quality judgment process for the rhinitis drug to be tested, including: inputting the original sample data of the latest batch of rhinitis drugs to be tested into the updated quality feature network to obtain the corresponding latest quality characterization map; multiplying the state value of each feature point in the latest quality characterization map by the weight of the feature point after reallocation to obtain a weighted state value; and calculating the sum of the weighted state values ​​of the drug matrix layer, the active ingredient distribution layer and the physical property expression layer respectively as the comprehensive score of each feature layer.

9. A method for testing the quality of a drug for treating rhinitis according to claim 8, characterized in that, The process of outputting a structured quality report includes: pre-setting the acceptable range interval for the comprehensive score of each feature layer; comparing the calculated comprehensive score of the drug matrix layer, the comprehensive score of the active ingredient distribution layer, and the comprehensive score of the physical property expression layer with the corresponding acceptable range intervals; generating a structured report containing the hierarchical judgment results. The structured report clearly records the comprehensive score value of each feature layer, the corresponding acceptable range interval, the judgment conclusion of whether it is acceptable, and when it is judged as unacceptable, the specific feature points whose weighted state values ​​showed significant abnormalities.

10. A drug quality testing system for treating rhinitis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of a drug quality testing method for treating rhinitis as described in any one of claims 1 to 9.

Images