Intelligent evaluation method and system for medical equipment extract
By combining standardized reagent kits and intelligent evaluation platforms with deep learning algorithms and graph neural networks, the inconsistencies and complexities in the evaluation of extracts from medical devices have been resolved. This has enabled efficient and accurate identification of chemical components and risk assessment, ensuring the safety and biocompatibility of the materials.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING MINGJIE BIOMEDICAL TESTING CO LTD
- Filing Date
- 2026-01-08
- Publication Date
- 2026-05-05
AI Technical Summary
Existing technologies for assessing the safety and biocompatibility of extracts from medical devices suffer from problems such as inconsistent extraction processes, complex analytical methods, difficulty in detecting non-target components, and lack of risk assessment, resulting in insufficient accuracy and reproducibility of analytical results.
Standardized reagent kits were used for extraction. Combined with a chromatography-mass spectrometry system and an intelligent evaluation platform, spectral features were automatically extracted using deep learning algorithms, chemical components were identified using graph neural networks, and toxicity and biocompatibility were evaluated using quantitative structure-activity relationship models and threshold toxicity concentration models, and reports were automatically generated.
It achieves standardization, automation, and efficiency in the analysis of extracts from medical devices, improves the consistency and accuracy of results, can identify known components and perform deep learning analysis on unknown components, provides scientific risk assessment, and ensures the safety and biocompatibility of materials.
Smart Images
Figure CN121978261A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of chemical analysis technology, specifically relating to an intelligent evaluation method and system for extracts from medical devices. Background Technology
[0002] With the advancement of medical technology and the widespread application of medical devices, an increasing number of medical device materials (such as polymers, metals, and coatings) are entering or coming into contact with the human body. Therefore, ensuring the safety and biocompatibility of these materials has become a critical issue, particularly regarding whether their extracts pose a potential risk to human health. Extracts refer to components that may dissolve or be released from medical device materials when they come into contact with solvents under specific conditions. These components may have adverse effects on the human body, such as causing allergic reactions, toxic effects, or immune responses.
[0003] Currently, assessing the safety of extracts from medical devices typically relies on standardized extraction processes and traditional experimental methods, such as chemical composition analysis and cell culture toxicity testing. However, these traditional methods have certain limitations, mainly in the following aspects: Inconsistent extraction process: Differences in the operation of the extraction process in different laboratories or under different experimental conditions may lead to inconsistencies in the composition of the extract, thereby affecting the accuracy and reproducibility of the analytical results.
[0004] Complexity of analytical methods: Existing analytical methods usually require complex sample pretreatment, lengthy analysis processes, and rely on manual operation, making them susceptible to human error.
[0005] The detection of non-target components is difficult: Most existing analytical methods are biased towards target analysis, which detects known chemical components, while there are significant challenges in the multi-component analysis of unknown components or complex samples.
[0006] Lack of risk assessment: Although component analysis can be performed, there is a lack of unified and standardized technical solutions for assessing the toxicity and biocompatibility of extracts, especially for comprehensive toxicity assessment based on chemical structure and concentration.
[0007] Therefore, current technical solutions urgently need a more standardized, efficient, and intelligent evaluation method to solve the above problems, thereby ensuring the safety and biocompatibility of medical device materials at a higher level. Summary of the Invention
[0008] In view of the above-mentioned problems in the prior art, the purpose of this invention is to provide an intelligent evaluation method for extracts from medical devices. Through innovative means such as standardized extraction, intelligent analysis, deep learning, and automated report generation, this method not only improves the accuracy and reliability of the extract evaluation process, but also solves the problems of inconsistent extraction, difficulty in component identification, low data processing efficiency, and insufficient toxicity assessment in the prior art.
[0009] A smart evaluation method for extracts from medical devices includes the following steps: Using a suitable standardized reagent kit, extracts from medical devices are extracted in a standardized manner under specific extraction conditions; the standardized reagent kit includes gradient polarity standard solvents, chemically stable monitoring standards, and low-adsorption extraction containers. Extract samples are sent to a chromatography-mass spectrometry system for analysis. Mass spectra and chromatograms of each component are acquired in real time and stored as raw data. The raw data is then input into the intelligent evaluation platform. After preprocessing the raw data, the intelligent evaluation platform automatically extracts spectral features using deep learning algorithms; The spectral characteristics of the extract are compared with a pre-set database to identify known chemical components; Based on the identified chemical components, the toxicity and biocompatibility of each chemical component were assessed using a toxicity analysis model; at the same time, the changes of the monitoring standard before and after extraction were analyzed to assess whether the extraction process was stable. If the extraction process is determined to be stable, a report will be automatically generated based on the evaluation results. The report includes chemical composition, risk assessment, toxicity level, and optimization recommendations.
[0010] Preferably, the gradient polarity standard solvent includes low-polarity solvent, medium-polarity solvent, and high-polarity solvent; depending on the type of medical device extract, a certain proportion of low-polarity solvent, medium-polarity solvent, and high-polarity solvent mixed solvent is used for the extraction of medical device extract samples.
[0011] Preferably, the deep learning algorithm is a convolutional neural network, and the process of obtaining spectral features through a convolutional neural network includes the following steps: The original data is preprocessed and standardized, and the standardized original data is converted into a two-dimensional matrix suitable for input to a convolutional neural network, where the first dimension represents the mass-to-charge ratio and the second dimension represents the sample dimension and / or the time dimension. The two-dimensional matrix is input into a convolutional neural network, and a convolution operation is performed on the two-dimensional matrix through at least one convolutional layer to generate a feature map; Perform multi-level convolution operations on the feature map to extract multi-level features from low to high levels; Pooling is performed after the convolutional layers to reduce the dimensionality of the feature maps; The pooled features are flattened, and the Lasso algorithm is used to select features from the flattened features to remove redundant features. The selected features are input into a fully connected layer and processed by an activation function, outputting a feature vector for subsequent prediction or classification tasks.
[0012] Preferably, the Lasso algorithm achieves feature selection by compressing the weights of irrelevant features to zero and outputs a subset of features that are relevant to subsequent chemical composition classification or toxicity prediction.
[0013] Preferably, the feature vector output by the convolutional neural network is a spectral feature, which includes at least peak intensity, retention time, and mass-to-charge ratio.
[0014] Preferably, the database stores standard spectra and corresponding chemical components. The similarity between the extract sample and the standard spectra in the database is calculated by a matching algorithm to identify the corresponding known chemical components.
[0015] Preferably, the matching algorithm includes one or more of the following: Pearson correlation coefficient algorithm, cosine similarity algorithm, and dynamic time warping algorithm.
[0016] Preferably, the toxicity analysis model includes a quantitative structure-activity relationship (QSAR) model and a threshold toxicity concentration (THC) model. The QSAR model is used to predict the biological activity of a chemical component based on its chemical structure, and the TTC model is used to determine the safety of the chemical component based on a set safety threshold.
[0017] Preferably, in the process of chemical component identification, for chemical components that do not match perfectly or are unknown, a graph neural network is used for deep learning analysis, and the specific process is as follows: When the spectral characteristics of the extract sample do not match the preset threshold with the preset chemical component database or there are unknown chemical components that cannot be identified, the chemical components to be identified are constructed as graph structure data and input into the graph neural network for deep learning analysis. Chemical molecules or components are used as graph nodes, chemical bonds or interrelationships between molecules are used as graph edges, and at least one of mass-to-charge ratio, molecular weight or peak intensity is used as node features. The graph convolutional layer performs weighted aggregation of the features of adjacent nodes and iteratively updates the node representation. Then, the updated node feature vector is input into a fully connected layer for feature fusion and optimization, and outputs at least one of the following: classification result, concentration prediction, or risk score of the unknown chemical component.
[0018] The second objective of this invention is to provide an intelligent evaluation system for extracts from medical devices, comprising: A standardized reagent kit, comprising gradient polarity standard solvents, chemically stable monitoring standards, and low-adsorption extraction containers, is used for standardized extraction of extracts from medical devices under specific extraction conditions. The non-target analysis module is used to analyze extract samples through a chromatography-mass spectrometry system, acquire mass spectra and chromatograms of each component in real time and store them as raw data, and input the raw data into the intelligent evaluation platform; The intelligent assessment platform includes: The data processing module is used to preprocess the raw data and automatically extract spectral features using deep learning algorithms; The data comparison and identification module is used to compare the spectral characteristics of the extract with a preset database to identify known chemical components; The risk assessment module is used to evaluate the toxicity and biocompatibility of each chemical component using a toxicity analysis model; at the same time, it analyzes and monitors the changes of the standard sample before and after extraction to assess whether the extraction process is stable. The report generation module is used to automatically generate reports based on the analysis results. The reports include chemical composition, risk assessment, toxicity level, and optimization recommendations.
[0019] The beneficial effects of this invention are: By using standardized reagent kits, the standardized extraction process was ensured, eliminating differences between different laboratories, experimental conditions, and procedures. This guaranteed the stability and reproducibility of each extraction process, greatly improving the consistency of extract analysis results and ensuring data reliability. Furthermore, the introduction of chemically stable monitoring standards helps distinguish between genuine extracts and experimental artifacts, making the data more closely reflect actual usage risks and enhancing the scientific rigor of the assessment.
[0020] By employing chromatography-mass spectrometry (GC-MS) and an intelligent evaluation platform, efficient and accurate analysis of chemical components in extracts from medical devices can be achieved. By introducing graph neural networks, not only can known chemical components be identified, but also unknown components in the extracts can be analyzed through deep learning, providing classification results, concentration predictions, or risk scores, thus improving the comprehensiveness and accuracy of the analysis.
[0021] By employing quantitative structure-activity relationship (QSAR) and threshold toxicity concentration (TTC) models to assess the toxicity and biocompatibility of each chemical component in the extract, scientific and quantitative risk assessments can be conducted based on the molecular characteristics and concentrations of the chemical components. This makes the assessment of the toxicity and biocompatibility of medical device materials more accurate and provides reliable safety predictions for subsequent product development. Attached Figure Description
[0022] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the present invention; Figure 2 This is the standard reagent kit of the present invention; Figure 3 This is the page display of the intelligent evaluation platform of the present invention. Detailed Implementation
[0023] Example 1 like Figure 1 As shown, a smart evaluation method for extracts from medical devices includes the following steps: S1: Standardized Extraction Select a suitable standardized reagent kit based on the extract of the medical device to be tested (such as polymers, metals, coatings, etc.), and use the standardized reagent kit to perform standardized extraction of the medical device extract under specific extraction conditions (such as temperature, extraction time, solvent concentration, etc.).
[0024] Among them, such as Figure 2 As shown, the standardized kit includes gradient polarity standard solvents, chemically stable monitoring standards, and specially treated low-adsorption extraction containers required for extraction. Using the standardized kit ensures the stability and reproducibility of the extraction process.
[0025] Its gradient polarity standard solvents include low-polarity solvents (such as n-hexane), medium-polarity solvents (such as alcohols including ethanol and isopropanol), and high-polarity solvents (water). In experiments such as liquid chromatography and extraction analysis, the gradient polarity solvents are mixed and used to gradually change the polarity of the solvents, thereby dissolving different components in the sample step by step. This gradient change helps to achieve the separation and extraction of complex samples, improving the sensitivity and accuracy of the analysis.
[0026] The monitoring standards (such as caffeine, benzoic acid, etc.) are chemically stable and will not react with medical device materials. By monitoring the changes of the monitoring standards before and after extraction (such as recovery rate, degradation products, etc.), the deviations introduced by the extraction process itself can be assessed and corrected, which can be used to distinguish between "extracts" and "process products".
[0027] Specifically, the extract of the medical device to be tested is mixed with a gradient polarity standard solvent in a specified ratio, and extraction is performed under specific extraction conditions following a standard operating procedure. During extraction, stirring, ultrasonic treatment, or heating are used to improve extraction efficiency and ensure sufficient contact between the solvent and the sample. After extraction, insoluble matter is removed by centrifugation or filtration to obtain a clear liquid extract, which is used as the sample for subsequent analysis.
[0028] Common extraction methods include ultrasound-assisted extraction and thermal reflux extraction. For example, in an ultrasonic extraction device, the temperature is set to 50°C, the extraction time to 30 minutes, and a specified solvent is used. Ensure that the extraction process does not exceed standardized operating conditions.
[0029] Furthermore, during the extraction process, online monitoring sensors (such as temperature sensors and solvent concentration sensors) are used to monitor key parameters in real time to ensure that the temperature and concentration of the solvent are maintained within the specified range.
[0030] By introducing standardized reagent kits, the inconsistency of extracts caused by different extraction conditions and differences between laboratories can be effectively improved, providing high-quality samples for subsequent analysis.
[0031] S2: Non-target analysis The extract sample is fed into a chromatography-mass spectrometry system (such as LC-MS or GC-MS), and appropriate chromatographic column, mobile phase conditions, and mass spectrometry scan mode are set. The chromatographic portion of the chromatography-mass spectrometry system is used to separate individual components in complex samples, while the mass spectrometry portion is used to accurately measure the mass and structure of each component.
[0032] During the chromatography-mass spectrometry analysis, the mass spectrum and chromatogram of each component are acquired in real time and stored as raw data, which is then input into the intelligent evaluation platform.
[0033] S3: Data Preprocessing and Standardization The intelligent evaluation platform performs noise reduction and smoothing on the raw data to remove interference signals caused by solvent residue, background noise, or instrument errors, ensuring data quality and reliability. After data preprocessing, important features, including peak intensity, retention time, and mass-to-charge ratio, are automatically extracted from the processed data using deep learning algorithms (such as convolutional neural networks). These features serve as the basis for subsequent comparison and identification. Using deep learning technology, key patterns in the data can be automatically identified and noise removed, improving the accuracy and effectiveness of the data.
[0034] S4: Data Comparison and Chemical Composition Identification The spectral characteristics of the extract are compared with a pre-set chemical composition database. The comparison process uses a matching algorithm to identify known chemical components by calculating the similarity between the extract and the standard spectra in the database.
[0035] The database needs to be built and updated regularly to include the latest chemical composition data, and its quality and accuracy must be ensured.
[0036] The standard spectral data stored in the database needs to cover a large number of common chemical components, including: Drug components: such as antibiotics, antiviral drugs, chemotherapy drugs, etc.
[0037] Chemical reagents: Common chemical reagents used in medical devices.
[0038] Additives: such as plasticizers and stabilizers.
[0039] Matching algorithms, including: Pearson correlation coefficient: measures the degree of linear correlation between two sets of data, and is used to determine the similarity between the spectral features of the extract and the standard spectra in the database.
[0040] Cosine similarity: Calculates the angle between two vectors to measure the similarity between two spectral features, and is especially suitable for complex high-dimensional data.
[0041] Dynamic Time Warping (DTW): When processing data of different lengths or irregularities, spectral data can be aligned to ensure accurate matching.
[0042] In the specific implementation process, one or more of the following algorithms can be selected: Pearson correlation coefficient matching algorithm, cosine similarity matching algorithm, and dynamic time warping matching algorithm, in order to improve the recognition accuracy.
[0043] S5: Risk Assessment Based on the identified chemical components, the toxicity and biocompatibility of each chemical component in the extract were assessed using quantitative structure-activity relationship (QSAR) and threshold toxicity concentration (TTC) models. Simultaneously, changes in monitoring standards before and after extraction were calculated and analyzed to assess the stability of the extraction process; only when the extraction process is stable is the generated report meaningful.
[0044] Specifically, the QSAR model is used to predict the toxicity of each chemical component, and its overall risk score is calculated based on its concentration in the extract. The TTC model is used to assess the toxicity threshold of the components, ensuring the safety of the extract in use. Based on these assessment results, the system generates a risk level report for each component, marking high-risk components as red warnings and low-risk components as green safe, and providing optimization suggestions for product developers.
[0045] QSAR model: This model models the relationship between chemical components and their biological activities using molecular descriptors (such as molecular weight, polarity, solubility, functional groups, etc.). It is a computational method for predicting molecular biological activities through chemical structure. The system inputs identified chemical components into the QSAR model to predict their potential biological activities (such as toxicity, carcinogenicity, allergenicity, etc.). For example, some chemical components may exhibit high cytotoxicity or carcinogenic potential.
[0046] The TTC model is used to assess the safety of chemical substances, especially when sufficient toxicological data is lacking. It sets a safety threshold; if the concentration of a chemical component is below this threshold, it is considered harmless to humans; if it exceeds the threshold, further evaluation is required. The system assesses the toxicity level of each component based on its concentration and the standard thresholds of the TTC model.
[0047] S6: Report Generation Detailed reports are automatically generated based on the analysis results, including chemical composition, risk assessment, toxicity level, and optimization recommendations. Users can select different report dimensions according to their needs and customize the report format and content. Reports support multiple formats such as PDF and Word, facilitating submission to relevant departments or use for subsequent analysis. Report generation not only improves work efficiency but also ensures data consistency and traceability.
[0048] Example 2 Based on the intelligent evaluation method for medical device extracts described in Example 1, a polymer infusion tube was tested and evaluated. The specific process is as follows: Standardized extraction: Take the infusion tube sample and cut it into small pieces. Using the three solvents (water, 50% ethanol, and n-hexane) provided in the standardized kit, soak the sample in the kit's dedicated container at 70°C for 24 hours. Simultaneously, add a known concentration of a monitoring standard (such as caffeine or benzoic acid) to each extract.
[0049] Non-target analysis: The cooled extract was filtered and analyzed using a standard laboratory-equipped liquid chromatography-quadrupole time-of-flight mass spectrometer (LC-QTOF-MS), with raw data acquired in full scan mode.
[0050] Intelligent Assessment: Raw mass spectrometry data files are uploaded to the intelligent assessment platform for medical device extracts. The platform automatically processes the data, extracts relevant spectral features, and matches them with antioxidant (e.g., BHT) and plasticizer (e.g., DEHP) fragments from its dedicated database. For several high-abundance unknown peaks, the platform predicts they may be oligomers based on mass spectrometry fragments and preliminarily determines their low genotoxicity risk using a QSAR model. The intelligent assessment platform also calculated the recoveries of the "monitoring standards," which were all between 85% and 115%, indicating a stable extraction process.
[0051] Report Output: The platform generates a report listing all identified and unidentified but assessed compounds, their estimated concentrations, toxicity risk levels (low, medium, high), and includes a data quality statement. The report concludes that the substances extracted from the infusion tubing under simulated harsh conditions are of acceptable risk according to the toxicological risk assessment.
[0052] Example 3 The intelligent evaluation method for medical device extracts described in Example 1, which employs a convolutional neural network (CNN) for feature extraction, specifically includes the following steps: Data Preprocessing and Formatting: The mass spectrometry data extracted from the medical device is preprocessed. This mass spectrometry data includes the mass-to-charge ratio (m / z) and corresponding peak intensities. Data standardization, such as normalizing peak intensities, removes biases between different experiments, ensuring all data are compared under a unified standard. The standardized mass spectrometry data is then transformed into a two-dimensional matrix format suitable for input to convolutional neural networks (CNNs). One dimension of the matrix represents the mass-to-charge ratio in the mass spectrometry data, while the second dimension corresponds to other variables such as samples or time. This matrix format allows the mass spectrometry data to be processed like an image.
[0053] Convolution Operations and Feature Extraction: The transformed mass spectrometry data matrix is processed using convolutional layers of a convolutional neural network. This convolutional layer employs multiple convolutional kernels to perform convolution operations on the data, each kernel having a predetermined size, such as 3×3 or 5×5. The convolutional kernels slide across the input mass spectrometry data matrix, extracting local feature patterns in the data through matrix multiplication of local regions, such as peak shape, intensity, and mass-to-charge ratio in the mass spectrum. Each convolution operation generates a new feature map, which represents the local features present in the input data.
[0054] Multi-layer convolutional operations: A multi-layer convolutional structure is employed, with each layer's convolutional kernel extracting features at different levels. The kernel of the first convolutional layer focuses on extracting basic patterns in the mass spectrometry data, such as the shape and amplitude of a single peak. As the network depth increases, the kernels of subsequent convolutional layers can recognize more complex patterns, such as relationships between peaks and combinations of multiple peaks. Through multi-layer convolutional operations, the convolutional neural network can progressively abstract high-level features from the mass spectrometry data.
[0055] Pooling Layers and Feature Dimensionality Reduction: After each convolutional layer, a pooling layer is applied to reduce the dimensionality of the generated feature map. The pooling operation uses a sliding window to select the maximum or average value of a local region, thereby reducing the dimensionality of the data while preserving the most salient features. Pooling helps reduce computational complexity and enhances the ability of convolutional neural networks to recognize local features.
[0056] Feature selection and dimensionality reduction: To improve processing efficiency and reduce model complexity, the Lasso algorithm is used to select and optimize the features extracted by the convolutional neural network. The Lasso algorithm can effectively compress the weights of irrelevant features to zero, remove redundant features, and retain the features most relevant to subsequent analysis, thereby reducing computation and improving model accuracy.
[0057] Fully Connected Layer and Final Feature Output: After processing through multiple convolutional and pooling layers, the extracted features are flattened and input into the fully connected layer. The fully connected layer linearly combines multiple features with the output, introduces non-linearity through an activation function (such as ReLU), and performs multiple feature abstractions and combinations, ultimately outputting a feature vector for prediction or classification. The output of the fully connected layer will serve as input for subsequent tasks (such as chemical composition classification or toxicity prediction).
[0058] By automatically extracting key information from mass spectrometry data using convolutional neural networks, the traditional manual feature selection process is avoided, significantly improving the automation and accuracy of data processing. Multiple convolutional layers extract low- to high-level features from the data, enabling in-depth exploration of complex patterns and details within the mass spectrometry data, further enhancing the accuracy of component identification and analysis. The combination of pooling layers and the Lasso algorithm reduces redundant features and computational complexity while ensuring the preservation of important features, improving model efficiency and interpretability.
[0059] Example 4 Furthermore, based on the intelligent evaluation method for medical device extracts described in Example 1, in the process of data comparison and chemical component identification, for chemical components that do not match completely or are unknown, deep learning analysis is performed using graph neural networks (GNNs), which can automatically identify the relationships between complex chemical components and improve the ability to identify unknown components.
[0060] The process of using graph neural networks (GNNs) for deep learning analysis of the chemical composition of extracts from medical devices is as follows: Unknown chemical components in the extract are constructed as graph-structured data, where each node represents a chemical molecule or component, and edges represent chemical bonds or relationships between molecules. Each node carries characteristic data of its molecule (such as mass-to-charge ratio, molecular weight, peak intensity, etc.), and each edge represents the chemical connection between nodes.
[0061] The input graph-structured data is passed into the network through graph convolutional layers (GCN). Each node propagates information based on the features of its neighboring nodes and, combined with its own features, calculates a new node representation. This process is iterated multiple times through graph convolution operations, thereby gradually updating the feature representation of the nodes.
[0062] In graph neural networks, graph convolutional layers aggregate information from neighboring nodes through a weighted average. Specifically, a node's representation is obtained by a weighted average of the features of its neighboring nodes, with the weights learned by the network. Graph convolutional layers can capture local and global dependencies between nodes, thereby improving the accuracy of relationship modeling between components.
[0063] After graph convolution, the feature vectors of the nodes are passed to fully connected layers for feature fusion and optimization. Fully connected layers learn high-level features between nodes, transforming node representations into outputs suitable for downstream tasks, such as classification results, concentration prediction, or risk scoring.
[0064] The model is trained and optimized using either cross-entropy loss or mean squared error loss. Backpropagation is then used to adjust the weights of the graph convolutional and fully connected layers, ensuring the model accurately learns the relationships and features between chemical components.
[0065] In summary, representing chemical components and their relationships using graph structures and utilizing graph neural networks to capture complex intermolecular relationships enhances the ability to identify components and learn chemical structures. Furthermore, through graph convolution operations, nodes can acquire information from their neighboring nodes, thereby progressively learning richer feature representations. Moreover, graph neural networks can automatically adjust the information propagation methods between nodes based on data, optimizing the feature learning process and enhancing the model's generalization ability.
[0066] Example 5 A smart evaluation system for extracts from medical devices includes a standardized reagent kit, a non-target analysis module, and a smart evaluation platform, wherein the standardized reagent kit and the non-target analysis module are located on the user end, and the smart evaluation platform is located on the server end.
[0067] Users select the appropriate standardized reagent kit based on the type of extract from medical devices and perform standardized extraction of the extracts under specific extraction conditions.
[0068] The non-target analysis module is a chromatography-mass spectrometry (GC-MS) system available on the user end. It analyzes the extract samples through the GC-MS system, acquires the mass spectra and chromatograms of each component in real time and stores them as raw data, and then inputs the raw data into the intelligent evaluation platform.
[0069] The intelligent assessment platform includes a data processing module, a data comparison and identification module, a risk assessment module, and a report generation module. The platform's page display is as follows: Figure 3 As shown.
[0070] The data processing module preprocesses the raw data and automatically extracts spectral features using deep learning algorithms. The data comparison and identification module compares the spectral features of the extracted material with a pre-set database to identify known chemical components. The risk assessment module evaluates the toxicity and biocompatibility of each chemical component using a toxicity analysis model; simultaneously, it analyzes and monitors changes in the standard sample before and after extraction to assess the stability of the extraction process. The report generation module automatically generates a report based on the analysis results, including chemical components, risk assessment, toxicity level, and optimization recommendations.
[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A smart evaluation method for extracts from medical devices, characterized in that, Includes the following steps: Using a suitable standardized reagent kit, extracts from medical devices are extracted in a standardized manner under specific extraction conditions; the standardized reagent kit includes gradient polarity standard solvents, chemically stable monitoring standards, and low-adsorption extraction containers. Extract samples are sent to a chromatography-mass spectrometry system for analysis. Mass spectra and chromatograms of each component are acquired in real time and stored as raw data. The raw data is then input into the intelligent evaluation platform. After preprocessing the raw data, the intelligent evaluation platform automatically extracts spectral features using deep learning algorithms; The spectral characteristics of the extract are compared with a pre-set database to identify known chemical components; Based on the identified chemical components, toxicity and biocompatibility of each chemical component are assessed using a toxicity analysis model; At the same time, we analyzed the changes in the monitoring standard before and after extraction to assess whether the extraction process was stable. If the extraction process is determined to be stable, a report will be automatically generated based on the evaluation results. The report includes chemical composition, risk assessment, toxicity level, and optimization recommendations.
2. The intelligent evaluation method for extracts from medical devices according to claim 1, characterized in that, The gradient polarity standard solvents include low-polarity solvents, medium-polarity solvents, and high-polarity solvents; depending on the type of medical device extract, a certain proportion of low-polarity solvents, medium-polarity solvents, and high-polarity solvents are mixed for the extraction of medical device extract samples.
3. The intelligent evaluation method for extracts from medical devices according to claim 1, characterized in that, The deep learning algorithm is a convolutional neural network. The process of obtaining spectral features through a convolutional neural network includes the following steps: The original data is preprocessed and standardized, and the standardized original data is converted into a two-dimensional matrix suitable for input to a convolutional neural network, where the first dimension represents the mass-to-charge ratio and the second dimension represents the sample dimension and / or the time dimension. The two-dimensional matrix is input into a convolutional neural network, and a convolution operation is performed on the two-dimensional matrix through at least one convolutional layer to generate a feature map; Perform multi-level convolution operations on the feature map to extract multi-level features from low to high levels; Pooling is performed after the convolutional layers to reduce the dimensionality of the feature maps; The pooled features are flattened, and the Lasso algorithm is used to select features from the flattened features to remove redundant features. The selected features are input into a fully connected layer and processed by an activation function, outputting a feature vector for subsequent prediction or classification tasks.
4. The intelligent evaluation method for extracts from medical devices according to claim 3, characterized in that, The Lasso algorithm achieves feature selection by compressing the weights of irrelevant features to zero and outputs a subset of features that are relevant to subsequent chemical composition classification or toxicity prediction.
5. The intelligent evaluation method for extracts from medical devices according to claim 3, characterized in that, The feature vector output by the convolutional neural network is a spectral feature, which includes at least peak intensity, retention time, and mass-to-charge ratio.
6. The intelligent evaluation method for extracts from medical devices according to claim 1, characterized in that, The database stores standard spectra and corresponding chemical components. By using a matching algorithm to calculate the similarity between the extract sample and the standard spectra in the database, the corresponding known chemical components can be identified.
7. The intelligent evaluation method for extracts from medical devices according to claim 6, characterized in that, The matching algorithm includes one or more of the following: Pearson correlation coefficient algorithm, cosine similarity algorithm, and dynamic time warping algorithm.
8. The intelligent evaluation method for extracts from medical devices according to claim 1, characterized in that, The toxicity analysis model includes a quantitative structure-activity relationship (QSAR) model and a threshold toxicity concentration (THC) model. The QSAR model is used to predict the biological activity of chemical components based on their chemical structures. The THC model is used to determine the safety of chemical components based on set safety thresholds.
9. The intelligent evaluation method for extracts from medical devices according to any one of claims 1 to 8, characterized in that, In the process of chemical component identification, for chemical components that do not match perfectly or are unknown, graph neural networks are used for deep learning analysis. The specific process is as follows: When the spectral characteristics of the extract sample do not match the preset threshold with the preset chemical component database or there are unknown chemical components that cannot be identified, the chemical components to be identified are constructed as graph structure data and input into the graph neural network for deep learning analysis. Chemical molecules or components are used as graph nodes, chemical bonds or interrelationships between molecules are used as graph edges, and at least one of mass-to-charge ratio, molecular weight or peak intensity is used as node features. The graph convolutional layer performs weighted aggregation of the features of adjacent nodes and iteratively updates the node representation. Then, the updated node feature vector is input into a fully connected layer for feature fusion and optimization, and outputs at least one of the following: classification result, concentration prediction, or risk score of the unknown chemical component.
10. An intelligent evaluation system for extracts from medical devices, characterized in that, include: A standardized reagent kit, comprising gradient polarity standard solvents, chemically stable monitoring standards, and low-adsorption extraction containers, is used for standardized extraction of extracts from medical devices under specific extraction conditions. The non-target analysis module is used to analyze extract samples through a chromatography-mass spectrometry system, acquire mass spectra and chromatograms of each component in real time and store them as raw data, and input the raw data into the intelligent evaluation platform; The intelligent assessment platform includes: The data processing module is used to preprocess the raw data and automatically extract spectral features using deep learning algorithms; The data comparison and identification module is used to compare the spectral characteristics of the extract with a preset database to identify known chemical components; The risk assessment module is used to evaluate the toxicity and biocompatibility of each chemical component using a toxicity analysis model; at the same time, it analyzes and monitors the changes of the standard sample before and after extraction to assess whether the extraction process is stable. The report generation module is used to automatically generate reports based on the analysis results. The reports include chemical composition, risk assessment, toxicity level, and optimization recommendations.