Raman spectrum pathogen rapid identification and detection system

Abstract

The invention discloses a Raman spectrum pathogen rapid identification and detection system, and relates to the technical field of pathogen detection, and the system comprises a spectrum collection module which obtains the Raman original spectrum data of a pathogen sample and the sample correlation multi-dimensional information, and generates a pathogen Raman fusion data set; the double-branch identification module is used for extracting cross-domain features based on the fusion data set and carrying out germ type matching to generate a preliminary germ type matching result; the risk level judgment module is used for constructing a risk assessment model and generating a pathogen risk level and a corresponding early warning tendency based on the preliminary matching result of the pathogen types in combination with a pathogen virulence factor database and clinical pathogenic probability statistical data; and the result verification module generates a detection report. According to the method, the Raman spectrum and the multi-dimensional sample information are fused, and rapid and accurate recognition and scientific risk grading of germs are realized by means of double-branch recognition, multi-dimensional risk assessment and an integrated learning verification mechanism.

Description

Technical Field

[0001] This invention relates to the field of pathogen detection technology, specifically to a rapid identification and detection system for pathogens using Raman spectroscopy. Background Technology

[0002] Rapid and accurate identification and risk assessment of pathogens are core aspects of clinical diagnosis and treatment, disease prevention and control, and public health security. Among traditional pathogen detection methods, culture is the gold standard, but it has a long detection cycle and cannot meet the rapid response needs of emerging infectious diseases. Nucleic acid amplification technology relies on specific primer design, which has poor adaptability to novel and variant pathogens and is easily affected by sample impurities, leading to false positives. Single Raman spectroscopy detection technology only focuses on spectral feature extraction, resulting in insufficient accuracy in pathogen identification and a high rate of missed and false detections.

[0003] At the same time, existing detection systems generally lack a comprehensive risk assessment system, and simply determine the risk level based on the type of pathogen without combining multi-dimensional indicators such as virulence factors, transmissibility and clinical cure rate for comprehensive evaluation, resulting in weak targeted early warning.

[0004] Solving this technical problem is a technical challenge that needs to be overcome by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned technical problems, a rapid identification and detection system for pathogens using Raman spectroscopy is provided. This technical solution resolves the problems described above.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A rapid identification and detection system for pathogens using Raman spectroscopy includes: a spectral acquisition module, a dual-branch identification module, a risk level determination module, and a result verification module; The spectral acquisition module is used to acquire the raw Raman spectral data of the pathogen samples and multi-dimensional information on sample association, and generate a pathogen Raman fusion dataset. The dual-branch recognition module is electrically connected to the spectral acquisition module and is used to extract cross-domain features and perform pathogen type matching based on the fused dataset to generate preliminary pathogen type matching results. The risk level determination module is electrically connected to the dual-branch identification module. It is used to construct a risk assessment model based on the preliminary matching results of pathogen types, combined with the pathogen virulence factor database and clinical pathogenicity probability statistics, and generate pathogen risk levels and corresponding early warning tendencies. The result verification module is electrically connected to the risk level determination module. It is used to generate pathogen identification and detection results and generate a detection report based on the pathogen risk level and early warning tendency, by using an integrated learning strategy to fuse the prediction confidence of multiple models and filtering through thresholds.

[0007] Preferably, the dual-branch recognition module includes: The cross-domain feature extraction unit employs an attention-enhanced convolutional neural network. First, it segments the spectral data in the pathogen Raman fusion dataset into fingerprint and feature peak regions, extracting the peak position, intensity distribution, and peak shape features. Simultaneously, it extracts texture, contour, and detail features from the morphological feature image. Through a modal attention weight allocation mechanism, it maps different modal features to the same feature space, transforming them into feature vectors of a unified dimension. The dual-branch modeling unit constructs a lightweight Transformer branch and a residual network branch. The Transformer branch performs global semantic association modeling on the feature vectors through a multi-head attention mechanism, capturing deep associations between cross-modal features. The residual network branch mines local detail features through residual connection structures, avoiding gradient vanishing. The output features of the two branches are concatenated according to dynamic weights to generate cross-domain fusion features that combine global associations and local details.

[0008] Preferably, the dual-branch recognition module further includes: The species matching unit calculates the cosine similarity between the cross-domain fusion features and the standard features in the pathogen spectral feature library. It then uses a K-nearest neighbor clustering matching algorithm to filter and verify the similarity ranking results, generating preliminary pathogen species matching results containing Top-N candidate species, corresponding matching similarities, and key feature matching items. The cosine similarity calculation formula is as follows: , In the formula, where, Cosine similarity between cross-domain fusion features and standard features; The cross-domain fusion feature vector output by the dual-branch modeling unit. For the vector of the first Feature values ​​in each dimension; This represents the standard feature vector of a certain type of pathogen in a pathogen spectral feature library. For the standard vector of the first Feature values ​​in each dimension; This is the uniform dimension of the feature vector.

[0009] Preferably, the risk level determination module includes: The virulence factor association unit, based on the preliminary matching results of pathogen types, calls the pathogen virulence factor database to extract the key virulence parameters of the corresponding pathogens, such as pathogenic genes, toxin secretion capacity, and invasiveness, and establishes an association mapping between virulence characteristics and pathogen types. The risk assessment model construction unit combines clinical pathogenicity probability statistics, epidemiological monitoring results, and clinical treatment effect data. It uses the analytic hierarchy process to assign scientific weights to assessment indicators such as virulence parameters, transmissibility, clinical cure rate, and susceptible population range, and establishes a multi-dimensional, multi-indicator risk assessment model. The early warning tendency generation unit calculates a comprehensive risk value for pathogens by comprehensively weighting various indicators using a risk assessment model. Based on preset standards, it classifies risks into high, medium, and low levels, generating targeted early warning tendencies for each level. High risk triggers emergency prevention and control warnings and priority treatment recommendations; medium risk suggests enhanced monitoring and precise intervention; and low risk provides routine prevention and control guidance and follow-up recommendations. The formula for calculating the comprehensive risk value is as follows: , In the formula, This represents the overall risk value of pathogens. For the first The weight of each evaluation indicator, For the first Standardized scores for each evaluation indicator This represents the total number of assessment indicators included in the risk assessment model.

[0010] Preferably, the result verification module includes: The ensemble learning fusion unit uses three basic models—random forest, support vector machine, and logistic regression—to perform secondary calculations on the prediction results of the dual-branch recognition module. Through a voting mechanism and a probability fusion algorithm, the prediction results of each model are combined to calculate the overall confidence score. The confidence threshold determination unit sets a graded confidence screening standard. Results with a comprehensive confidence level of ≥95% are directly confirmed, results with a confidence level of 85%-94% are subject to secondary verification through feature matching, and samples with a confidence level of <85% are marked as pending verification. At the same time, the key reasons for not meeting the standard are recorded. The test result calibration unit establishes an error correction model based on historical test data, performs systematic deviation calibration and accuracy optimization on the confirmed identification results, and adjusts the correction coefficients using clinical feedback data.

[0011] Preferably, the spectral acquisition module further includes: The sample adaptation unit is equipped with replaceable dedicated sampling probes and sample preprocessing components for different forms of pathogen samples, including solid, liquid, and aerosol samples. Through modular design, it enables rapid adaptation and efficient collection of samples in multiple scenarios. The real-time monitoring unit integrates a visual interface for spectral acquisition progress, providing real-time feedback on the success rate of characteristic peak identification, data signal-to-noise ratio, and acquisition completion indicators. It also establishes an anomaly monitoring mechanism, automatically triggering a re-acquisition command when the indicators fall below the preset standards and recording the abnormal situation.

[0012] Preferably, the dual-branch recognition module further includes: The feature enhancement unit uses generative adversarial networks to augment the spectral features of rare pathogens and simulates spectral changes under different detection environments and sample conditions through style transfer technology. The rapid response unit performs quantization compression and hardware acceleration adaptation on the dual-branch model, optimizes the computation process of feature extraction and matching algorithms, reduces redundant calculations, and improves data processing efficiency through parallel computing technology, shortening the single-sample detection cycle to the range required for rapid clinical diagnosis.

[0013] Preferably, the risk level determination module further includes: The data update unit establishes a data synchronization mechanism to synchronize the latest clinical pathogenic data, pathogen mutation monitoring results, the latest epidemiological developments, and treatment plan updates in real time. It adopts an online learning approach to dynamically adjust the weight parameters and level determination criteria of the risk assessment model to ensure the timeliness and accuracy of risk assessment. Personalized early warning units allow for customization of the level of detail and presentation format of warning content based on the needs of different application scenarios. In clinical scenarios, the focus is on supplementing medication recommendations and treatment pathway guidance; in disease control scenarios, the analysis of transmission routes and suggestions on the scope of prevention and control are strengthened; and in food safety supervision scenarios, pollution source tracing prompts and control measures are added.

[0014] Preferably, the result verification module further includes: The anomaly marking unit automatically locates spectral data anomalies for samples marked as to be reviewed. These anomalies include missing characteristic peaks, excessive noise, baseline drift, feature matching discrepancies, and model prediction divergence points. It then generates a review guidance report to identify key review points and detection directions. The traceability unit establishes a data log for the entire testing process, recording sample information, instrument parameters, model version, operator information, and testing time information, supporting full traceability of test results and historical data backtracking and query.

[0015] Preferably, the spectral acquisition module includes: The precision acquisition unit matches the optimal combination of excitation light parameters based on the type and state of the pathogen sample. Through iterative acquisition optimization and signal enhancement processing, it acquires high signal-to-noise ratio and low interference Raman raw spectral data. The multi-dimensional information synchronization unit simultaneously collects information on the culture environment of pathogen samples, morphological feature images under a microscope, and clinical relevance information to form multi-dimensional supplementary sample data. The data calibration unit establishes a calibration model based on a standard pathogen spectral library, and uses standard spectral data as a reference in real time to perform wavelength shift correction, intensity normalization, and baseline correction on the raw Raman spectral data.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The dual-branch identification module integrates multimodal information such as spectral and morphological data through a cross-domain feature extraction unit. It combines lightweight Transformer and residual network dual-branch modeling to capture global semantic associations and mine local detailed features. With the help of cosine similarity calculation and K-nearest neighbor clustering matching, it significantly improves the accuracy of pathogen identification. The rapid response unit significantly shortens the single-sample detection cycle through model quantization compression and parallel computing technology, meeting the needs of rapid clinical diagnosis and emergency response to public health emergencies.

[0017] 2. The risk level determination module, based on the preliminary matching results of pathogen types, integrates the virulence factor database, clinical pathogenicity statistics, and epidemiological monitoring information. It assigns scientific weights to each assessment indicator through the analytic hierarchy process and combines the comprehensive risk value calculation formula to achieve multi-dimensional quantitative assessment. It classifies high, medium, and low risk levels according to standards and generates targeted early warning tendencies, providing accurate decision-making basis for clinical diagnosis and disease control, and solving the problems of the traditional assessment method being too singular and subjective.

[0018] 3. The result verification module introduces multi-model integrated learning strategies such as random forest and support vector machine. It calculates the comprehensive confidence level through voting mechanism and probability fusion, and effectively reduces the false detection and false negative rates by combining hierarchical threshold screening and error correction model. The anomaly labeling unit and the source tracing unit record the data of the entire detection process, realize the accurate location of abnormal samples, the full source tracing of detection results and the backtracking of historical data, and meet the needs of medical quality control and dispute tracing.

[0019] 4. The sample adaptation unit of the spectral acquisition module adopts a modular design and is equipped with replaceable sampling probes and preprocessing components to achieve rapid adaptation of samples in various forms such as solid, liquid, and aerosol. The cross-device compatibility unit establishes a standardized data interface and parsing protocol to support data interconnection and interoperability of different brands and models of Raman spectrometers. Together with the remote interaction module, it enables real-time sharing of test results and remote expert review, thereby improving the coverage and convenience of testing services. Attached Figure Description

[0020] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0021] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0022] Reference Figure 1As shown, a rapid identification and detection system for pathogens using Raman spectroscopy includes: Spectrum acquisition module: The spectral acquisition module is used to acquire the raw Raman spectral data of pathogen samples and multi-dimensional information related to the samples, and generate a unified and complete pathogen Raman fusion dataset. Specifically, it includes a precise acquisition unit, a multi-dimensional information synchronization unit, a data calibration unit, a sample adaptation unit, and a real-time monitoring unit. Each unit works together to achieve efficient acquisition and data preprocessing of samples in multiple scenarios. Precision acquisition unit: Based on the type and state of the pathogen sample, the optimal combination of excitation light parameters is adaptively selected through the system's preset parameter matching library. For the signal intensity differences of different types of pathogen samples, the excitation light related parameters are adjusted to improve signal quality. Through repeated acquisition and signal superposition processing, random noise interference is reduced, and high signal-to-noise ratio and low interference Raman raw spectral data are obtained. Multi-dimensional information synchronization unit: While acquiring Raman spectra, it simultaneously acquires multi-dimensional supplementary information about the sample. Among these, the culture environment information covers key conditions such as sample culture temperature, pH value, and culture time; morphological feature images are acquired through an integrated microscope module, recording intuitive features such as the cell structure and colony morphology of pathogens; and clinical correlation information is obtained by connecting with the hospital information system to acquire background information such as the sending department, the patient's underlying diseases, and the sampling site, forming multi-dimensional and comprehensive supplementary sample data. Data calibration unit: Based on the national standard pathogen spectral library, a dynamic calibration model is established, which calls standard spectral data in real time as a reference standard to systematically calibrate the raw Raman spectral data. First, wavelength shift is corrected to eliminate the characteristic peak position deviation caused by instrument optical path fluctuations; second, intensity normalization is performed to unify the spectral intensity scale of different samples; finally, baseline correction is used to eliminate fluorescence background interference to ensure the accuracy and comparability of spectral data. Sample Adaptation Unit: Equipped with replaceable dedicated sampling probes for pathogen samples in different forms, including solid, liquid, and aerosol, to adapt to the collection needs of various samples. It also includes a sample preprocessing component, whose modular design enables rapid adaptation and efficient collection of samples in multiple scenarios, eliminating the need for complex manual operations and improving the convenience and versatility of the collection process. Real-time monitoring unit: Integrates a visual interface for spectral acquisition progress, providing real-time feedback on key indicators such as characteristic peak identification success rate, data signal-to-noise ratio, and acquisition completion rate. An anomaly monitoring mechanism is established; when indicators fall below preset standards, a re-acquisition command is automatically triggered, and the anomaly is recorded, providing a basis for subsequent analysis and troubleshooting. Data integration and processing: The calibrated raw Raman spectral data and multi-dimensional supplementary information are associated and aligned according to the unique identifier of the sample, redundant data is removed, and data format is standardized to finally generate a pathogen Raman fusion dataset, which provides comprehensive and reliable data support for subsequent feature extraction and identification.

[0023] Dual-branch recognition module: The dual-branch identification module is electrically connected to the spectral acquisition module. Based on the pathogen Raman fusion dataset, it generates preliminary pathogen species matching results through cross-domain feature extraction, dual-branch collaborative modeling, and species matching operations. Specifically, it includes a cross-domain feature extraction unit, a dual-branch modeling unit, a species matching unit, a feature enhancement unit, and a fast response unit. Cross-domain feature extraction unit: Feature extraction is performed using an attention-enhanced convolutional neural network. First, the spectral data in the pathogen Raman fusion dataset is segmented into regions to locate the fingerprint region and feature peak region of the Raman spectrum, and core spectral features such as feature peak position, intensity distribution, and peak shape parameters are extracted. Simultaneously, multi-scale convolution operations are performed on the morphological feature image to extract texture features, contour features, and detail features. Through a modal attention weight allocation mechanism, the weight ratio of spectral features and morphological features is dynamically adjusted, and the feature fusion effect is optimized according to the sample quality. Different modal features are mapped to the same high-dimensional feature space and transformed into a feature vector of a unified dimension. The dual-branch modeling unit constructs a lightweight Transformer branch and a residual network branch to achieve deep feature mining. The Transformer branch uses a multi-head attention mechanism to model global semantic associations between feature vectors, capturing deep correlations between cross-modal features. The residual network branch avoids gradient vanishing through residual connection structures, focusing on mining local detail features. The output features of the two branches are concatenated with dynamic weights to generate cross-domain fused features that combine global correlations and local details, improving the comprehensiveness of feature representation.

[0024] Species matching unit: Cosine similarity is calculated between the cross-domain fusion features and the standard features in the pathogen spectral feature library. The formula for calculating cosine similarity is: , In the formula, where, Cosine similarity between cross-domain fusion features and standard features; The cross-domain fusion feature vector output by the dual-branch modeling unit. For the vector of the first Feature values ​​in each dimension; This represents the standard feature vector of a certain type of pathogen in a pathogen spectral feature library. For the standard vector of the first Feature values ​​in each dimension; This is the uniform dimension of the feature vector.

[0025] The similarity ranking results are filtered and verified using the K-nearest neighbor clustering matching algorithm. The Top-N candidate species with the highest similarity are selected, and the corresponding matching similarity and key feature matching items are recorded to generate preliminary matching results for pathogen species. Feature Enhancement Unit: For rare samples such as rare pathogens and novel variants, generative adversarial networks are used to augment the data. Based on the spectral features of existing rare pathogens, style transfer technology is used to simulate spectral changes under different detection environments and sample states, enriching the diversity of training samples. At the same time, data augmentation operations are performed on the augmented data to improve the model's generalization ability to complex scenarios and rare pathogens, and reduce the recognition bias caused by insufficient sample size. Rapid Response Unit: This unit performs quantization compression and hardware acceleration adaptation on the dual-branch model, optimizes the computational flow of feature extraction and matching algorithms, eliminates redundant computational steps, and reduces unnecessary computation. It improves data processing efficiency through parallel computing technology, shortens the single-sample detection cycle, and meets the timeliness requirements of rapid clinical diagnosis and emergency response to public health emergencies.

[0026] Risk level assessment module: The risk level determination module is electrically connected to the dual-branch identification module. Based on the preliminary matching results of pathogen types, a risk assessment model is constructed by combining multi-source data to generate pathogen risk levels and corresponding early warning tendencies. Specifically, it includes a virulence factor association unit, a risk assessment model construction unit, an early warning tendency generation unit, a data update unit, and a personalized early warning unit. Virulence factor association unit: Based on the preliminary matching results of pathogen types, the system calls the pathogen virulence factor database through the API interface to extract the key virulence parameters of the corresponding pathogens, including core information such as pathogenic genes, toxin secretion capacity, invasiveness, and drug resistance genes, and establishes an association mapping between virulence characteristics and pathogen types, providing core basis for risk assessment; Risk assessment model construction unit: Combining clinical pathogenicity probability statistics, epidemiological monitoring results and clinical treatment effect data, the analytic hierarchy process is used to assign scientific weights to the assessment indicators. The assessment indicators cover key dimensions such as virulence parameters, transmissibility, clinical cure rate, and susceptible population range, with a total weight of 1. A multi-dimensional, multi-indicator risk assessment model is established to ensure the comprehensiveness and objectivity of the assessment results. Early warning tendency generation unit: This unit calculates the overall risk value of pathogens by comprehensively weighting various indicators using a risk assessment model. The formula for calculating the overall risk value is as follows: , In the formula, This represents the overall risk value of pathogens. For the first The weight of each evaluation indicator, For the first Standardized scores for each evaluation indicator This represents the total number of assessment indicators included in the risk assessment model.

[0027] The system categorizes risks into three levels—high, medium, and low—based on preset standards, and generates targeted early warnings for each level: high risk triggers an emergency prevention and control warning and simultaneously pushes priority treatment recommendations; medium risk prompts enhanced monitoring and precise intervention; and low risk provides routine prevention and control guidance and follow-up recommendations, offering precise guidance for prevention and control decisions in different scenarios. Data Update Unit: Establishes a real-time data synchronization mechanism, connecting with clinical data platforms, CDC monitoring systems, and international pathogen mutation databases to synchronize the latest clinical pathogenic data, pathogen mutation monitoring results, the latest epidemiological developments, and updated treatment plans in real time. Employs online learning methods to adjust the weight parameters and grading criteria of the risk assessment model, ensuring the timeliness and accuracy of risk assessments and adapting to changes in pathogen mutations and prevention and control needs. Personalized early warning unit: Supports customization of the level of detail and presentation format of early warning content according to different application scenarios. In clinical scenarios, it focuses on supplementing medication advice and treatment pathway guidance; in disease control scenarios, it strengthens transmission path analysis and prevention and control scope suggestions; in food safety supervision scenarios, it adds pollution source tracing prompts and control measures to meet the personalized needs of different use scenarios.

[0028] Result verification module: The result verification module is electrically connected to the risk level determination module. Based on the pathogen risk level and warning tendency, it generates accurate and reliable detection results and outputs a detection report through integrated learning strategy and threshold screening. Specifically, it includes an integrated learning fusion unit, a confidence threshold determination unit, a detection result calibration unit, an anomaly marking unit, and a source tracing unit. The ensemble learning fusion unit introduces multiple base models, each performing secondary calculations on the prediction results of the dual-branch recognition module. Each base model has its own specific function: capturing nonlinear correlations between features, classifying and recognizing high-dimensional features, and predicting the probability of linear features. Through a voting mechanism and a probability fusion algorithm, the prediction results of each model are combined to calculate a comprehensive confidence score, improving the stability and reliability of the prediction results. Confidence threshold determination unit: Sets graded confidence screening criteria. Results with a comprehensive confidence level of high are directly confirmed and judged as reliable identification results. Results with medium confidence are subject to secondary feature matching verification. After the verification is passed, the results are confirmed. Samples with low confidence are marked as pending review. At the same time, the key reasons for not meeting the standards are recorded to provide direction for the review work. Test Result Calibration Unit: Based on historical test data, an error correction model is established to calibrate and optimize the accuracy of confirmed identification results. The correction coefficient is dynamically adjusted in conjunction with clinical feedback data to continuously optimize test accuracy, reduce the impact of systematic errors on results, and further improve the accuracy and reliability of test results. Anomaly Marking Unit: For samples marked as requiring review, it automatically locates spectral data anomalies, feature matching discrepancies, and model prediction divergences, generates a detailed review guidance report, clarifies the review focus and detection direction, assists staff in efficiently completing the review work, and reduces the difficulty and workload of review. Traceability Unit: Establishes a data log for the entire testing process, adopts reliable data storage technology to ensure that the data is tamper-proof, records key information such as sample information, instrument parameters, model version, operators, testing time, and processing results of each module, supports full traceability of test results and historical data backtracking query, and meets the needs of medical dispute tracing, quality control and scientific research analysis; Test report generation: Integrates pathogen identification results, risk level, warning tendency, calibration information, etc., to generate a standardized test report. The report includes core contents such as basic sample information, test method description, identification results, risk level and warning suggestions, test basis, and review prompts. It supports standardized format export and cloud storage, and can be connected to relevant information systems to achieve real-time sharing, improving the practicality and dissemination efficiency of the report.

[0029] Reference Figure 2 As shown, the usage process of the Raman spectroscopy-based rapid identification and detection system for pathogens described in this invention is as follows: 101. The sample adaptation unit of the spectral acquisition module adapts to samples of different shapes, the precision acquisition unit acquires the raw Raman spectral data, the multi-dimensional information synchronization unit acquires supplementary information, and after calibration by the data calibration unit, a pathogen Raman fusion dataset is generated. 102. The cross-domain feature extraction unit of the dual-branch recognition module extracts multimodal features and transforms them into a unified feature vector. The dual-branch modeling unit generates cross-domain fusion features. The species matching unit obtains the preliminary matching results of pathogen species through cosine similarity calculation and clustering matching algorithm. 103. The risk level determination module calls the virulence factor database, constructs a risk assessment model to calculate the comprehensive risk value, classifies the risk level, and generates personalized early warning tendencies. 104. The result verification module integrates the prediction results of multiple models through integrated learning, and obtains the final detection result after threshold screening and error calibration. The anomaly marking unit and the source tracing unit ensure the reliability and traceability of the results, and finally output a standardized detection report.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A rapid identification and detection system for pathogens using Raman spectroscopy, characterized in that, include: Spectral acquisition module, dual-branch identification module, risk level determination module, and result verification module; The spectral acquisition module is used to acquire the raw Raman spectral data of the pathogen samples and multi-dimensional information on sample association, and generate a pathogen Raman fusion dataset. The dual-branch recognition module is electrically connected to the spectral acquisition module and is used to extract cross-domain features and perform pathogen type matching based on the fused dataset to generate preliminary pathogen type matching results. The risk level determination module is electrically connected to the dual-branch identification module. It is used to construct a risk assessment model based on the preliminary matching results of pathogen types, combined with the pathogen virulence factor database and clinical pathogenicity probability statistics, and generate pathogen risk levels and corresponding early warning tendencies. The result verification module is electrically connected to the risk level determination module. It is used to generate pathogen identification and detection results and generate a detection report based on the pathogen risk level and early warning tendency, by using an integrated learning strategy to fuse the prediction confidence of multiple models and filtering through thresholds.

2. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 1, characterized in that, The dual-branch recognition module includes: The cross-domain feature extraction unit employs an attention-enhanced convolutional neural network. First, it segments the spectral data in the pathogen Raman fusion dataset into fingerprint and feature peak regions, extracting the peak position, intensity distribution, and peak shape features. Simultaneously, it extracts texture, contour, and detail features from the morphological feature image. Through a modal attention weight allocation mechanism, it maps different modal features to the same feature space, transforming them into feature vectors of a unified dimension. The dual-branch modeling unit constructs a lightweight Transformer branch and a residual network branch. The Transformer branch performs global semantic association modeling on the feature vectors through a multi-head attention mechanism, capturing deep associations between cross-modal features. The residual network branch mines local detail features through residual connection structures, avoiding gradient vanishing. The output features of the two branches are concatenated according to dynamic weights to generate cross-domain fusion features that combine global associations and local details.

3. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 2, characterized in that, The dual-branch recognition module also includes: The species matching unit calculates the cosine similarity between the cross-domain fusion features and the standard features in the pathogen spectral feature library. It then uses a K-nearest neighbor clustering matching algorithm to filter and verify the similarity ranking results, generating preliminary pathogen species matching results containing Top-N candidate species, corresponding matching similarities, and key feature matching items. The cosine similarity calculation formula is as follows: ， In the formula, where, Cosine similarity between cross-domain fusion features and standard features; The cross-domain fusion feature vector output by the dual-branch modeling unit. For the vector of the first Feature values ​​in each dimension; This represents the standard feature vector of a certain type of pathogen in a pathogen spectral feature library. For the standard vector of the first Feature values ​​in each dimension; This represents the uniform dimension of the feature vectors.

4. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 3, characterized in that, The risk level determination module includes: The virulence factor association unit, based on the preliminary matching results of pathogen types, calls the pathogen virulence factor database to extract the key virulence parameters of the corresponding pathogens, such as pathogenic genes, toxin secretion capacity, and invasiveness, and establishes an association mapping between virulence characteristics and pathogen types. The risk assessment model construction unit combines clinical pathogenicity probability statistics, epidemiological monitoring results, and clinical treatment effect data. It uses the analytic hierarchy process to assign scientific weights to assessment indicators such as virulence parameters, transmissibility, clinical cure rate, and susceptible population range, and establishes a multi-dimensional, multi-indicator risk assessment model. The early warning tendency generation unit calculates a comprehensive risk value for pathogens by comprehensively weighting various indicators using a risk assessment model. Based on preset standards, it classifies risks into high, medium, and low levels, generating targeted early warning tendencies for each level. High risk triggers emergency prevention and control warnings and priority treatment recommendations; medium risk suggests enhanced monitoring and precise intervention; and low risk provides routine prevention and control guidance and follow-up recommendations. The formula for calculating the comprehensive risk value is as follows: ， In the formula, This represents the overall risk value of pathogens. For the first The weight of each evaluation indicator, For the first Standardized scores for each evaluation indicator This represents the total number of assessment indicators included in the risk assessment model.

5. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 4, characterized in that, The result verification module includes: The ensemble learning fusion unit uses three basic models—random forest, support vector machine, and logistic regression—to perform secondary calculations on the prediction results of the dual-branch recognition module. Through a voting mechanism and a probability fusion algorithm, the prediction results of each model are combined to calculate the overall confidence score. The confidence threshold determination unit sets a graded confidence screening standard. Results with a comprehensive confidence level of ≥95% are directly confirmed, results with a confidence level of 85%-94% are subject to secondary verification through feature matching, and samples with a confidence level of <85% are marked as pending verification. At the same time, the key reasons for not meeting the standard are recorded. The test result calibration unit establishes an error correction model based on historical test data, performs systematic deviation calibration and accuracy optimization on the confirmed identification results, and adjusts the correction coefficients using clinical feedback data.

6. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 5, characterized in that, The spectral acquisition module also includes: The sample adaptation unit is equipped with replaceable dedicated sampling probes and sample preprocessing components for different forms of pathogen samples, including solid, liquid, and aerosol samples. Through modular design, it enables rapid adaptation and efficient collection of samples in multiple scenarios. The real-time monitoring unit integrates a visual interface for spectral acquisition progress, providing real-time feedback on the success rate of characteristic peak identification, data signal-to-noise ratio, and acquisition completion indicators. It also establishes an anomaly monitoring mechanism, automatically triggering a re-acquisition command when the indicators fall below the preset standards and recording the abnormal situation.

7. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 6, characterized in that, The dual-branch recognition module also includes: The feature enhancement unit uses generative adversarial networks to augment the spectral features of rare pathogens and simulates spectral changes under different detection environments and sample conditions through style transfer technology. The rapid response unit performs quantization compression and hardware acceleration adaptation on the dual-branch model, optimizes the computation process of feature extraction and matching algorithms, reduces redundant calculations, and improves data processing efficiency through parallel computing technology, shortening the single-sample detection cycle to the range required for rapid clinical diagnosis.

8. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 7, characterized in that, The risk level determination module also includes: The data update unit establishes a data synchronization mechanism to synchronize the latest clinical pathogenic data, pathogen mutation monitoring results, the latest epidemiological developments, and updated treatment plans in real time. It adopts an online learning approach to dynamically adjust the weight parameters and level determination criteria of the risk assessment model to ensure the timeliness and accuracy of risk assessment. Personalized early warning units allow for customization of the level of detail and presentation format of warning content based on the needs of different application scenarios. In clinical scenarios, the focus is on supplementing medication recommendations and treatment pathway guidance; in disease control scenarios, the analysis of transmission routes and suggestions on the scope of prevention and control are strengthened; and in food safety supervision scenarios, pollution source tracing prompts and control measures are added.

9. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 8, characterized in that, The result verification module also includes: The anomaly marking unit automatically locates spectral data anomalies for samples marked as to be reviewed. These anomalies include missing characteristic peaks, excessive noise, baseline drift, feature matching discrepancies, and model prediction divergence points. It then generates a review guidance report to identify key review points and detection directions. The traceability unit establishes a data log for the entire testing process, recording sample information, instrument parameters, model version, operator information, and testing time information, supporting full traceability of test results and historical data backtracking query.

10. The rapid identification and detection system for pathogens using Raman spectroscopy according to claim 9, characterized in that, The spectral acquisition module includes: The precision acquisition unit matches the optimal combination of excitation light parameters based on the type and state of the pathogen sample. Through iterative acquisition optimization and signal enhancement processing, it acquires high signal-to-noise ratio and low interference Raman raw spectral data. The multi-dimensional information synchronization unit simultaneously collects information on the culture environment of pathogen samples, morphological feature images under a microscope, and clinical relevance information to form multi-dimensional supplementary sample data. The data calibration unit establishes a calibration model based on a standard pathogen spectral library, and uses standard spectral data as a reference in real time to perform wavelength shift correction, intensity normalization, and baseline correction on the raw Raman spectral data.

Images