Search

CN-121997192-A - Raman spectrum germ quick identification detection system

CN121997192ACN 121997192 ACN121997192 ACN 121997192ACN-121997192-A

Abstract

The invention discloses a Raman spectrum germ quick identification detection system, which relates to the technical field of germ detection and comprises a spectrum acquisition module, a double-branch identification module, a risk level judgment module, a result verification module and a detection report, wherein the spectrum acquisition module is used for acquiring Raman original spectrum data of germ samples and sample association multidimensional information to generate germ Raman fusion data sets, the double-branch identification module is used for extracting cross-domain features and performing germ type matching based on the fusion data sets to generate germ type preliminary matching results, the risk level judgment module is used for constructing a risk assessment model based on the germ type preliminary matching results and combining a germ virulence factor database and clinical pathogenic probability statistical data to generate germ risk levels and corresponding early warning trends, and the result verification module is used for generating detection reports. According to the invention, by fusing Raman spectrum and multidimensional sample information, rapid and accurate identification and scientific risk classification of germs are realized by means of double-branch identification, multidimensional risk assessment and integrated learning verification mechanisms.

Inventors

  • ZHAO XIANMING

Assignees

  • 纽勤生物科技(上海)有限公司

Dates

Publication Date
20260508
Application Date
20251231

Claims (10)

  1. 1. The Raman spectrum germ rapid identification detection system is characterized by comprising a spectrum acquisition module, a double-branch identification module, a risk level judgment module and a result verification module; The spectrum acquisition module is used for acquiring Raman original spectrum data of a pathogen sample and sample-related multidimensional information and generating a pathogen Raman fusion data set; The dual-branch identification module is electrically connected with the spectrum acquisition module and is used for extracting cross-domain characteristics and performing pathogen type matching based on the fusion data set to generate a pathogen type preliminary matching result; The risk level judging module is electrically connected with the double-branch identifying module and is used for constructing a risk assessment model based on a germ type preliminary matching result and combining a germ virulence factor database and clinical pathogenic probability statistical data to generate germ risk levels and corresponding early warning trends; The result verification module is electrically connected with the risk level judgment module and is used for fusing the multi-model prediction confidence coefficient by utilizing the integrated learning strategy based on the pathogen risk level and the early warning tendency, generating a pathogen identification detection result through threshold screening and generating a detection report.
  2. 2. The rapid identification and detection system for raman spectrum pathogens according to claim 1, wherein the dual branch identification module comprises: The cross-domain feature extraction unit adopts an attention mechanism enhanced convolutional neural network, firstly performs fingerprint region and characteristic peak region segmentation on spectrum data in a germ Raman fusion data set, extracts characteristic peak positions, intensity distribution and peak shape features, simultaneously performs texture, contour and detail feature extraction on morphological feature images, maps different modal features to the same feature space through a modal attention weight distribution mechanism, and converts the different modal features into feature vectors with uniform dimensions; The dual-branch modeling unit is used for constructing a lightweight transducer branch and a residual network branch, wherein the transducer branch carries out global semantic association modeling on feature vectors through a multi-head attention mechanism, captures deep association among cross-modal features, the residual network branch excavates local detail features through a residual connection structure to avoid gradient disappearance, and the output features of the two branches are cascaded according to dynamic weights to generate cross-domain fusion features with the global association and the local detail.
  3. 3. The rapid identification and detection system for raman spectrum pathogens according to claim 2, wherein the dual branch identification module further comprises: the category matching unit performs cosine similarity calculation on the cross-domain fusion features and standard features in the pathogen spectral feature library, screens and verifies similarity sorting results through a K neighbor cluster matching algorithm, and generates a pathogen category preliminary matching result containing Top-N candidate categories, corresponding matching similarity and key feature matching items, wherein a cosine similarity calculation formula is as follows: , Wherein, in the formula, wherein, Cosine similarity between cross-domain fusion features and standard features; The feature vectors are fused for the cross-domain output by the dual-branch modeling unit, Is the vector of Characteristic values of the individual dimensions; Is a standard feature vector of a certain type of bacteria in a bacteria spectrum feature library, Is the standard vector Characteristic values of the individual dimensions; Is the unified degree of the feature vector.
  4. 4. A raman spectrum pathogen rapid identification detection system according to claim 3, wherein said risk level determination module comprises: The virulence factor association unit is used for calling a pathogen virulence factor database based on a pathogen type preliminary matching result, extracting key virulence parameters of pathogenic genes, toxin secretion capacity and invasiveness of corresponding pathogens, and establishing association mapping of virulence characteristics and pathogen types; The risk assessment model construction unit combines clinical pathogenic probability statistical data, epidemiological monitoring results and clinical treatment effect data, adopts a analytic hierarchy process to endow scientific weight to assessment indexes such as virulence parameters, transmission capacity, clinical cure rate, susceptible crowd range and the like, and establishes a multi-dimensional and multi-index risk assessment model; The early warning trend generating unit carries out comprehensive weighted calculation on each index through the risk assessment model to obtain a germ comprehensive risk value, three risk levels of high, medium and low are divided according to preset standards, targeted early warning trends are generated aiming at different risk levels, high risk triggers emergency prevention and control early warning and priority diagnosis and treatment suggestions, medium risk prompts strengthen monitoring and accurate intervention, low risk gives conventional prevention and control guidance and follow-up suggestions, and the comprehensive risk value calculation formula is as follows: , in the formula, Is a comprehensive risk value of bacteria; Is the first The weight of the individual evaluation index is determined, Is the first The normalized score of each of the evaluation indicators, The total number of the evaluation indexes contained in the risk evaluation model.
  5. 5. The rapid identification and detection system of raman spectrum pathogens according to claim 4, wherein the result verification module comprises: The integrated learning fusion unit uses three basic models of random forest, support vector machine and logistic regression to respectively carry out secondary operation on the prediction results of the double-branch recognition module, and synthesizes the prediction results of the models through a voting mechanism and a probability fusion algorithm to calculate and obtain comprehensive confidence coefficient; The confidence threshold judging unit is used for setting a hierarchical confidence screening standard, directly confirming the results with the comprehensive confidence of more than or equal to 95%, performing feature matching secondary verification on 85% -94% of the results, marking the samples with the confidence of <85% as to-be-checked, and recording key reasons which do not reach the standard; And the detection result calibration unit establishes an error correction model based on the historical detection data, performs systematic deviation calibration and precision optimization on the confirmed recognition result, and adjusts a correction coefficient by using clinical feedback data.
  6. 6. The rapid identification and detection system of raman spectrum pathogens according to claim 5, wherein the spectrum acquisition module further comprises: The sample adaptation unit is used for configuring a replaceable special sampling probe and a sample pretreatment component aiming at solid, liquid and aerosol pathogen samples in different forms, and realizing rapid adaptation and efficient collection of multi-scene samples through a modularized design; And the real-time monitoring unit is integrated with the spectrum acquisition progress visualization interface, feeds back characteristic peak identification success rate, data signal to noise ratio and acquisition completion index in real time, establishes an abnormal monitoring mechanism, automatically triggers a re-acquisition instruction when the index is lower than a preset standard, and records abnormal conditions.
  7. 7. The rapid identification and detection system of raman spectrum pathogens according to claim 6, wherein the dual branch identification module further comprises: The characteristic enhancement unit is used for carrying out data expansion on the spectrum characteristics of the rare germs by adopting a generated countermeasure network, and simulating spectrum changes under different detection environments and different sample states through a style migration technology; And the quick response unit is used for carrying out quantization compression and hardware acceleration adaptation on the double-branch model, optimizing the calculation flow of the feature extraction and matching algorithm, reducing redundant operation, improving the data processing efficiency through a parallel calculation technology, and shortening the single-sample detection period to a range required by clinical quick diagnosis.
  8. 8. The rapid identification and detection system of raman spectrum pathogens according to claim 7, wherein the risk level determination module further comprises: The data updating unit establishes a data synchronization mechanism, synchronizes the latest clinical pathogenic data, the latest germ variation monitoring result, the latest epidemiology dynamics and the updating information of the treatment scheme in real time, dynamically adjusts the weight parameters and the grade judgment standard of the risk assessment model in an online learning mode, and ensures the timeliness and the accuracy of risk assessment; the personalized early warning unit is used for customizing the detail degree and presentation form of early warning contents according to the requirements of different application scenes, the clinical scenes are used for supplementing medication suggestions and diagnosis and treatment path guidance, the disease control scenes are used for strengthening propagation path analysis and prevention and control range suggestions, and the food safety supervision scenes are used for adding pollution tracing prompt and management and control measures.
  9. 9. The rapid identification and detection system of raman spectrum pathogens according to claim 8, wherein the result verification module further comprises: The abnormal marking unit is used for automatically positioning abnormal points of spectrum data for a sample marked to be checked, wherein the abnormal points of the spectrum data comprise characteristic peak missing, noise exceeding standard, baseline drifting, characteristic matching difference items and model prediction bifurcation points, and generating a check guide report to acquire check key points and detection directions; And the tracing unit is used for establishing a detection full-flow data log, recording sample information, instrument parameters, model versions, operators and detection time information and supporting the whole-course tracing and historical data tracing inquiry of detection results.
  10. 10. The rapid identification and detection system of raman spectrum pathogens according to claim 9, wherein the spectrum acquisition module comprises: The accurate acquisition unit is used for matching the optimal excitation light parameter combination based on the type and the sample state of a pathogen sample, and acquiring Raman original spectrum data with high signal-to-noise ratio and low interference through acquisition iterative optimization and signal enhancement processing; The multidimensional information synchronization unit is used for synchronously collecting culture environment information, morphological characteristic images under a microscope and clinical related information of the pathogen sample to form multidimensional sample supplementary data; And the data calibration unit is used for establishing a calibration model based on the standard pathogen spectrum library, calling the standard spectrum data as a reference standard in real time, and carrying out wavelength offset correction, intensity normalization processing and baseline correction on the Raman original spectrum data.

Description

Raman spectrum germ quick identification detection system Technical Field The invention relates to the technical field of germ detection, in particular to a Raman spectrum germ rapid identification detection system. Background The rapid and accurate identification and risk assessment of germs are the key links of clinical diagnosis and treatment, disease prevention and control and public health safety guarantee, in the traditional germ detection method, although the culture method is a gold standard, the detection period is long, the rapid response requirement of sudden infectious diseases is difficult to meet, the nucleic acid amplification technology is poor in adaptability to novel variant germs and is easy to be interfered by sample impurities to cause false positive depending on the design of specific primers, and the single Raman spectrum detection technology only focuses on spectral feature extraction, so that the identification precision of germ types is insufficient and the false detection rate is high. Meanwhile, the existing detection system generally lacks a perfect risk assessment system, the risk level is simply judged according to the types of germs, comprehensive assessment is carried out on multidimensional indexes of uncombined virulence factors, transmissibility and clinical cure rate, and early warning pertinence is not strong. How to solve the technical problem is a technical problem that a person skilled in the art needs to overcome. Disclosure of Invention In order to solve the technical problems, the quick identification and detection system for the Raman spectrum pathogens is provided. In order to achieve the above purpose, the invention adopts the following technical scheme: A Raman spectrum germ rapid identification detection system comprises a spectrum acquisition module, a double-branch identification module, a risk level judgment module and a result verification module; The spectrum acquisition module is used for acquiring Raman original spectrum data of a pathogen sample and sample-related multidimensional information and generating a pathogen Raman fusion data set; The dual-branch identification module is electrically connected with the spectrum acquisition module and is used for extracting cross-domain characteristics and performing pathogen type matching based on the fusion data set to generate a pathogen type preliminary matching result; The risk level judging module is electrically connected with the double-branch identifying module and is used for constructing a risk assessment model based on a germ type preliminary matching result and combining a germ virulence factor database and clinical pathogenic probability statistical data to generate germ risk levels and corresponding early warning trends; The result verification module is electrically connected with the risk level judgment module and is used for fusing the multi-model prediction confidence coefficient by utilizing the integrated learning strategy based on the pathogen risk level and the early warning tendency, generating a pathogen identification detection result through threshold screening and generating a detection report. Preferably, the dual branch identification module includes: The cross-domain feature extraction unit adopts an attention mechanism enhanced convolutional neural network, firstly performs fingerprint region and characteristic peak region segmentation on spectrum data in a germ Raman fusion data set, extracts characteristic peak positions, intensity distribution and peak shape features, simultaneously performs texture, contour and detail feature extraction on morphological feature images, maps different modal features to the same feature space through a modal attention weight distribution mechanism, and converts the different modal features into feature vectors with uniform dimensions; The dual-branch modeling unit is used for constructing a lightweight transducer branch and a residual network branch, wherein the transducer branch carries out global semantic association modeling on feature vectors through a multi-head attention mechanism, captures deep association among cross-modal features, the residual network branch excavates local detail features through a residual connection structure to avoid gradient disappearance, and the output features of the two branches are cascaded according to dynamic weights to generate cross-domain fusion features with the global association and the local detail. Preferably, the dual branch identification module further comprises: the category matching unit performs cosine similarity calculation on the cross-domain fusion features and standard features in the pathogen spectral feature library, screens and verifies similarity sorting results through a K neighbor cluster matching algorithm, and generates a pathogen category preliminary matching result containing Top-N candidate categories, corresponding matching similarity and key feature matching items, wherein