Rapid meningitis diagnosis system based on combination of SERS (Surface Enhanced Raman Scattering) technology and machine learning method

Abstract

The invention belongs to the technical field of machine learning auxiliary diagnosis, and particularly relates to a meningitis rapid diagnosis system based on the combination of an SERS technology and a machine learning method. The system comprises: an input module, which is used for inputting SERS spectral data of cerebrospinal fluid; the preprocessing module is used for preprocessing the SERS spectrum data; and the prediction module is used for calculating the preprocessed SERS spectral data through a machine learning model to obtain a meningitis classification prediction result. According to the method and the system, the SERS technology and the machine learning method are combined for the first time, the method and the system for rapidly and automatically diagnosing the meningitis are realized, and the method and the system have the advantages of simplicity, rapidness, accuracy, capability of carrying out subtype classification and the like, and have a good application prospect in clinical diagnosis of the meningitis, particularly in an early screening process.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of machine learning assisted diagnosis, and particularly relates to a meningitis rapid diagnosis system based on combination of SERS technology and machine learning method. BACKGROUND

[0002] As a kind of "fingerprint" spectrum, Raman spectrum reflects the structure and vibration of molecules. Compared with infrared spectrum, Raman spectrum is more suitable for detection of biomolecules in biological systems because it requires less sample and the water itself has a weak Raman signal. However, the spontaneous Raman effect is relatively weak, generally only about 10 -10 of the incident light intensity. Therefore, the Raman signal is generally very weak, and the sample needs to be enhanced for Raman spectrum research. SERS (surface-enhanced Raman spectroscopy) spectrum technology is a kind of molecular vibration fingerprint spectrum technology in which the molecules or substances are adsorbed to the surface of certain specific noble metals or semiconductors (SERS enhancement substrate), and the Raman scattering signal is greatly enhanced. The enhancement effect mainly depends on the interaction between light-plasmonic nanomaterials-targets, and has the advantages of high sensitivity, good specificity, narrow spectral band, simple operation, less sample requirement, and no interference from spontaneous fluorescence of biological samples and water. In recent years, SERS technology has proved its ability as a label-free detection tool, which can sensitively detect various physiological indicators in different types of clinical biological samples (including blood, urine, tears, sweat, cerebrospinal fluid and tissue sections), and has been applied to the identification and diagnosis of various diseases, such as various cancers, diabetes and leukemia.

[0003] The SERS data contains rich molecular vibration "fingerprint" information of the measured substance, and these spectral information or imaging data sets can be used for chemical system analysis, but the amount of data is too large to make some classic linear information processing methods difficult to effectively extract data. Therefore, the classic linear information processing method cannot meet the requirements of SERS data processing and analysis. In the field of analytical science, machine learning provides unprecedented opportunities for extracting information from complex or large data sets in the fields of chromatography, mass spectrometry, nuclear magnetic resonance and spectroscopy. At present, there have been relevant researches on machine learning for classification of SERS spectral data, for example, "CN114216851A An acute pancreatitis evaluation device based on surface enhanced Raman spectroscopy technology" discloses a method for analyzing SERS spectral data by using machine learning to diagnose acute pancreatitis. However, there are complex relationships and spectral characteristics in the SERS spectrum without samples, and different preprocessing methods of SERS spectral data may have a significant impact on the prediction results of the machine learning model. On the other hand, a large number of machine learning algorithms have been developed in the prior art, and these algorithms are suitable for different data characteristics and have different adaptability for different prediction tasks. Therefore, for specific SERS spectral data of a sample and a specific prediction task, the selection of different machine learning algorithms will also have a significant impact on the prediction performance. For specific SERS spectral data of a sample and a specific prediction task, not all machine learning algorithms can achieve accurate prediction, and the machine learning algorithm needs to be selected optimally.

[0004] Currently, cerebrospinal fluid (CSF) pathogen detection is the standard method for diagnosing various meningitis, and is combined with patient clinical manifestations to assist in diagnosis. Cerebrospinal fluid is generally obtained by lumbar puncture, and is a colorless transparent liquid existing in the ventricle and subarachnoid space, with a total volume of 130-150 mL and an average daily production of 400-500 mL. It surrounds and supports the entire brain and spinal cord. Normal cerebrospinal fluid has certain chemical composition and pressure, and plays an important role in maintaining the relative stability of intracranial pressure. Therefore, changes in the color or pressure of cerebrospinal fluid, the content of biochemical indicators (such as protein, glucose, and chloride), cell microscopy, and polymerase chain reaction (PCR) abnormalities can indicate related diseases. In the prior art, it has been found that the SERS technology can be used to detect the chloride content in the cerebrospinal fluid, so as to realize the diagnosis of meningitis by using the results of the chloride content (CN117491331A, a portable Raman spectrum-based quantitative detection method for cerebrospinal fluid chloride). However, the analysis of the SERS spectrum data of the cerebrospinal fluid still relies on traditional manual interpretation, which depends on experience and subjective judgment, and may lead to misdiagnosis or missed diagnosis. At the same time, the use of SERS spectrum data is to quantitatively calculate the content of chloride, and the rich fingerprint information contained in the SERS spectrum cannot be effectively utilized, which may also lead to errors in the diagnosis results.

[0005] Therefore, there is an urgent need in the art to develop a system for automatically diagnosing meningitis by classifying the SERS spectrum data of cerebrospinal fluid through a machine learning algorithm. SUMMARY

[0006] In view of the problems of the prior art, the present application provides a rapid meningitis diagnosis system based on the combination of SERS technology and machine learning methods.

[0007] A rapid meningitis diagnosis system based on the combination of SERS technology and machine learning methods, comprising:

[0008] An input module for inputting SERS spectrum data of cerebrospinal fluid;

[0009] A preprocessing module for preprocessing the SERS spectrum data;

[0010] A prediction module for calculating the preprocessed SERS spectrum data through a machine learning model to obtain a meningitis classification prediction result.

[0011] Preferably, the preprocessing includes at least one of removing cosmic rays or denoising.

[0012] Preferably, the meningitis classification prediction result is to classify the sample into four categories of tubercular meningitis, viral meningitis, cryptococcal meningitis and autoimmune meningitis.

[0013] Preferably, the algorithm of the machine learning model is selected from KNN or SVM.

[0014] Preferably, it further comprises a SERS spectrum data detection module for detecting SERS spectrum data of the cerebrospinal fluid sample.

[0015] Preferably, in the SERS spectrum data detection module, the conditions for detecting the SERS spectrum data include that the laser wavelength is 785 nm, the excitation laser power is set to 20-310 mW, and the spectral range is 300-2500 cm -1 .

[0016] Preferably, in the SERS spectrum data detection module, the SERS enhancement reagent for detecting the SERS spectrum data is Ag nanoparticles with I - ion modification.

[0017] Preferably, in the SERS spectrum data detection module, the Ag nanoparticles with I - ion modification are prepared by co-incubating Ag nanoparticles with a solution containing I - for 20-60 minutes at a temperature range of 25-90℃.

[0018] Preferably, in the SERS spectrum data detection module, the method for detecting the SERS spectrum data is to mix the SERS enhancement reagent and the cerebrospinal fluid sample, and then detect by using a Raman spectrum device.

[0019] The application also provides a computer readable storage medium having stored thereon a computer program for implementing the above-mentioned diagnostic system.

[0020] The application combines SERS technology and machine learning method for the first time, and realizes a method and system for rapid and automatic diagnosis of meningitis.

[0021] 1. The whole detection and diagnosis process is simple to operate, rapid in detection, high in sensitivity and excellent in stability.

[0022] 2. The amount of cerebrospinal fluid sample required is small, and the detection can be performed on a sample amount of 30 μL.

[0023] 3、In the preferred scheme, the fingerprint information in the SERS spectrum data is identified by a machine learning model, which can realize the rapid detection and identification of four subtypes of meningitis, namely Tubercular Meningitis (TM), Viral Meningitis (VM), Cryptococcal Meningitis (CM) or Autoimmune Encephalitis (AE). In the prior art, neither the quantitative analysis of the chloride content obtained by SERS spectrum nor the manual interpretation of the SERS spectrum can realize the identification of the subtypes of meningitis. Therefore, in the present application, the combination of SERS technology and machine learning method produces a powerful diagnostic ability beyond expectation.

[0024] 4、The present application optimizes the type and synthesis method of SERS enhancing reagent, the collection conditions of SERS spectrum, the pretreatment method of input data and the algorithm of machine learning model, etc. factors to obtain the best prediction model with prediction performance.

[0025] In summary, the present application has the advantages of simplicity, rapidity, accuracy and subtype classification, etc. In the clinical diagnosis of meningitis, especially in the early screening process, it has good application prospect.

[0026] Obviously, according to the above content of the present application, according to the ordinary technical knowledge and conventional means in the art, other various forms of modification, replacement or change can be made without departing from the above basic technical idea of the present application.

[0027] The above content of the present application will be further described in detail through the specific embodiments. However, it should not be understood that the above subject matter of the present application is limited to the following examples. Any technology realized based on the above content of the present application belongs to the scope of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 UV-Vis absorption spectrum and TEM image of SERS enhancing reagent synthesized in Example 1.

[0029] Figure 2 Confusion matrix and ROC curve for four types of meningitis classification: (a) confusion matrix; (b) ROC curve for KNN training cerebrospinal fluid SERS spectrum of meningitis; (c) confusion matrix; (d) ROC curve for SVM training baseline corrected SERS spectrum of cerebrospinal fluid of meningitis.

[0030] Figure 3 Effect of laser power on SERS spectrum characteristic peak and baseline of cerebrospinal fluid.

[0031] Figure 4 Typical SERS spectra of various meningitis.

[0032] Figure 5 Schematic diagram of SERS spectrum baseline processing process of cerebrospinal fluid.

[0033] Figure 6 Stability of 5 batches of cerebrospinal fluid samples detected by solution method. DETAILED DESCRIPTION

[0034] In the following examples and experimental examples, the reagents and raw materials not specifically explained are commercially available. It should be particularly noted that the algorithms for data acquisition, transmission, storage and processing steps not specifically explained in the examples, as well as the hardware structure, circuit connection, etc. not specifically explained, can be realized through the existing technology disclosed.

[0035] Example 1 Rapid diagnosis system for meningitis based on combination of SERS technology and machine learning method

[0036] The system of the present embodiment comprises:

[0037] A SERS spectrum data detection module for detecting SERS spectrum data of a cerebrospinal fluid sample;

[0038] An input module for inputting SERS spectrum data of cerebrospinal fluid;

[0039] A preprocessing module for preprocessing the SERS spectrum data;

[0040] A prediction module for calculating the preprocessed SERS spectrum data through a machine learning model to obtain a meningitis classification prediction result.

[0041] In the above system, the method for SERS spectrum data detection by the SERS spectrum data detection module specifically comprises the following steps:

[0042] 1. Synthesis of SERS enhancement reagent: 100 mL of AgNO3 solution (1 mM) was heated to boiling, and then 3 mL of 1% (w / w) sodium citrate solution was added under vigorous stirring. The reaction lasted for 1 hour, and the solution color changed to yellow-green, obtaining Ag colloid enhancement reagent. The synthesized Ag colloid UV-visible absorption spectrum and morphology characteristics are shown in Figure 1 After the reaction, the Ag colloid was stored in an environment of 4℃.

[0043] 2. Collection of SERS spectrum of clinical cerebrospinal fluid sample:

[0044] Step 1: Centrifuge 1 mL of Ag colloid (8000 rpm, 10 minutes) to remove supernatant, and dilute to 100 μL with ultrapure water;

[0045] Step 2, the concentrated colloidal nanoparticles of step 1 were mixed with KI aqueous solution (1 mM) at a volume ratio of 1:1, and co-incubated at room temperature (25℃) for 20 min to modify iodide onto the surface of silver nanoparticles, referred to as Ag@KI solution;

[0046] Step 3, the Ag@KI solution was mixed with clinical cerebrospinal fluid samples in a microcuvette at a volume ratio of 1:1, with a total volume of 60 μL;

[0047] Step 4, the mixed solution was directly detected using a portable Raman spectrometer, with a laser wavelength of 785 nm, an excitation laser power setting of 100 mW, and a spectral range of 300-2500 cm -1 The Raman spectrum was measured within 5 min, and the integration time was set to 1 s.

[0048] In the pretreatment module, the method for pretreating the SERS spectral data includes removing cosmic rays and denoising.

[0049] In the prediction module, the algorithm of the machine learning model is selected from KNN or SVM, and the meningitis classification prediction result is to classify the sample (known to be a meningitis sample) into four categories of tuberculous meningitis, viral meningitis, cryptococcal meningitis, or autoimmune meningitis.

[0050] The technical solutions of the present application are further described below through experiments.

[0051] Experimental Example 1: Preferred pretreatment method and machine learning model algorithm

[0052] I. Experimental method

[0053] This experimental example compares the prediction performance of different pretreatment methods and machine learning model algorithms, and each experimental group classifies the samples according to the system of Example 1.

[0054] The difference lies in the method of pretreating the SERS spectral data and the algorithm of the machine learning model in each experimental group, and the pretreatment method uses two methods respectively:

[0055] 1) The pretreatment method includes removing cosmic rays;

[0056] 2) The pretreatment method includes removing cosmic rays and baseline subtraction.

[0057] The algorithms of the machine learning models used are as follows: Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Artificial Neural Network (ANN) algorithm, Random Forest (RF) and Linear Discrimination (LD), respectively. The training set and the test set are randomly divided in the ratio of 80% / 20% for comparison.

[0058] In this experimental example, 40 cerebrospinal fluid samples of clinical meningitis patients (more than 400 spectral data were collected) were used, of which 10 samples of TM, VM, CM and AE inflammation were used. Before the experiment, the samples were stored at-80°C.

[0059] II. Experimental results

[0060] The classification accuracy of different pretreatment methods and different machine learning models is shown in Table 1.

[0061] Table 1 Classification accuracy of meningitis disease based on different machine models and different data features

[0062]

[0063] From the data in Table 1, it can be seen that after deducting the baseline of the SERS spectral data, all machine algorithm models cannot accurately classify the four types of meningitis. Using SERS spectral data without deducting the baseline, some machine learning models can achieve ideal accuracy. Among them, the best machine learning model is the KNN model.

[0064] The confusion matrix and ROC curve of the four types of meningitis classification are shown in Figure 2 , wherein (a) and (b) represent the confusion matrix and ROC curve of KNN training meningitis cerebrospinal fluid original SERS spectrum, respectively, and (c) and (d) represent the confusion matrix and ROC curve of SVM training meningitis cerebrospinal fluid baseline corrected SERS spectrum.

[0065] It can be seen that for the prediction task of the present application, KNN is the best model algorithm.

[0066] Experimental Example 2 Methodological verification in SERS spectral data detection

[0067] In this experimental example, the steps of collecting SERS spectral data are optimized, and the steps and conditions not specifically described are performed according to the description in Example 1.

[0068] I. Excitation laser power optimization

[0069] To obtain stable and reproducible SERS spectra, the SERS spectral data of the cerebrospinal fluid were collected by adjusting different laser powers. As shown in Figure 3 , the intensity of the Raman characteristic peaks and the height of the baseline of the SERS spectrum of the Ag@KI and cerebrospinal fluid mixture can be adjusted by changing the excitation laser power, and high signal-to-noise ratio SERS spectral data can also be obtained as the laser power increases. Therefore, to ensure consistency of the detection conditions and prevent the SERS spectrum from exceeding the range or causing damage to the sample due to excessive power, the preferred condition is to collect the SERS spectral data of the cerebrospinal fluid under the condition that the excitation laser power is set to 100 mW.

[0070] II. Reproducibility verification of SERS spectral data

[0071] The typical SERS spectrum of the cerebrospinal fluid sample is shown in Figure 4 . The measured SERS data was preprocessed, and the asymmetric least squares smoothing method was used to extract the spectral baseline features, such as the baseline intensity or the integral data Figure 5 . According to the characteristic peak intensity, baseline intensity and baseline integral intensity of the cerebrospinal fluid spectrum, the same cerebrospinal fluid sample was repeatedly sampled and tested five times, and the batch reproducibility was good, as shown in Figure 6 . This indicates that the prediction performance of the system provided by the present application has good stability.

[0072] As can be seen from the above examples and experimental examples, the present application combines SERS technology with machine learning methods to realize a method and system for rapid and automatic diagnosis of meningitis, which has the advantages of simplicity, speed, accuracy and ability to classify subtypes, etc. In the clinical diagnosis of meningitis, especially in the early screening process, it has a good application prospect.

Claims

1. A rapid diagnostic system for meningitis based on a combination of SERS technology and machine learning methods, characterized in that, include: The input module is used to input SERS spectral data of cerebrospinal fluid; The preprocessing module is used to preprocess the SERS spectral data; The prediction module is used to calculate the classification prediction results of meningitis by using a machine learning model on the preprocessed SERS spectral data.

2. The diagnostic system according to claim 1, characterized in that: The preprocessing includes at least one of cosmic ray removal or noise reduction.

3. The diagnostic system according to claim 1, characterized in that: The meningitis classification prediction results classify the samples into four categories: tuberculous meningitis, viral meningitis, cryptococcal meningitis, or autoimmune meningitis.

4. The diagnostic system according to claim 1, characterized in that: The algorithm for the machine learning model is selected from KNN or SVM.

5. The diagnostic system according to claim 1, characterized in that, Also includes: The SERS spectral data detection module is used to detect the SERS spectral data of cerebrospinal fluid samples.

6. The diagnostic system according to claim 5, characterized in that: The conditions for detecting the SERS spectral data in the SERS spectral data detection module include: a laser wavelength of 785 nm, an excitation laser power set to 20–310 mW, and a spectral range of 300–2500 cm⁻¹. -1 .

7. The diagnostic system according to claim 5, characterized in that: In the SERS spectral data detection module, the SERS enhancement reagent used to detect the SERS spectral data is a reagent with I... - Ion-modified Ag nanoparticles.

8. The diagnostic system according to claim 7, characterized in that: In the SERS spectral data detection module, the one with I - The preparation method of ion-modified Ag nanoparticles involves reacting Ag nanoparticles with ion-containing... - The solution was incubated for 20–60 minutes at a temperature range of 25–90°C.

9. The diagnostic system according to any one of claims 6-8, characterized in that: In the SERS spectral data detection module, the method for detecting the SERS spectral data is to mix the SERS enhancement reagent and the cerebrospinal fluid sample and then use a Raman spectroscopy device for detection.

10. A computer-readable storage medium, characterized in that, It stores a computer program for implementing the diagnostic system according to any one of claims 1-9.

Images