A multi-channel raman spectroscopy system for tumor marker detection

Abstract

The application belongs to the technical field of marker detection, and discloses a multi-channel Raman spectrum analysis system for tumor marker detection, which comprises a pretreatment module, an acquisition module, an enhancement processing module, an analysis module and an output module; the pretreatment module is used for uniformly coating electromagnetic nanomaterial on the surface of a target tissue section; the acquisition module is used for acquiring the Raman spectrum of the target tissue section, and the Raman spectrum comprises a plurality of Raman scattering lights with different wavelengths; the enhancement processing module is used for converting the Raman spectrum acquired by the acquisition module into an electric signal, and also used for enhancing the electric signal; the analysis module is used for identifying tumor markers by analyzing the enhanced electric signal; and the output module is used for marking tumor markers according to the identification result of the analysis module, and also used for outputting a target tissue pathological image with the tumor markers marked.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of marker detection, and particularly relates to a multi-channel Raman spectrum analysis system for tumor marker detection. BACKGROUND

[0002] Tumor markers are substances synthesized and released by tumor cells or generated by the body in response to tumor cells during the occurrence and growth of tumors, covering proteins, enzymes, gene products and metabolites, etc. The detection of tumor markers is of key significance for the early diagnosis, treatment and prognosis evaluation of malignant tumors.

[0003] At present, malignant tumors have become the primary cause of death among Chinese residents, so it is of extremely important clinical value to develop a method capable of realizing sensitive, accurate and specific detection of tumor markers, while meeting the needs of simplicity, rapidity and non-invasive diagnosis. However, existing tumor marker detection methods such as enzyme-linked immunosorbent assay (ELISA), polymerase chain reaction (PCR), fluorescence method, electrochemical method and mass spectrometry, although widely used in clinical practice, have a series of shortcomings, such as insufficient sensitivity, complex operation, long detection period, etc., and most of these methods are invasive detection, which is not conducive to the early screening of low-concentration tumor markers.

[0004] Raman spectrum technology, including surface-enhanced Raman scattering (SERS), tip-enhanced Raman spectroscopy (TERS) and stimulated Raman scattering (SRS), is a powerful label-free tool that can provide excellent dynamic performance and high spatial resolution imaging in the fields of biology and medicine. Although Raman spectrum technology has these advantages, due to the weakness of spontaneous Raman signal, slow data acquisition speed and high equipment cost, these factors limit its wide application in clinical practice.

[0005] Protein molecules are an important class of tumor markers, however, surface-enhanced Raman spectroscopy (SERS) performs poorly in directly detecting high molecular weight biomolecules, with complex biological macromolecule spectral bands overlapping and limited selectivity for proteins.

[0006] In addition, the current Raman spectrum reconstruction algorithm is mainly based on the reconstruction of spectral reflectance curves, such as pseudo-inverse method, Wiener estimation method and finite-dimensional model method, etc. These methods are established on the baseline of linear systems, and have weak resistance to nonlinear interference.

[0007] Therefore, it is particularly important to develop a calibration-reconstruction algorithm specially applicable to Raman spectrum, which can further improve the accuracy and specificity of detection and realize sensitive detection of trace tumor markers. SUMMARY

[0008] In view of this, in order to solve the problems raised in the background art, the purpose of the present application is to provide a multi-channel Raman spectrum analysis system for tumor marker detection.

[0009] To achieve the above purpose, the present application provides the following technical solutions.

[0010] A multi-channel Raman spectrum analysis system for tumor marker detection, comprising a pretreatment module, an acquisition module, an enhancement processing module, an analysis module and an output module.

[0011] The pretreatment module is used for modifying the target tissue section by magnetic nanoparticle adaptive modification.

[0012] The acquisition module is used for acquiring the Raman spectrum of the target tissue section, and the Raman spectrum includes a plurality of Raman scattering lights of different wavelengths.

[0013] The enhancement processing module is used for converting the Raman spectrum collected by the acquisition module into an electrical signal, and is also used for enhancing the electrical signal.

[0014] The analysis module is used for identifying tumor markers by analyzing the enhanced electrical signal.

[0015] The output module is used for marking tumor markers according to the identification result of the analysis module, and is also used for outputting the target tissue pathological image marked with tumor markers.

[0016] Preferably, the pretreatment step of the pretreatment module comprises:

[0017] Selecting aptamers based on tumor markers in the target tissue section;

[0018] Modifying magnetic nanoparticles by the aptamers;

[0019] Coating a layer of chitosan on the surface of the modified magnetic nanoparticles as a capture probe;

[0020] Incubating the target tissue section with the capture probe to form a sandwich complex of aptamer-target tissue section-aptamer;

[0021] Magnetic separation and washing the sandwich complex with a buffer solution, and then transferring it to a SERS substrate.

[0022] Preferably, the step of modifying magnetic nanoparticles by the aptamers comprises:

[0023] Rinsing the streptavidin-coated magnetic nanoparticles with a buffer solution, and then resuspending them in the buffer solution to obtain a suspension;

[0024] The aptamer is added into the suspension and reacted under smooth oscillation at room temperature for 30 min.

[0025] The magnetic force separation obtains the magnetic nanoparticle with the aptamer fixed on the surface.

[0026] Preferably, the collection module comprises a spectrum collection device provided with a plurality of independent optical channels, and each independent optical channel is provided with a laser source, a spectrometer, a detector and a TE refrigeration module.

[0027] Preferably, the enhancement processing module enhances the electrical signal through a Raman signal separation algorithm based on a convolutional neural network.

[0028] Preferably, the Raman signal separation algorithm based on the convolutional neural network is expressed as:

[0029] R = L * B + N; wherein R represents an input measured spectrum signal, L represents an independent spectral peak signal of a Raman spectrum, B represents an instrument broadening signal and a background baseline signal, N represents a measurement noise, and "*" represents a convolution operation.

[0030] Preferably, the independent spectral peak signal L of the Raman spectrum is expressed as:

[0031]

[0032] wherein x represents a Raman shift, c represents a center of a Lorentz function, wL represents a half-height half-width of the Lorentz function, and S represents a full spectrum area of the Lorentz function.

[0033] Preferably, the instrument broadening signal and the background baseline signal B are expressed as:

[0034]

[0035] wherein w G represents a half-height half-width of a Gaussian function .

[0036] Preferably, the tumor marker comprises:

[0037] A Raman spectrum database of a tissue slice containing a tumor marker is constructed, a spectrum data matrix is obtained, after data normalization processing, a characteristic vector of a Raman spectrum is calculated, and a formula is used:

[0038] ;

[0039] The Raman spectrum characteristic vector is calculated, and a recognition model is constructed based on a deep learning algorithm.

[0040] wherein represents the a characteristic vector of the Raman spectrum, a Raman signal intensity of the th data point in the th group of Raman spectra, a mean signal intensity of the th group of Raman spectra, a total number of sampling points of the Raman spectrum, a regularization parameter for preventing the denominator from approaching zero;

[0041] training the recognition model using the database;

[0042] inputting an electrical signal into the recognition model, using the formula:

[0043]

[0044] calculating a specific signal feature, extracting the specific signal feature, and identifying a tumor marker according to the specific signal feature;

[0045] wherein, represents a specific signal feature value, represents the th enhanced electrical signal data value, represents the mean value of the enhanced electrical signal data, represents the th background signal data value, represents the mean value of the background signal data, represents the total number of enhanced electrical signal data, represents the total number of background signal data, represents the index of the enhanced electrical signal data, represents the index of the background signal data.

[0046] Preferably, the tumor marker includes a spatial position marker and a quantity concentration marker.

[0047] As a general inventive concept, the present application also provides the following technical solutions:

[0048] A multi-channel Raman spectrum analysis method for tumor marker detection, comprising the following steps:

[0049] S1. Hardware preprocessing

[0050] Calibration of the spectrum acquisition device with multiple independent optical channels, each of which is provided with a laser source, a spectrometer, a detector, and a TE refrigeration module.

[0051] The overall spectrum acquisition system has a size of 144x92x46 mm, and is as​Figure 3 The spectral instrument light path is built by using the Czerny-Turner system, two spherical mirrors are used as collimating lenses and converging lenses respectively, the paths of incident light and imaging light cross each other, and the slit width is 25 microns, so that the light flux is improved as much as possible while the resolution is not greatly affected.

[0052] The spectral acquisition device is calibrated and debugged by using a mercury lamp method, and multiple measurement results prove that the spectral acquisition device has a spectral coverage range of 785nm-1050nm, a resolution of 6.45cm -1 (0.54nm, full width at half maximum at 912nm), improves the resolution under the premise of ensuring that the Raman spectrum can fully reflect the tumor marker concentration information. At the same time, the signal-to-noise ratio of the system is higher than 6000:1, which ensures that the collected spectral signal can be used as a reliable information basis for quantitative analysis of tumor markers, and a silicon wafer is used as a standard substance to correct the wave number accuracy.

[0053] The laser source uses a quantum nanometer pulse laser (QNPL), each laser can generate a specific wavelength of light beam, which is used to excite the Raman scattering of the sample;

[0054] The beam splitter uses a "precision beam splitter (PBS)" and a "high-resolution spectral sorter (HSS)" to ensure accurate focusing of the laser and clear separation of the spectral signal. The PBS is responsible for uniformly distributing the incident laser to different areas of the sample, and the HSS is responsible for screening useful spectral information from complex scattered light; for example, a grating with 830l / mm of ruling number.

[0055] The detector uses a "synchronous spectral sensor (SSS)". After the laser irradiates the sample, the scattered Raman spectrum is captured by the SSS, realizing efficient signal acquisition. In order to further improve the quality and accuracy of the data, the application also introduces a "dynamic spectral enhancement module (DSEM)" in the detector, which is used to amplify and enhance weak Raman signals; for example, a 2048-pixel back-illuminated linear array CCD with a pixel size of 14x900 microns.

[0056] The refrigeration temperature of the TE refrigeration module is -50℃, so as to improve the signal-to-noise ratio and stability of the spectrum of each optical channel.

[0057] S2. Sample pretreatment

[0058] Selecting an aptamer based on the tumor marker in the target tissue section;

[0059] The streptavidin-coated magnetic nanoparticles are washed with a buffer solution and then resuspended in the buffer solution to obtain a suspension; the aptamer is added to the suspension and reacted at room temperature under smooth oscillation conditions for 30 minutes; the magnetic nanoparticles modified with the aptamer on the surface are obtained by magnetic separation.

[0060] A layer of chitosan is coated on the surface of the modified magnetic nanoparticles as a capture probe;

[0061] The target tissue slice is incubated with the capture probe to form an aptamer-target tissue slice-aptamer sandwich complex;

[0062] The sandwich complex is separated by magnetic force, washed with a buffer, and then transferred to a SERS substrate.

[0063] The magnetic nanoparticles create an enhanced spectral reaction interface on the surface of the target tissue slice. This interface significantly improves the efficiency of Raman scattering, enabling effective detection of even extremely low concentrations of tumor markers. Using multi-frequency resonant Raman scattering (MFRS) technology, a multi-wavelength laser system is configured to simultaneously excite multiple tumor markers in the sample. Each wavelength of laser light is precisely adjusted to match the resonant frequency of a specific tumor marker, maximizing the generation of Raman signals.

[0064] S3. Software preprocessing

[0065] A Raman signal separation algorithm based on a convolutional neural network, a Raman spectrum database of tissue slices containing tumor markers, and a recognition model based on a deep learning algorithm are constructed, and the recognition model is trained using the database.

[0066] S4. Raman spectrum acquisition of the target tissue slice using a spectral acquisition device. Multiple different wavelengths of Raman scattered light are collected through multiple channels to improve the speed and efficiency of data acquisition.

[0067] S5. The collected Raman spectrum is converted into an electrical signal, and in this conversion process, the signal is also preliminarily processed by the spectral deep weaving technology (SDW), such as signal amplification, noise filtering, and signal stabilization, etc.

[0068] S6. Enhancement processing of the electrical signal;

[0069] S7. The electrical signal after enhancement processing is input into the recognition model, specific signal features are extracted, and tumor markers are identified according to the specific signal features;

[0070] S8. Construction of an enhanced Raman image of the target tissue slice based on the collected Raman spectrum and image reconstruction technology;

[0071] S9. Using the recognition results of tumor markers and spatial resolution algorithms, the spatial position and concentration of tumor markers are marked in the enhanced Raman image to obtain a target tissue pathological image marked with tumor markers;

[0072] S10. Output the target tissue pathological image, so that the doctor can see the specific distribution of the tumor marker in the biological sample, thereby providing more accurate diagnostic information.

[0073] Compared with the prior art, the present application has the following beneficial effects:

[0074] (1) The high automation of the system reduces the complexity of operation and the possibility of human error. The standardized sample processing, spectral acquisition, data analysis and result output process greatly reduces the operation threshold, so that non-professionals can easily use it, improving the user friendliness of the system.

[0075] (2) The system of the present application improves the detection sensitivity of tumor markers by using multi-channel spectral acquisition and magnetic nano-enhanced methods, so as to ensure that trace biomarkers related to tumors can be effectively detected at low concentrations.

[0076] (3) The system of the present application realizes the excitation and collection of Raman scattering light of multiple different wavelengths through the design of multiple optical channels, thereby greatly improving the speed and efficiency of data acquisition; and the beam splitter and high-resolution spectral sorter are used in each optical channel of the spectrometer, thereby effectively ensuring the precise focusing of the laser and the high-quality output of the spectral signal.

[0077] (4) The system of the present application realizes the optimization and enhancement of Raman spectral signals through the Raman signal separation processing of convolutional neural network, thereby accurately eliminating the noise and interference introduced by sample preparation and environmental factors, and ensuring the high consistency and repeatability of spectral data.

[0078] (5) The system of the present application uses directly collected Raman spectra as training data, without involving complex spectral enhancement processing, and uses the trained recognition model based on deep learning algorithm to extract and effectively recognize the features of tumor markers, without the need for antibody reagents and staining process, greatly shortening the cycle of pathological diagnosis and reducing the cost of diagnosis.

[0079] (6) The system of the present application can present the distribution of tumor markers on the target tissue slice in the form of an image through Raman imaging processing, providing visual pathological images for doctors. This can seamlessly connect and transition for pathologists, help doctors make clinical interpretation and treatment decisions, overcome the shortcomings of existing spectral detection techniques that are not easy for doctors to understand, and improve the efficiency and accuracy of diagnosis. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 Structure diagram of the multi-channel Raman spectral analysis system for tumor marker detection of the present application;

[0081] Figure 2A flow chart of the multi-channel Raman spectrum analysis method for tumor marker detection of the present application;

[0082] Figure 3 A light path structure diagram of the acquisition module of the present application. DETAILED DESCRIPTION

[0083] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the protection scope of the present application.

[0084] Embodiment 1

[0085] A multi-channel Raman spectrum analysis system for tumor marker detection, comprising an acquisition module, an analysis module and an output module.

[0086] Regarding the acquisition module, the Raman spectrum of the target tissue section is acquired.

[0087] The acquisition module comprises a spectrum acquisition device provided with a plurality of independent optical channels, and a laser source, a beam splitter, a detector and a TE refrigeration module are arranged in each independent optical channel.

[0088] The size of the overall spectrum acquisition system is 144x92x46 mm, and the optical path of the spectrometer is built using a Czerny-Turner system as shown in Figure 3 Two spherical mirrors are used as collimating lenses and converging lenses respectively, the paths of the incident light and the imaging light cross each other, and the slit width is 25 μm, so as to improve the luminous flux as much as possible while ensuring that the resolution is not greatly affected.

[0089] The laser source uses a quantum nanometer pulse laser (QNPL), each laser can generate a light beam of a specific wavelength, which is used to excite the Raman scattering of the sample, and based on this, multiple independent optical channels can excite multiple Raman scattering lights of different wavelengths.

[0090] The beam splitter uses a “precision beam splitter (PBS)” and a “high-resolution spectrum sorter (HSS)” to ensure accurate focusing of the laser and clear separation of the spectral signal. The PBS is responsible for uniformly distributing the incident laser to different areas of the sample, and the HSS is responsible for filtering useful spectral information from complex scattered light; for example, a grating with a ruling number of 830 l / mm.

[0091] The detector uses a "synchronous spectral sensor (SSS)". When the laser irradiates the sample, the scattered Raman spectrum is captured by the SSS, achieving efficient signal acquisition. To further improve the quality and accuracy of the data, the present application also introduces a "dynamic spectral enhancement module (DSEM)" in the detector, which is used to amplify and enhance weak Raman signals; for example, a 2048-pixel back-illuminated linear array CCD with a pixel size of 14x900μm.

[0092] The TE refrigeration module has a refrigeration temperature of -50℃, which improves the signal-to-noise ratio and stability of the spectrum of each optical channel.

[0093] Regarding the analysis module - identify the tumor markers of the target tissue section through Raman spectrum analysis.

[0094] The analysis module uses high-dimensional spectral analysis technology to analyze and process the collected Raman spectrum, which combines complex multi-dimensional quantum analysis (such as quantum support vector mechanism) and deep learning algorithm. Specifically, through deep learning of the Raman spectrum database, the specific spectral characteristics of the tumor markers are accurately extracted, so that trace information can be extracted from complex biological samples, and tumor markers can be identified with ultra-high sensitivity even at very low concentrations.

[0095] Regarding the output module - for marking tumor markers according to the identification results of the analysis module, and for outputting the target tissue pathology image marked with tumor markers.

[0096] Marking: Introduce biomarker spectral mapping technology, which combines image reconstruction technology and spatial analysis algorithm. Specifically, by constructing an enhanced Raman image of the target tissue section and marking the spatial position and number concentration of the tumor markers in the enhanced Raman image, the target tissue pathology image is obtained.

[0097] Output: Output the target tissue pathology image (the spatial position and number concentration of the tumor markers are marked in the image). In addition, the possible clinical significance of these tumor markers can also be analyzed in the image, providing necessary decision support for doctors and providing reference for formulating early diagnosis and personalized treatment plan.

[0098] Example 2

[0099] This embodiment is a multi-channel Raman spectrum analysis method for tumor marker detection based on the above-mentioned embodiment 1, specifically comprising the following steps:

[0100] S1. Hardware pretreatment

[0101] Calibrate the spectral acquisition device with multiple independent optical channels, each of which is equipped with a laser source, a spectrometer, a detector, and a TE refrigeration module.

[0102] The spectral acquisition device was calibrated and adjusted using the mercury lamp method. Multiple measurements confirmed that the spectral acquisition device of this invention has a spectral coverage range of 785nm-1050nm and a resolution of 6.45cm. -1 (Full width at half maximum at 0.54 nm and 912 nm), improving resolution while ensuring that Raman spectroscopy can fully reflect tumor marker concentration information. Simultaneously, the system's signal-to-noise ratio is higher than 6000:1, ensuring that the acquired spectral signals can serve as a reliable information basis for quantitative analysis of tumor markers, and wavenumber accuracy is corrected using a silicon wafer as a standard.

[0103] S2. Software Preprocessing

[0104] A Raman spectroscopy database of tissue slices containing tumor markers and a recognition model based on a deep learning algorithm were constructed, and the recognition model was trained using the database.

[0105] A Raman spectral database containing tumor markers was constructed, the spectral data matrix was obtained, and after data normalization, the eigenvectors of the Raman spectra were calculated using the following formula:

[0106] ;

[0107] Calculate the eigenvectors of the Raman spectrum A recognition model is built based on deep learning algorithms;

[0108] in, Representing the Eigenvectors of Raman spectra Representing the The first group of Raman spectra Raman signal intensity at each data point Representing the The average signal intensity of the group of Raman spectra, The total number of sampling points representing the Raman spectrum. This represents the regularization parameter to prevent the denominator from approaching zero;

[0109] To construct a Raman spectral database of tissue sections containing tumor markers, Raman spectral data of target tissue sections were first collected using a high-sensitivity Raman spectrometer under standardized experimental conditions, such as a laser wavelength of 785 nm, a laser power of 100 mW, and an integration time of 10 seconds. The raw spectral data may contain baseline drift and noise; therefore, data preprocessing was necessary. Baseline correction was performed using polynomial fitting, with a third-order polynomial used for baseline fitting. The fitted baseline was then subtracted from the raw spectrum. Finally, a moving average filtering method was applied to remove high-frequency confounding markers. To eliminate high-frequency noise, data points with a window size of 5 were selected for smoothing. After preprocessing, the spectral data were normalized using maximum value normalization, dividing the intensity value of each spectrum by its maximum value to ensure that the intensity range of all spectra is between 0 and 1. The processed spectral data was stored in a database, with each spectral data entry containing relevant sample information such as sample number, collection date, and experimental conditions. The database was designed using a relational database structure, with tables used to store the spectral data and related information, ensuring data retrieval and manageability. Finally, a recognition model was built based on a deep learning algorithm.

[0110] In the above formula, Indicates the first Eigenvectors of Raman spectra Indicates the first The first group of Raman spectra Raman signal intensity at each data point Indicates the first The average signal intensity of the group of Raman spectra, This represents the total number of sampling points in the Raman spectrum. This represents the regularization parameter to prevent the denominator from approaching zero.

[0111] To calculate these parameters, firstly, the first... Data points of a group of Raman spectra, assuming This indicates that each spectrum contains 1000 data points. Next, the calculation of the... Average signal intensity of the group of Raman spectra The calculation formula is as follows:

[0112] ;

[0113] Assume the first Group of spectral data point signal intensity The sum is 5000, then Then, calculate the sum of the absolute values ​​of the differences between each data point and the mean, i.e. Assuming the calculated result is 800, then calculate the sum of squares of the differences between each data point and the mean, i.e. , assuming the calculation result is 1200, the regularization parameter is selected, these values are substituted into the formula to obtain:

[0114] ;

[0115] Therefore, the eigenvector of the first group of Raman spectra is about 23.09.

[0116] The innovation of the formula is that the sum of the absolute values of the differences between the Raman spectrum data points and the mean value, and the square root of the sum of the squares of these differences, are calculated, and a regularization parameter is introduced, which comprehensively considers the deviation degree and fluctuation amplitude of the spectrum signal, generates an eigenvector that reflects both the overall deviation degree of the spectrum signal and the severity of the signal fluctuation, which helps to more comprehensively characterize the features of the Raman spectrum.

[0117] The calculated eigenvector is compared with the preset reference value, assuming the reference value is 20, since is greater than the reference value, it indicates that the signal deviation degree and fluctuation amplitude of the Raman spectrum are large, which may correspond to the high expression of tumor markers or other significant characteristics in the sample.

[0118] The electrical signal is input to the identification model, and the formula is used:

[0119] ;

[0120] The specific signal feature is calculated, the specific signal feature is extracted, and the tumor marker is identified according to the specific signal feature;

[0121] wherein, represents the specific signal feature value, represents the th enhanced processed electrical signal data value, represents the mean value of the enhanced processed electrical signal data, represents the th background signal data value, represents the mean value of the background signal data, represents the total number of enhanced processed electrical signal data, represents the total number of background signal data, represents the index of the enhanced processed electrical signal data, represents the index of the background signal data.

[0122] The electrical signal is input to the identification model, and the electrical signal refers to a time-varying signal detected by a sensor, the value of which changes continuously over time. In the specific implementation process, first, the original electrical signal data is obtained through a high-precision signal acquisition device, which needs to have microvolt-level precision to capture weak bioelectrical signals. Then, the electrical signal data is preprocessed, which includes filtering and denoising. In the filtering process, a band-pass filtering method is used, with a low-frequency cutoff frequency of 0.5 Hz and a high-frequency cutoff frequency of 100 Hz, to ensure that noise in non-target frequency bands is filtered out. In the denoising process, a neighborhood mean denoising method is used, which calculates the mean of each of the five points before and after the current signal point, and replaces the original value of the current point with the mean, so that the signal curve is smoother, thereby reducing random noise interference. After filtering and denoising, the electrical signal is subjected to amplitude normalization processing. The normalization method is linear normalization, that is, the formula: ;

[0123] wherein, represents the normalized electrical signal value, is the original electrical signal value, and represent the minimum and maximum signal values in the current sampling time window, respectively. This normalization operation ensures that electrical signals of different intensities can be analyzed within a unified numerical range. Subsequently, the normalized electrical signal is input to the identification model, which is a neural network structure composed of a feature extraction unit and a classification unit. The feature extraction unit uses a time-frequency analysis method to extract features from the electrical signal. Specifically, first, the energy value of the electrical signal in a short time window is calculated using the formula:

[0124] ;

[0125] wherein, represents the signal energy value in the short time window, is the window start time, is the window length. This energy feature can effectively reflect the local changes of the electrical signal. Subsequently, the calculated energy feature value is input to the classification unit, which uses multiple thresholds for judgment. Set the thresholds , and . When , it is judged to be a background noise signal, when , it is judged to be a normal physiological signal, and when , it is judged to be a suspicious feature signal, and further analysis is performed to determine whether it has tumor marker characteristics. Finally, specific signal features are obtained, and tumor markers are identified based on the specific signal features.

[0126] ;

[0127] wherein, representing the specific signal characteristic value;

[0128] representing the first enhanced processed electric signal data value, which is derived from the normalized processed electric signal;

[0129] representing the enhanced processed electric signal data mean value, whose calculation formula is as follows:

[0130] ;

[0131] wherein representing the total number of the enhanced processed electric signal data;

[0132] representing the first background signal data value, which is derived from the original background signal without enhancement processing;

[0133] representing the background signal data mean value, whose calculation formula is as follows:

[0134] ;

[0135] wherein representing the total number of the background signal data.

[0136] Parameter assignment and calculation

[0137] Set the number of electric signal sampling data , the data are as follows:

[0138] ;

[0139] Calculate :

[0140] ;

[0141] Set the number of background signal sampling data , the data are as follows:

[0142] ;

[0143] Calculate :

[0144] ;

[0145] Calculate the molecular part:

[0146] ;

[0147] ;

[0148] Calculate the denominator:

[0149] ;

[0150] ;

[0151] ;

[0152] ;

[0153] ;

[0154] Calculate the final result:

[0155] ;

[0156] This result indicates the current signal eigenvalues. A high level may indicate tumor marker characteristics.

[0157] Regarding the training of the aforementioned recognition model:

[0158] Its deep learning algorithm adopts the classic convolutional neural network model, representing each spectral sample as a one-dimensional array. In the first convolutional layer, two one-dimensional convolutional kernels with a kernel size of 2 are used. Then, Flatten is used to flatten the features extracted from the upper layer. Finally, a fully connected layer with a unit number of 2 is used to predict the output through the softmax activation function.

[0159] This invention uses stochastic gradient descent to train the network. The network input size is m*475*2, where m is the number of samples, 475 is the spectral length of the region of interest, the number of input channels is 2, and the network output is one-hot data.

[0160] After training is completed, the model is validated using a validation set, and the convolutional neural network model with the smallest loss is selected as the optimal recognition model.

[0161] S3. Use a spectral acquisition device to acquire Raman spectra of the target tissue slices.

[0162] S4. Convert the Raman spectrum into an electrical signal and input it into the recognition model, extract specific signal features, and identify tumor markers based on the specific signal features;

[0163] During the recognition process:

[0164] The training data of the above model retains the first dimension, while all other dimensions are covered. This data is then input into the trained recognition model for prediction, and the prediction accuracy and LOSS value are recorded.

[0165] Then, 1 is changed to 1+1, the above process is repeated, and 1 is selected to be added to the important region sequence so that the accuracy is maximum and the LOSS value is minimum.

[0166] Then, all dimensions appearing in the important region in the input data are reserved, the above process is repeated, and 1 is selected to be added to the important region sequence again so that the accuracy is maximum and the LOSS value is minimum. Iteration is continuously performed until N important points successively added in the important region sequence no longer change the prediction accuracy, for example, N=5, and finally the important region sequence is obtained;

[0167] After obtaining the important region sequence, the discrete data is continuous to obtain the important region of the final Raman spectrum. The density of the points in the important region sequence represents the importance of the region where the point is located. The idea of step-by-step propagation is used to assign weights to each point on the full spectrum. The important points in the important region sequence are taken out in turn, and the point is assigned a weight w. Then the weight is continuously attenuated to the leftmost end and the rightmost end of the whole spectrum step by step from the left and right of the point, that is, the weight of the kth point to the left and right of the point is Wxp k , wherein p is the attenuation rate, for example, p=0.9, and the same applies to each point in the important region sequence. The sequence weight of each point in the full spectrum is added after the important region sequence is traversed, and the final spectrum region sequence weight is obtained. The above process can ensure that when the important points are adjacent, the weights of these points and adjacent points are high. The final generated spectrum region weight can be used to accurately identify specific signal characteristics and corresponding tumor markers.

[0168] S5. Constructing an enhanced Raman image of the target tissue section according to the collected Raman spectrum and image reconstruction technology;

[0169] S6. Labeling the spatial position and quantity concentration of the tumor marker in the enhanced Raman image by using the identification result of the tumor marker and the spatial resolution algorithm, and obtaining a target tissue pathology image labeled with the tumor marker;

[0170] S7. Outputting the target tissue pathology image.

[0171] In the above embodiment 1 and embodiment 2:

[0172] The present application realizes the excitation and collection of Raman scattered light of multiple different wavelengths through the design of multiple optical channels, thereby greatly improving the speed and efficiency of data collection. Moreover, the beam splitter and high-resolution spectrum sorter are used for each optical channel splitter, thereby effectively ensuring the accurate focusing of the laser and the high-quality output of the spectrum signal.

[0173] The present application takes directly collected Raman spectrum as training data, does not involve a complex spectrum enhancement process, uses a trained recognition model based on a deep learning algorithm to extract and effectively recognize tumor markers, does not need antibody reagents and staining processes, greatly shortens the cycle of pathological diagnosis, and reduces the cost of diagnosis.

[0174] The present application can present the distribution of tumor markers on a target tissue slice in the form of an image through Raman imaging processing, provide a visual pathological image for doctors, seamlessly connect and transition for pathologists, help doctors obtain clinical interpretation and make treatment decisions, overcome the shortcoming that existing spectroscopy detection technologies are not easy for doctors to understand, and improve the efficiency and accuracy of diagnosis.

[0175] Embodiment 3

[0176] A multi-channel Raman spectrum analysis system for tumor marker detection, comprising an acquisition module, an enhancement processing module, an analysis module and an output module.

[0177] Regarding the acquisition module, the Raman spectrum of a target tissue slice is acquired.

[0178] The acquisition module comprises a spectrum acquisition device provided with a plurality of independent optical channels, and a laser source, a spectrometer, a detector and a TE refrigeration module are arranged in each independent optical channel.

[0179] Regarding the enhancement processing module, the acquired Raman spectrum is converted into an electrical signal, and the electrical signal is also used for enhancement processing through a neural network.

[0180] The enhancement processing module specifically enhances the electrical signal through a Raman signal separation algorithm based on a convolutional neural network.

[0181] Regarding the analysis module, the tumor markers of a target tissue slice are recognized through Raman spectrum analysis.

[0182] The analysis module analyzes and processes the acquired Raman spectrum by using a high-dimensional spectrum analysis technology, which combines a composite multi-dimensional quantum analysis (such as a quantum support vector mechanism) and a deep learning algorithm. The specific extraction of specific spectral characteristics of tumor markers is realized through deep learning of a Raman spectrum database, so as to ensure that trace information can be extracted from complex biological samples, and tumor markers can be recognized with ultra-high sensitivity even in the case of extremely low concentration.

[0183] Regarding the output module, the tumor markers are marked according to the recognition result of the analysis module, and the target tissue pathological image with the tumor markers marked is output.

[0184] Labeling: Introduce biomarker spectral mapping technology which combines image reconstruction technology and spatial resolution algorithm. Specifically, by constructing the enhanced Raman image of the target tissue section and marking the spatial position and number concentration of the tumor marker in the enhanced Raman image, the target histopathological image is obtained.

[0185] Output: Output the target histopathological image (the spatial position and number concentration of the tumor marker are marked in the image). In addition, the possible clinical significance of these tumor markers can be analyzed in the image to provide necessary decision support for doctors and provide reference for formulating early diagnosis and individualized treatment plan.

[0186] Example 4

[0187] This embodiment is a multi-channel Raman spectral analysis method for tumor marker detection based on the above-mentioned embodiment 3, specifically comprising the following steps:

[0188] S1. Hardware pretreatment

[0189] Calibrate the spectral acquisition device provided with multiple independent optical channels, and each independent optical channel is provided with a laser source, a spectrometer, a detector and a TE refrigeration module.

[0190] The size of the overall spectral acquisition system is 144x92x46 mm, and the optical path of the spectrometer is built with a Czerney-Turner system as shown in Figure 3 Two spherical mirrors are used as collimating lenses and converging lenses respectively, the paths of incident light and imaging light cross each other, and the slit width is 25 μm, so as to improve the luminous flux as much as possible while ensuring that the resolution is not greatly affected.

[0191] Specifically, the spectral acquisition device is calibrated and debugged by using mercury lamp method, and the multiple measurement results prove that the spectral acquisition device has a spectral coverage range of 785nm-1050nm, a resolution of 6.45cm -1 (0.54nm, FWHM at 912nm), which improves the resolution on the premise of ensuring that the Raman spectrum can fully reflect the tumor marker concentration information. At the same time, the signal-to-noise ratio of the system is higher than 6000:1, which ensures that the acquired spectral signal can be used as a reliable information basis for quantitative analysis of tumor markers, and silicon wafer is used as a standard substance to correct the wave number accuracy.

[0192] S2. Software pretreatment

[0193] Construct a Raman signal separation algorithm based on convolutional neural network, a Raman spectral database of tissue sections containing tumor markers, and an identification model based on deep learning algorithm, and train the identification model using the database.

[0194] The expression of the Raman signal separation algorithm is:

[0195] R=L*B+N; wherein R is the input measured spectrum signal, L represents the Raman spectrum independent spectral peak signal, B represents the instrument broadening signal and background baseline signal, N represents the measurement noise, and "*" represents convolution operation;

[0196] ; wherein x represents the Raman shift, c represents the center of the Lorentz function, wL represents the half-height half-width of the Lorentz function, and S represents the full spectrum area of the Lorentz function;

[0197] ; wherein w G represents the half-height half-width of the Gaussian function .

[0198] Regarding the training of the identification model:

[0199] The deep learning algorithm thereof adopts a classical convolutional neural network model, each spectrum sample is represented as a one-dimensional array, 2 one-dimensional convolution kernels with a kernel size of 2 are used in the first convolution layer, then the features extracted in the upper layer are flattened by using a Flatten, and finally a fully connected layer with a unit number of 2 is used, and a prediction output is performed through a softmax activation function.

[0200] The network is trained using a stochastic gradient descent. The input size of the network is m*475*2, wherein m is the sample number, 475 is the spectrum length of the region of interest, the input channel number is 2, and the output of the network is one-hot type data.

[0201] After the training is completed, the verification set is verified, and the convolutional neural network model with the smallest loss is used as the optimal identification model.

[0202] S3. Raman spectrum acquisition is performed on the target tissue slice by using a spectrum acquisition device.

[0203] S4. The collected Raman spectrum is converted into an electrical signal, and in the conversion process, the signal is also preliminarily processed by using a spectrum depth weaving technology (SDW), such as signal amplification, noise filtering and signal stabilization.

[0204] S5. The electrical signal is subjected to enhancement processing;

[0205] When the electrical signal is subjected to enhancement processing by using the above Raman signal separation algorithm:

[0206] The Lorentz spectral peak signal I and the parameter vector β thereof and the instrument broadening signal B and the half-width parameter W thereof are separated from the original spectrum G ;

[0207] Set the initial value of the Lorentz spectral peak signal parameter vector β, β = ∅, and the initial value of the number of Raman spectral peaks contained therein k is 0. Set the initial value of the half-width parameter WG of the instrument broadening signal, WG = 5.

[0208] From the above, the initial value of the estimated spectrum is calculated:

[0209] Since β = ∅, there is , and the initial value of the residual spectrum is the original spectrum.

[0210] A new Raman spectral peak is added to the Lorentz spectral peak signal, and the number of Raman spectral peaks is updated, k = k + 1. The initial value of the parameter of the newly added spectral peak is set as follows: find the maximum value hk of the residual spectrum signal and the wave number ck at which it is located. Then search from both sides of the wave number ck to find the nearest xleft and xright, respectively, such that Y(x left +1) ≥ hk / 2 and Y(x left ) < hk / 2, and Y(x right -1) ≥ hk / 2 and Y(x right ) < hk / 2. Set the initial value of the parameter vector βk, w LK = min{ck-x left , x right -ck}, SK=2w LK h K , βk= (S K ,Ck,w LK ). Extend the parameter vector, β = (β 1， β 2，…, β k-1， β k ).

[0211] Optimize β and wG to obtain the estimated spectrum after parameter update. According to the original spectrum and the estimated spectrum, the residual spectrum signal r(v) is calculated. The residual spectrum is calculated as follows:

[0212] r(x) = Y(x) - R(x, β, w G ) for x = x b , x b+1 ,..., x e .

[0213] When the maximum value of the residual spectrum is less than the termination threshold or the number of iterations is greater than or equal to the maximum number of iterations, output the Lorentz spectral peak signal L and its parameters, the vector β, and the instrument broadening signal B and its half-width parameter; otherwise, continue iteration.

[0214] S6. input the enhanced signal into the identification model, extract specific signal features, and identify tumor markers according to the specific signal features;

[0215] S7. Constructing an enhanced Raman image of the target tissue section according to the collected Raman spectrum and image reconstruction technology;

[0216] S8. Labeling the spatial position and number concentration of the tumor marker in the enhanced Raman image by using the recognition result of the tumor marker and the spatial resolution algorithm, to obtain a target tissue pathological image labeled with the tumor marker;

[0217] S9. Outputting the target tissue pathological image.

[0218] Compared with Embodiment 1 and Embodiment 2, the above-mentioned Embodiment 3 and Embodiment 4 are additionally provided with an enhancement processing module, so that the present application can optimize the processing parameters of Raman spectrum signal enhancement in real time through the alternative convergence of the neural network, and further accurately eliminate the noise and interference introduced by sample preparation and environmental factors, to ensure the high consistency and repeatability of the spectrum data.

[0219] Embodiment 5

[0220] A multi-channel Raman spectrum analysis system for tumor marker detection, comprising a pretreatment module, an acquisition module, an enhancement processing module, an analysis module and an output module.

[0221] Regarding the pretreatment module, it is used for adapting and modifying the target tissue section by magnetic nanoparticles.

[0222] The magnetic nanoparticles at least include one of gold nanoparticles and silver nanoparticles.

[0223] Regarding the acquisition module, it is used for acquiring the Raman spectrum of the target tissue section.

[0224] The acquisition module comprises a spectrum acquisition device provided with a plurality of independent optical channels, and a laser source, a spectrometer, a detector and a TE refrigeration module are arranged in each independent optical channel.

[0225] Regarding the enhancement processing module, it is used for converting the collected Raman spectrum into an electrical signal and enhancing the electrical signal through a neural network.

[0226] The enhancement processing module specifically enhances the electrical signal through a Raman signal separation algorithm based on a convolutional neural network.

[0227] Regarding the analysis module, it is used for recognizing the tumor marker of the target tissue section through Raman spectrum analysis.

[0228] The analysis module analyzes and processes the collected Raman spectrum by using a high-dimensional spectral analysis technology, which combines a composite multi-dimensional quantum analysis (such as a quantum support vector machine mechanism) and a deep learning algorithm. Specifically, the specific spectral characteristics of the tumor markers are accurately extracted through deep learning of the Raman spectrum database, so as to ensure that trace information can be extracted from complex biological samples, and the tumor markers can be recognized with ultra-high sensitivity even at a very low concentration.

[0229] As to the output module, it is used for marking the tumor markers according to the recognition result of the analysis module, and also used for outputting the target histopathological image marked with the tumor markers.

[0230] Marking: the biomarker spectral mapping technology is introduced, which combines an image reconstruction technology and a spatial analysis algorithm. Specifically, the target histopathological image is obtained by constructing an enhanced Raman image of the target tissue slice and marking the spatial position and number concentration of the tumor markers in the enhanced Raman image.

[0231] Output: the target histopathological image (in which the spatial position and number concentration of the tumor markers are marked) is output. In addition, the possible clinical significance of the tumor markers can be analyzed in the image, which provides necessary decision support for doctors and provides a reference for formulating early diagnosis and individualized treatment plans.

[0232] Embodiment 6

[0233] This embodiment is a multi-channel Raman spectrum analysis method for tumor marker detection based on the method provided in Embodiment 5, and specifically includes the following steps:

[0234] S1. Hardware pretreatment

[0235] The spectral acquisition device is calibrated, and a plurality of independent optical channels are provided in the spectral acquisition device, and a laser source, a spectrometer, a detector and a TE refrigeration module are provided in each independent optical channel.

[0236] The size of the overall spectral acquisition system is 144x92x46 mm, and the Czerny-Turner system is used to build the optical path of the spectrometer as shown in Figure 3 Two spherical mirrors are used as collimating lenses and converging lenses respectively, the paths of the incident light and the imaging light cross each other, and the slit width is 25μm, so as to improve the luminous flux as much as possible while ensuring that the resolution is not greatly affected.

[0237] The spectral acquisition device is calibrated and debugged by using the mercury lamp method, and the multiple measurement results prove that the spectral acquisition device has a spectral coverage range of 785nm-1050nm and a resolution of 6.45cm -1(0.54 nm, full width at half maximum at 912 nm), which improves the resolution while ensuring that the Raman spectrum can fully reflect the concentration information of the tumor marker. At the same time, the signal-to-noise ratio of the system is higher than 6000:1, which ensures that the collected spectrum signal can be used as a reliable information basis for quantitative analysis of the tumor marker, and the wave number accuracy is corrected using a silicon wafer as a standard.

[0238] S2. Sample pretreatment

[0239] Select aptamers with high affinity and high specificity to breast cancer biomarkers;

[0240] The streptavidin-coated magnetic nanoparticles are rinsed with buffer solution and then resuspended in buffer solution. Biotinylated TBA15 is added and combined with the buffer solution, and reacted at room temperature for 30 min under smooth oscillation conditions. TBA15 is fixed on the surface of the magnetic nanoparticles through avidin-biotin interaction. Streptavidin and small ligand biotin exhibit extremely high binding affinity and specificity. A magnetic separator and B&W buffer solution are used to wash away unreacted reagents. The magnetic nanoparticles are resuspended in a buffer solution. In this way, the aptamer is modified on the surface of the magnetic beads magnetic nanoparticles;

[0241] A layer of chitosan is coated on the surface of the modified magnetic nanoparticles to improve biocompatibility and stability;

[0242] The aptamer-modified magnetic nanoparticles are used as capture probes;

[0243] The sample is incubated with the capture probes to form a sandwich complex of aptamer-tumor marker-aptamer;

[0244] A magnetic separator is used to remove unbound components, and the complex is washed with buffer solution;

[0245] The sandwich complex is transferred to the SERS substrate for Raman spectrum measurement.

[0246] The magnetic nanoparticles create an enhanced spectral reaction interface on the surface of the target tissue section. This interface significantly improves the efficiency of Raman scattering, enabling effective detection of even extremely low concentrations of tumor markers. The multi-frequency resonant Raman scattering (MFRS) technique is used, which configures a multi-wavelength laser system to simultaneously excite multiple tumor markers in the sample. Each wavelength of laser light is precisely adjusted to match the resonant frequency of a specific tumor marker, thereby maximizing the generation of Raman signals.

[0247] S3. Software pretreatment

[0248] A Raman signal separation algorithm based on a convolutional neural network, a Raman spectrum database of a tissue slice containing a tumor marker, and an identification model based on a deep learning algorithm are constructed, and the identification model is trained using the database.

[0249] S4. Raman spectrum acquisition of the target tissue slice is performed using a spectrum acquisition device. Specifically, multiple different wavelengths of Raman scattered light are collected through multiple channels to improve the speed and efficiency of data acquisition.

[0250] S5. The collected Raman spectrum is converted into an electrical signal, and in this conversion process, the signal is also preliminarily processed by a spectral deep weaving technology (SDW), such as signal amplification, noise filtering, and signal stabilization.

[0251] S6. The electrical signal is enhanced;

[0252] S7. The enhanced electrical signal is input into the identification model, specific signal features are extracted, and the tumor marker is identified according to the specific signal features;

[0253] S8. An enhanced Raman image of the target tissue slice is constructed according to the collected Raman spectrum and image reconstruction technology;

[0254] S9. The identification result of the tumor marker and the spatial resolution algorithm are used to mark the spatial position and concentration of the tumor marker in the enhanced Raman image, and a target histopathological image marked with the tumor marker is obtained;

[0255] S10. The target histopathological image is output, so that the doctor can see the specific distribution of the tumor marker in the biological sample, and thus provide more accurate diagnostic information.

[0256] Compared with Embodiment 6 and Embodiments 3 and 4, the above-mentioned Embodiment 5 adds a pretreatment module, so that the present application can improve the detection sensitivity of the tumor marker by means of magnetic nanoparticles, thereby further ensuring that the tumor-related trace biomarkers can be effectively detected at low concentrations.

[0257] As described above, the present application also takes breast cancer biomarkers as an example, and the breast cancer biomarkers can be represented by circulating tumor cells (CTCs), microRNAs, and specific proteins:

[0258] S1. Sample collection

[0259] The serum sample is collected from the patient's body, pretreated with a buffer reagent to remove impurities and interfering substances, and stored at low temperature.

[0260] S2. Sample processing

[0261] Select aptamers with high affinity and high specificity to breast cancer biomarkers;

[0262] The streptavidin-coated magnetic nanoparticles are rinsed with buffer and then resuspended in buffer. Biotinylated TBA15 is added and allowed to bind to the buffer for 30 min at room temperature with smooth shaking. TBA15 is immobilized on the surface of the magnetic nanoparticles through avidin-biotin interaction. Streptavidin and small ligand biotin exhibit extremely high binding affinity and specificity. Unreacted reagents are washed away using a magnetic separator and B&W buffer. The magnetic nanoparticles are resuspended in buffer solution. In this way, the aptamer is modified on the surface of the magnetic beads magnetic nanoparticles;

[0263] A layer of chitosan is coated on the surface of the modified magnetic nanoparticles to improve biocompatibility and stability;

[0264] The aptamer-modified magnetic nanoparticles are used as capture probes;

[0265] The sample is incubated with the capture probes to form a sandwich complex of aptamer-tumor marker-aptamer;

[0266] The unbound components are removed using a magnetic separator, and the complex is washed with buffer;

[0267] The sandwich complex is transferred to a SERS substrate for Raman spectrum measurement.

[0268] S3. Use a spectrum acquisition device to collect the Raman spectrum of the processed sample.

[0269] S4. A series of pretreatments such as cosmic ray muon noise removal, convolution smoothing filtering, fluorescence background correction, and normalization are performed on the collected Raman spectrum.

[0270] S5. Convert the Raman spectrum into an electrical signal.

[0271] S6. Perform enhancement processing on the electrical signal.

[0272] A specific convolutional neural network Raman signal separation algorithm is proposed for enhancement processing.

[0273] The algorithm expression is R=L*B+N; where R is the input measured spectrum signal, L represents the independent spectral peak signal of the Raman spectrum, B represents the instrument broadening signal and background baseline signal, N represents the measurement noise, and "*" represents convolution operation;

[0274] ; where x represents the Raman shift, c represents the center of the Lorentz function, wL represents the half-height half-width of the Lorentz function, and S represents the full spectrum area of the Lorentz function;

[0275] ; wherein w G represents the half-height half-width of the Gaussian function .

[0276] In the processing of the electric signal enhancement using the above algorithm:

[0277] The Lorentzian spectral peak signal I and its parameter vector β and the instrument broadening signal B and its half-width parameter W separated from the original spectrum G ;

[0278] Set the initial value of the parameter vector β of the Lorentzian spectral peak signal, β = ∅, and the initial value of the number of Raman spectral peaks contained therein k is 0. Set the initial value of the half-width parameter WG of the instrument broadening signal, WG = 5.

[0279] From the above, calculate the initial value of the estimated spectrum:

[0280] Since β = ∅, we have , and the initial value of the residual spectrum is the original spectrum.

[0281] Add a new Raman spectral peak to the Lorentzian spectral peak signal, update the number of Raman spectral peaks, k = k + 1. Set the initial value of the parameter of the newly added spectral peak, as follows: find the maximum value hk of the residual spectrum signal and the wave number ck at which it is located. Then search from both sides of the wave number ck to find the nearest xleft and xright, respectively, such that Y(x left +1) ≥ hk / 2 and Y(x left ) < hk / 2, and Y(x right -1) ≥ hk / 2 and Y(x right ) < hk / 2. Set the initial value of the parameter vector βk, w LK = min{ck-x left , x right -ck}, SK=2w LK h K , βk= (S K , Ck, w LK ). Extend the parameter vector, β = (β 1， β 2，…, β k-1， β k ).

[0282] Optimize β and wG to obtain the estimated spectrum after updating the parameters. According to the original spectrum and the estimated spectrum, calculate the residual spectrum signal r(v). The residual spectrum is calculated as follows:

[0283] r(x) = Y(x) - R(x, β, w G ) for x = x b , x b+1..., x e .

[0284] When the maximum value of the residual spectrum is less than the termination threshold or the number of iterations is greater than or equal to the maximum number of iterations, the Lorentz spectral peak signal L and its parameters, the vector β and the instrument broadening signal B and its half-width parameter are output; otherwise, iteration is continued.

[0285] S7. Constructing a Raman spectrum database of a tissue section containing a tumor marker and an identification model based on a deep learning algorithm; it is known in the art that the Raman spectrum database in the present application can be a Raman spectrum database of a biomarker, a classified Raman spectrum database of all biomarkers of a certain type of cancer, or a hierarchical classified Raman spectrum database of all biomarkers of multiple types of cancer;

[0286] S8. Training the identification model using the database. The database is divided into a training set and a validation set, and the identification model is trained by the training set, and the trained identification model is verified by the validation set, and the verification passes to be a usable identification model;

[0287] Regarding the training of the identification model:

[0288] The deep learning algorithm thereof adopts a classic convolutional neural network model, each spectrum sample is represented as a one-dimensional array, 2 one-dimensional convolution kernels with a kernel size of 2 are used in the first convolution layer, then the features extracted from the upper layer are flattened by Flatten, and finally a fully connected layer with 2 units is used to output the prediction by a softmax activation function.

[0289] The present application uses stochastic gradient descent to train the network. The input size of the network is m*475*2, where m is the number of samples, 475 is the spectral length of the region of interest, the input channel number is 2, and the output of the network is one-hot type data.

[0290] After training, the validation set is verified, and the convolutional neural network model with the smallest loss is used as the optimal identification model.

[0291] S9. Inputting the enhanced electric signal into the identification model, extracting specific signal characteristics of breast cancer biomarkers, and identifying tumor markers (such as circulating tumor cells (CTCs), microRNAs, and specific proteins) according to the specific signal characteristics.

[0292] In the identification process:

[0293] The above model training data is retained in the first dimension, and all other dimensions are covered, and it is input into the trained identification model for prediction, and the prediction identification accuracy and LOSS value are recorded.

[0294] Then, 1 is changed to 1+1, the above process is repeated, and 1 is selected to be added to the important region sequence so as to maximize the accuracy and minimize the LOSS value.

[0295] Then, all dimensions in the input data that appear in the important region are reserved, the above process is repeated, and 1 is selected to be added to the important region sequence so as to maximize the accuracy and minimize the LOSS value. Iteration is continuously performed until N important points that are continuously added to the important region sequence no longer change the prediction accuracy, for example, N=5, and finally, the important region sequence is obtained.

[0296] After the important region sequence is obtained, the discrete data is continuous to obtain the important region of the final Raman spectrum. The density of the points in the important region sequence represents the importance of the region where the point is located. The idea of step-by-step propagation is used to assign weights to each point on the full spectrum. The important points in the important region sequence are taken out in turn, and the point is assigned a weight w. Then the weight is continuously attenuated to the leftmost end and the rightmost end of the whole spectrum from the left and right of the point, that is, the weight of the kth point to the left and right of the point is Wxp k , where p is the attenuation rate, for example, p=0.9, and the same is true in turn. Each point in the important region sequence will assign a weight to the whole spectrum sequence once. After the important region sequence is traversed, the sequence weight of each point in the full spectrum is added to obtain the final spectrum region sequence weight. The above process can ensure that when the important points are adjacent, the weights of these points and adjacent points are high. The finally generated spectrum region weight can be used to accurately identify specific signal characteristics and corresponding tumor markers.

[0297] S10. Constructing an enhanced Raman image of the breast tissue section sample according to the collected Raman spectrum and image reconstruction technology;

[0298] S11. Labeling the spatial position and quantity concentration of the tumor markers in the enhanced Raman image by using the identification result of the tumor markers and the spatial resolution algorithm (color brightness can be used for labeling), to obtain a breast tissue pathological image in which the tumor markers are labeled.

[0299] S12. Outputting the breast tissue pathological image. The spatial position and quantity concentration of the circulating tumor cells, microRNAs, and specific proteins are indicated in the breast tissue pathological image. In addition, the possible clinical significance of the circulating tumor cells, microRNAs, and specific proteins can be analyzed in the breast tissue pathological image.

[0300] In addition, to achieve the above object, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the multi-channel Raman spectrum analysis system for tumor marker detection.

[0301] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, embodiments of the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0302] Embodiments of the present application are described herein with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable data processing terminal devices to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal devices, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0303] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing terminal device to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0304] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device to cause a series of operational steps to be performed on the computer or other programmable terminal device to produce a computer implemented process such that the instructions which execute on the computer or other programmable terminal device provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams.

[0305] It should be noted that the above examples are only used to illustrate the technical solutions of the embodiments of the present application, but not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent ones, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application.

[0306] Although the embodiments of the present application have been shown and described, it should be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-channel Raman spectroscopy analysis system for tumor marker detection, characterized in that: It includes a data acquisition module, an enhancement processing module, an analysis module, and an output module; The acquisition module is used to acquire the Raman spectrum of the target tissue slice, and the Raman spectrum includes Raman scattered light of multiple different wavelengths; The enhancement processing module is used to convert the Raman spectrum acquired by the acquisition module into an electrical signal, and also to enhance the electrical signal; The analysis module is used to identify tumor markers by analyzing the enhanced electrical signals. The output module is used to label tumor markers according to the identification results of the analysis module, and is also used to output the target tissue pathological image labeled with tumor markers; The enhancement processing module enhances the electrical signal using a Raman signal separation algorithm based on a convolutional neural network; the expression for the Raman signal separation algorithm based on a convolutional neural network is as follows: R = L*B + N; where R is the measured spectral signal, L represents the independent peak signal of the Raman spectrum, B represents the instrument broadening signal and the background baseline signal, N represents the measurement noise, and "*" represents the convolution operation. The independent peak signal L of the Raman spectrum is represented as: Where x represents the Raman shift, c represents the center of the Lorentz function, wL represents the half-width and half-height of the Lorentz function, and S represents the area of ​​the full spectrum of the Lorentz function. The instrument broadening signal and background baseline signal B are represented as follows: Among them, w G Represents the half-height and half-width of the Gaussian function 2. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 1, characterized in that: It also includes a preprocessing module; The preprocessing module is used to modify the target tissue slices by adapting them with magnetic nanoparticles.

3. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 2, characterized in that, The preprocessing steps of the preprocessing module include: Aptamers are selected based on tumor markers in the target tissue slices; Magnetic nanoparticles are modified using the aptamer; A layer of chitosan was coated on the surface of the modified magnetic nanoparticles to serve as a capture probe; The target tissue slice was incubated with the capture probe to form a sandwich complex of aptamer-target tissue slice-aptamer; After magnetic separation and washing with buffer, the sandwich complex was transferred to a SERS substrate.

4. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 3, characterized in that: The step of modifying magnetic nanoparticles with the aptamer includes: The streptavidin-coated magnetic nanoparticles were rinsed with buffer solution and then resuspended in buffer solution to obtain a suspension. The aptamer was added to the suspension and reacted at room temperature under smooth oscillation for 30 min. Magnetic nanoparticles with aptamers fixed on their surface were obtained by magnetic separation.

5. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 1, characterized in that: The acquisition module includes a spectral acquisition device with multiple independent optical channels, and each independent optical channel is equipped with a laser source, a spectrometer, a detector and a TE cooling module.

6. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 1, characterized in that, The tumor markers mentioned include: A Raman spectral database containing tumor markers was constructed, the spectral data matrix was obtained, and after data normalization, the eigenvectors of the Raman spectra were calculated using the following formula: Calculate the Raman spectral feature vector V, and construct a recognition model based on a deep learning algorithm; Among them, V i S represents the eigenvector of the i-th Raman spectrum. ij This represents the Raman signal intensity of the j-th data point in the i-th Raman spectrum. λ represents the average signal intensity of the i-th group of Raman spectra, n represents the total number of sampling points of the Raman spectra, and λ represents the regularization parameter to prevent the denominator from approaching zero. The recognition model is trained using a database; The electrical signal is input into the recognition model, using the formula: Calculate specific signal features, extract specific signal features, and identify tumor markers based on the specific signal features; in, E represents the characteristic value of a specific signal. P μ represents the p-th enhanced electrical signal data value. E B represents the mean value of the enhanced electrical signal data. q μ represents the q-th background signal data value. B represents the mean of the background signal data, N represents the total number of electrical signal data after enhancement processing, M represents the total number of background signal data, p represents the index of the electrical signal data after enhancement processing, and q represents the index of the background signal data.

7. The multi-channel Raman spectroscopy analysis system for tumor marker detection according to claim 1, characterized in that, The tumor markers mentioned include spatial location markers and quantity concentration markers.

Images