ICCD and fiber-optic gated raman spectroscopy system

Abstract

This invention discloses an application method and system for a Raman spectroscopy system based on ICCD and fiber optic gating, comprising: S1, a control unit switching the working state of a laser source to generate a laser that excites a Raman signal in the target tissue; S2, the laser reaching the part to be detected through a flexible fiber optic probe, and the control unit switching the working states of the gating unit and the ICCD detector to acquire the Raman spectral signal of the target tissue through the fiber optic probe; S3, the control unit outputting the acquired Raman spectral signal to a spectral analysis module, which processes the acquired Raman spectral signal based on a built-in deep learning model to analyze and evaluate the malignancy of the acquired tissue. This invention provides an application method and system for a Raman spectroscopy system based on ICCD and fiber optic gating, which, by combining ICCD and gating technology, enhances the Raman signal intensity, reduces fluorescence background, and thus improves detection sensitivity and efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical testing technology. More specifically, this invention relates to a Raman spectroscopy system based on ICCD and fiber optic gating. Background Technology

[0002] Cancer is a common challenge facing humanity today. Early diagnosis of cancer is crucial for improving treatment outcomes and patient survival rates. Traditional histopathological examination methods typically require surgical biopsy or needle biopsy, which are invasive, can lead to complications, and are time-consuming.

[0003] Of course, existing technologies also employ non-invasive detection methods. For example, imaging techniques (such as ultrasound) are used to detect certain areas (such as the breast). For instance, patent application number 202110038828.6, entitled "Cancer Clinical Indicator Assessment System Based on Radiomics Qualitative Algorithm," is used to provide non-invasive, accurate, and non-subjective judgments on early diagnosis, prognosis, and drug efficacy subtypes of cancer. However, this technology has a problem: the system can only achieve tumor morphological judgment by extracting multiple imaging features. Tumors are highly heterogeneous and have multiple morphological manifestations. Radiomics analysis, based on one dimension, is difficult to comprehensively and accurately reflect the molecular biological characteristics and immunological characteristics of tumors, thus affecting the accuracy of diagnosis.

[0004] In addition, some products use test strips to diagnose cancer. For example, patent application number 202211408042.X, entitled "A Diagnostic Product and Model for Gastric Cancer," is used to diagnose early-stage gastric cancer. It has very high sensitivity and specificity and can be used for early screening of gastric cancer, thereby reducing the mortality rate of gastric cancer. However, the problem is that this test strip is an in vitro diagnostic product and cannot achieve real-time monitoring or accurately locate the cancerous tissue. Further confirmation with other detection methods is still required.

[0005] Raman spectroscopy, as a non-invasive molecular imaging technique, can analyze the relative content and distribution of proteins, lipids, nucleic acids, and other components in cancerous cells at the molecular level, providing a deeper understanding of tumor heterogeneity. It can be applied to tumor identification and classification, leading to more accurate diagnosis. Furthermore, combining Raman spectroscopy with fiber optic probes allows for real-time acquisition of tissue Raman spectral information during surgery, helping doctors quickly determine tumor boundaries and properties, thus enabling more precise surgical resection. However, traditional Raman spectrometers, when used in vivo, are limited in their detection sensitivity and efficiency due to the fluorescence background of biological tissues and the relatively low Raman signal intensity. Traditional Raman spectrometers use continuous lasers as excitation sources and detectors with exposure times on the order of milliseconds (or even longer) to acquire signals. This leads to problems such as strong fluorescence interference and high background noise. In order to reduce the excitation of natural fluorescence and reduce the intensity of the fluorescence signal, a longer laser wavelength (such as 785nm) is usually selected. However, the Raman scattering cross section is inversely proportional to the fourth power of the laser wavelength, which further weakens the Raman signal intensity. The combination of weak signal and strong background makes the rapid acquisition of Raman signals from biological samples a major challenge. Therefore, there is currently no mature technology to apply Raman spectroscopy to the early diagnosis of cancer. Summary of the Invention

[0006] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0007] To achieve these objectives and other advantages of the present invention, a method for applying an ICCD-based and fiber-optic gated Raman spectroscopy system is provided, comprising:

[0008] S1. The control unit switches the working state of the laser source to generate a laser that excites the target tissue to produce a Raman signal.

[0009] S2. The laser reaches the part to be detected through a flexible fiber optic probe. The control unit switches the working state of the gating unit and the ICCD detector to acquire the Raman spectrum signal of the target tissue through the fiber optic probe.

[0010] S3. The control unit outputs the acquired Raman spectral signal to the spectral analysis module. The spectral analysis module processes the acquired Raman spectral signal based on the built-in deep learning model to analyze and evaluate the malignancy of the acquired tissue.

[0011] S4. The control unit outputs and / or displays the processing results of the spectral analysis module.

[0012] Preferably, in S2, the method for acquiring the Raman spectral signal of the target tissue is as follows: within a predetermined area of ​​the tissue to be acquired, 10 points are randomly selected to acquire Raman spectra, and the acquisition time for each point is at least 10 seconds.

[0013] Preferably, in S2, the gating unit is used to control the exposure time of the ICCD to ensure that the shutter speed of the ICCD is 500 ps and the gate width is 5 ns.

[0014] Preferably, in S3, the method for obtaining the deep learning model includes:

[0015] S30. The acquired Raman spectral signals are preprocessed and features are extracted to obtain the acquired tissue component spectra;

[0016] S31. The standard Raman spectra of basic components are learned through a neural network model, and the collected tissue component spectra and normal tissue spectra are used as training data for the neural network model.

[0017] The contribution of basic components in the collected tissue composition spectrum is analyzed using a trained neural network model to provide the proportional relationship of basic components in the tissue.

[0018] S32. Based on the characteristics of Raman spectroscopy data, a fusion neural network model based on convolutional neural network and long short-term memory network is constructed and trained. The performance of the neural network model is evaluated by cross-validation method. Then, the weights W1 and W2 of the convolutional neural network and long short-term memory network are assigned values ​​to obtain the trained deep learning model.

[0019] The basic components include DNA, fat, histones, collagen, and actin.

[0020] Preferably, in S3, the trained deep learning model is compressed and deployed on the spectral analysis module to enable real-time analysis and feedback of the acquired Raman spectral data.

[0021] Preferably, in S30, the preprocessing includes: noise reduction, baseline correction, normalization, and spectral distortion correction operations on the acquired Raman spectra.

[0022] The feature extraction method includes:

[0023] S301. Extract effective feature information from multiple dimensions in the preprocessed Raman spectrum signal image;

[0024] S302. The spectral peak feature extraction algorithm based on continuous wavelet transform extracts key information that is biologically significant for tumor diagnosis from effective feature information, and then obtains spectral peak feature data.

[0025] The effective feature information refers to feature peak fragments, which are used to characterize the attribution of biologically significant peaks for tumor diagnosis.

[0026] The key information includes the peak position, peak intensity, and peak width of the spectral peaks in the characteristic spectral peak segment.

[0027] Preferably, in S301, the attribution of biologically significant spectral peaks for tumor diagnosis is characterized by the following characteristic spectral peak fragments:

[0028] Through 616-626cm -1 Characteristic peak fragments characterize amide IV;

[0029] Passing through 637-647cm -1 Characteristic spectral peak fragments characterize CS stretching vibrations;

[0030] Passing 785-795cm -1 Characteristic peak fragments characterize DNA;

[0031] Through 855-865cm -1 Characteristic peak fragments characterize lipids;

[0032] Passing 998-1008cm -1 Characteristic peak fragments characterize phenylalanine;

[0033] Through 1075-1085cm -1 Characteristic peak fragments characterize DNA;

[0034] Passing through 1235-1245cm -1 Characteristic peak fragments characterize amide III;

[0035] Through 1318-1328cm -1 Characteristic spectral peak fragments characterize CH deformation vibrations;

[0036] Through 1444-1454cm -1 Characteristic spectral peak fragments characterize CH2 bending vibrations;

[0037] Through 1547-1557cm -1 Characteristic peak fragments characterize amide II;

[0038] Through 1611-1621cm -1 Characteristic peak fragments characterize tryptophan;

[0039] Through 1664-1674cm -1 Characteristic peak fragments are used to characterize amine I.

[0040] Preferably, in S31, the neural network model is trained to determine the proportional relationships of basic components in the tissue in the following manner:

[0041] S310. Obtain the simulated tissue spectrum S using the following formula:

[0042]

[0043] In the above formula, S 0 represents the spectrum of normal tissue. S 1- S 5 represent the spectra of DNA, lipids, histones, collagen, and actin, respectively. a 0- a 5 represents the coefficients corresponding to the normal tissue spectrum, DNA spectrum, lipid spectrum, histone spectrum, collagen spectrum, and actin spectrum, respectively. a 0- a The value of 5 satisfies the following formula:

[0044] S311. Random noise is superimposed on the simulated tissue spectrum S using the following formula to obtain the generated spectrum. S m :

[0045]

[0046] In the above formula, n It is random noise;

[0047] S312. A model is constructed using the convolutional neural network AlexNet to generate the spectrum. S m The input data for the convolutional neural network model is the input data, and the output data is the coefficients of each component in the spectrum.

[0048] Preferably, in S32, the fused neural network model is based on the Boosting model to fuse the convolutional neural network and the long short-term memory network, and the output F of the fused neural network model is obtained by the following formula:

[0049]

[0050] Among them, O CNN and O LSTM These are the outputs of the CNN model and the LSTM model, respectively. Indicates the corrected linear unit, and .

[0051] A Raman spectroscopy detection system, applied in a method for using ICCD and fiber-optic gated Raman spectroscopy systems, includes:

[0052] Laser source used to generate Raman signals that excite tissue;

[0053] ICCD detectors used for rapid acquisition of Raman spectral signals;

[0054] Control unit that communicates with the laser source and ICCD detector;

[0055] A spectral analysis module that provides spectral analysis results to the control unit;

[0056] The laser source and ICCD detector are respectively bundled together to the fiber optic probe via corresponding optical fibers;

[0057] The end of the fiber optic probe is designed to be flexible.

[0058] The present invention has at least the following beneficial effects:

[0059] Firstly, the application method of this invention provides a non-invasive tumor detection method that combines ICCD and gating technology to enhance Raman signal intensity and reduce fluorescence background, thereby improving detection sensitivity and efficiency.

[0060] Secondly, this invention provides a set of deep learning-based spectral analysis algorithms to achieve rapid and accurate analysis results of tumors and their malignancy under non-invasive conditions in vivo, which can serve as an auxiliary means to help doctors diagnose tumors.

[0061] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0062] Figure 1 This is a block diagram illustrating the principle of the Raman spectroscopy detection system based on ICCD and fiber optic gating of the present invention.

[0063] Figure 2 This is a comparison diagram of tissue spectra acquired using the Raman spectroscopy detection system of this invention and a conventional fiber Raman system;

[0064] Figure 3 A flowchart illustrating the application method of the fiber-optic gated Raman spectrometer based on ICCD according to the present invention;

[0065] Figure 4 This is a schematic diagram of the deep learning model used in the spectral analysis of this invention;

[0066] Figure 5 In the process of applying the deep learning model of this invention, standard Raman spectra of five basic components in the tissue are obtained;

[0067] Figure 6 This is a schematic diagram comparing the Raman spectra of malignant gastric tumor tissue and normal gastric body tissue acquired using the Raman spectroscopy detection system of the present invention;

[0068] Figure 7This is a schematic diagram showing the component analysis results of malignant tumor tissue and normal gastric body tissue using the algorithm of this invention. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0070] This patent aims to develop an ICCD (Intensified Charge-Coupled Device) and fiber-optic gated Raman spectrometer. First, the ICCD is applied to the fiber-optic Raman spectrometer. By utilizing the high sensitivity and fast readout capability of the ICCD, the acquisition efficiency and quality of the Raman signal are improved. At the same time, by combining the ICCD and gated technology, the Raman signal intensity is increased and the fluorescence background is reduced, thereby improving the detection sensitivity and efficiency.

[0071] Example 1

[0072] like Figure 1 As shown, the Raman spectrometer of the present invention mainly includes:

[0073] (1) Fiber optic probe 1: Composed of a bundle of optical fibers, used to transmit the excitation laser and the Raman spectral signal from the target tissue to the spectrometer. The end of the fiber optic probe is designed to be flexible so as to adapt to different in vivo detection paths.

[0074] (2) Laser source 2: The laser used to generate Raman signals to excite tissue is a high-repetition-rate pulsed laser source with a wavelength of 785 nm, a maximum power of 300 mW, a pulse width of 5 ns, and a repetition rate of 1 MHz.

[0075] (3) ICCD Detector 3: A high-sensitivity ICCD detector is used for rapid acquisition of Raman spectral signals. It should be noted that ICCD technology, or Enhancement Charge-Coupled Device technology, is a high-sensitivity photodetector technology. It enhances the signal by adding a gain layer to the traditional CCD sensor. ICCD has the characteristics of high sensitivity, fast readout capability, low noise performance, high dynamic range, and good linear response. This scheme mainly utilizes the fast electronic gating capability of the ICCD detector, which can collect a large number of photons in a short time, thereby improving the signal-to-noise ratio.

[0076] (4) Gating Unit 4: It should be noted that gating technology is a gated detection technology implemented on a time scale. By matching and synchronizing with the laser in time, it can detect effective Raman signals and compress background noise signals. This scheme mainly uses gating technology to control the exposure time of the detector, which can distinguish between Raman signals and fluorescence signals in terms of time resolution, so as to suppress fluorescence background. The shutter speed is 500 ps and the gate width is 5 ns. That is, this invention opens the camera shutter every time a Raman signal appears through gating technology, and sets the shutter width to allow only the Raman signal to enter the detector. By gating, strong background signals are suppressed and the signal-to-noise ratio of the Raman signal is improved, thereby realizing Raman testing in a strong background.

[0077] (5) Spectral Analysis Module 5: Built-in deep learning-based algorithms for processing and analyzing acquired Raman spectral data. This module includes a high-performance computing unit for running deep learning models.

[0078] (6) Control Unit 6: Used to control the operation of the entire system, including the position adjustment of the fiber optic probe, the working mode of the ICCD detector and the gating unit, and the data processing of the spectral analysis module.

[0079] It should be noted that since the generation and decay rates of Raman signals are much faster than those of fluorescence signals, precise control of the gating time can effectively suppress fluorescence background and improve the quality of the Raman spectrum. Therefore, this scheme, through the combination of ICCD and gating technology, can not only enhance the Raman signal but also reduce fluorescence background, thereby significantly improving detection sensitivity.

[0080] Furthermore, the rapid readout capability of ICCD also helps to improve the acquisition speed of Raman spectra, further enhancing detection efficiency. The design of fiber optic Raman spectrometers allows for convenient insertion into in vivo tissues for detection, while the application of ICCD and gating technology ensures the acquisition of high-quality Raman spectral data even in the in vivo environment, enabling real-time, non-invasive in vivo detection and greatly promoting the clinical application of Raman spectrometers.

[0081] like Figure 2 As shown, a comparison was made between the Raman spectra of tissues acquired using the Raman spectrometer of this invention and those acquired using a conventional non-ICCD gated system. It can be seen that the fluorescence background in the Raman spectra acquired by this invention is significantly suppressed, and the tissue characteristic peaks are more obvious.

[0082] Furthermore, this invention provides a deep learning-based Raman spectroscopy application method to achieve rapid and accurate analysis of tumors and their malignancy under non-invasive in vivo conditions, providing assistance to doctors in diagnosis.

[0083] Example 2

[0084] like Figure 3 As shown, the Raman spectrometer application method of the present invention mainly includes the following steps:

[0085] S1. Preparation and Start-up Stage: Place the fiber optic probe directly or through a visual imaging guidance system into the part to be detected, ensuring that the end of the fiber optic probe is close to the target tissue (distance less than 1 cm).

[0086] S2. Raman Spectroscopy Acquisition: The gating unit and ICCD detector are activated, and the Raman spectral signal of the target tissue is acquired through the fiber optic probe. The high-speed readout capability of the ICCD detector greatly shortens the signal acquisition time. However, in practical applications, Raman spectroscopy acquisition here refers to randomly selecting 10 points within a certain area of ​​the tissue to acquire Raman spectra, with each point acquiring for 10 seconds.

[0087] S3. Spectral data processing and analysis after signal transmission: The acquired Raman spectral signal is transmitted to the spectral analysis module through the fiber optic probe, ICCD detector, and control unit.

[0088] After receiving the signal, the spectral analysis module processes the spectral data using a built-in deep learning algorithm. The trained deep learning model can identify specific Raman spectral features of tumor cells, assess their malignancy, and provide component analysis results.

[0089] It should be noted that the deep learning spectral analysis algorithm used in S3 needs to be adapted to the characteristics of Raman spectra acquired by in vivo tissues and ICCD and gating systems, and has the ability to be highly adaptable, highly accurate and highly efficient.

[0090] Specifically, such as Figure 3 As shown, the processing methods of deep learning models include specific construction and implementation methods:

[0091] S31. Data Preprocessing Module: This module performs preprocessing operations on the acquired Raman spectra, including noise reduction, baseline correction, normalization, and spectral distortion correction. Since Raman spectra acquired by ICCD and gating systems may contain noise, noise reduction is first performed using the Savitzky-Golay filtering algorithm. Secondly, Raman spectra are often affected by environmental factors, instrument conditions, and sample preparation, leading to baseline drift. Baseline correction is performed using methods such as polynomial fitting and iterative least squares to eliminate this effect. Simultaneously, because spectral intensities may differ between different tissues and under different acquisition conditions, normalization is required to eliminate the impact of these differences on model training. Finally, the ICCD itself may cause distortion in the acquired spectra; wavelength calibration techniques are needed to reduce the impact of distortion on the analysis results.

[0092] S32. Feature Extraction Module: Extracts effective feature information from Raman spectra from multiple dimensions. Combining expert experience and theoretical knowledge, it selects characteristic spectral peak fragments from tissue Raman spectra that are biologically significant for tumor diagnosis, as shown in the table below:

[0093]

[0094] Then, a peak feature extraction algorithm based on continuous wavelet transform is used to accurately extract key information such as peak position, peak intensity, and peak width of the spectral peaks in these spectral segments. After this processing, the originally complex spectral data with thousands of channels is simplified to peak feature data with only dozens of channels.

[0095]

[0096] Among them, X s For the raw spectral data, X f For the selected spectral fragments, PFE is the peak feature extraction algorithm, X c The extracted spectral peak feature data.

[0097] S33, Tissue Component Analysis Module: Constructs a neural network model to learn the standard Raman spectra of five basic components: DNA, fat, histones, collagen, and actin. Figure 5 As shown, the trained model can analyze the contribution of these spectra in the tissue spectrum, and thus give the proportion of these components in the tissue, providing support for more in-depth pathological analysis.

[0098] The specific construction process of this model is as follows:

[0099] Tissue component spectra and a small amount of normal tissue spectra were collected to generate model training data;

[0100] The simulated tissue spectrum is obtained by multiplying the spectra of each tissue component by a certain coefficient compared to the spectrum of normal tissue.

[0101]

[0102] in, S 0 S 1- S 5 represent the spectra of DNA, lipids (trioleylglycerol), histones, collagen, and actin, respectively. All spectra were normalized. The six coefficients satisfy the following formula:

[0103]

[0104] To increase the robustness of the model, a certain amount of random noise is superimposed on the generated spectrum:

[0105]

[0106] Based on existing research, the proportion of DNA and protein increases and the proportion of lipid decreases in malignant tumors. Therefore, the coefficient ranges for each component are set as follows: DNA: 0~0.05, lipid: 0.05~0.8, collagen: 0.05~0.4, actin: 0.01~0.4, and the remainder is histone.

[0107] The model is constructed using a convolutional neural network, AlexNet. During training, the input data is the generated spectrum, and the output data is the coefficients of each component in the spectrum. During training, the neural network learns the correlation between the presented tissue spectrum and the coefficients of each component in the spectrum. For a trained model, given a tissue spectrum, it can analyze and output the proportional coefficients of each tissue component reflected in that spectrum.

[0108] S34, Model Training Module: Based on the characteristics of Raman spectroscopy data, a neural network model fusing Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) is constructed. The CNN is used to implement the tissue composition analysis function described in S43; it can automatically learn deep features of Raman spectroscopy, which are more effective in distinguishing different tissues and tumor types. The LSTM is used to establish the interrelationships between the features extracted in S42, improving the model's recognition accuracy and robustness. Boosting model fusion technology is used to combine the prediction results of the two models to improve the final prediction accuracy.

[0109]

[0110] Where F is the final output of the dual-model fusion algorithm, and O CNN and O LSTM W1 and W2 are the outputs of the CNN and LSTM models, respectively, and the weights of the two models are W1 and W2, respectively. These two weights are trainable parameters, determined by the neural network model through the evaluation and optimization module described in S45.

[0111] S35, Model Evaluation and Optimization Module: This module uses cross-validation to evaluate the model's performance, ensuring good performance across datasets of different organs and tissues. Hyperparameters are adjusted using methods such as grid search, random search, or Bayesian optimization to further improve model performance.

[0112] S36. For applications requiring real-time monitoring at the surgical site, the algorithm must possess real-time prediction capabilities. To achieve real-time prediction, the trained model needs to be compressed to reduce its computational complexity and memory usage. The model is then deployed on a Raman spectrometer to enable real-time analysis and feedback of the acquired Raman spectral data.

[0113] S4. Result Output: The control unit displays the processing results, allowing doctors to quickly determine whether the target tissue is a tumor and its malignancy based on the displayed Raman spectrum and the analysis results of the deep learning algorithm. Doctors then make a final diagnostic decision based on the system's results and clinical information.

[0114] This embodiment 2 develops a matching deep learning algorithm tailored to the characteristics of in vivo tissues and the spectra acquired by ICCD-gated Raman spectrometers, making the identification of in vivo tumors and the assessment of their malignancy more accurate and faster. In addition, the use of a fiber optic flexible probe design reduces the invasiveness and trauma to patients.

[0115] Verification example:

[0116] The Raman spectrometer based on ICCD and fiber optic gating of this invention was used to perform Raman detection and classification of gastric tissue biopsy samples. The specific results are as follows:

[0117] like Figure 6 As shown, Raman spectroscopy detection of malignant gastric tumor tissue and normal gastric body tissue using the present invention can be seen that there are distinguishable differences in the Raman spectra collected by the present invention, that is, the adaptability meets the detection requirements.

[0118] like Figure 7 As shown, the algorithm model of this invention can determine whether a tissue is a tumor, and can also provide tissue composition analysis results. The results show that tumor tissue has higher DNA and protein content and lower fat content, which is consistent with the conclusions of existing research, and can fully verify the effectiveness of this invention.

[0119] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0120] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A Raman spectroscopy system based on ICCD and fiber-optic gating, characterized in that, include: Laser source used to generate Raman signals that excite tissue; ICCD detectors used for rapid acquisition of Raman spectral signals; Control unit that communicates with the laser source and ICCD detector; A spectral analysis module that provides spectral analysis results to the control unit; The laser source and ICCD detector are respectively bundled together to the fiber optic probe via corresponding optical fibers; The end of the fiber optic probe is designed to be flexible. The application methods of the Raman spectroscopy system include: S1. The control unit switches the working state of the laser source to generate a laser that excites the target tissue to produce a Raman signal. S2. The laser reaches the part to be detected through a flexible fiber optic probe. The control unit switches the working state of the gating unit and the ICCD detector to acquire the Raman spectrum signal of the target tissue through the fiber optic probe. S3. The control unit outputs the acquired Raman spectral signal to the spectral analysis module. The spectral analysis module processes the acquired Raman spectral signal based on the built-in deep learning model to analyze and evaluate the malignancy of the acquired tissue. S4. The control unit outputs and / or displays the processing results of the spectral analysis module; In S3, the methods for obtaining the deep learning model include: S30. The acquired Raman spectral signals are preprocessed and features are extracted to obtain the acquired tissue component spectra; S31. The standard Raman spectra of basic components are learned through a neural network model, and the collected tissue component spectra and normal tissue spectra are used as training data for the neural network model. The contribution of basic components in the collected tissue composition spectrum is analyzed using a trained neural network model to provide the proportional relationship of basic components in the tissue. S32. Based on the characteristics of Raman spectroscopy data, a fusion neural network model based on convolutional neural network and long short-term memory network is constructed and trained. The performance of the neural network model is evaluated by cross-validation method. Then, the weights W1 and W2 of the convolutional neural network and long short-term memory network are assigned values ​​to obtain the trained deep learning model. The basic components include DNA, fat, histones, collagen, and actin. In S3, the trained deep learning model is compressed and deployed on the spectral analysis module to enable real-time analysis and feedback of the collected Raman spectral data. In S31, the neural network model is trained to determine the proportional relationships of basic components in the tissue as follows: S310. Obtain the simulated tissue spectrum S using the following formula: In the above formula, S 0 represents the spectrum of normal tissue. S 1- S 5 represent the spectra of DNA, lipids, histones, collagen, and actin, respectively. a 0- a 5 represents the coefficients corresponding to the normal tissue spectrum, DNA spectrum, lipid spectrum, histone spectrum, collagen spectrum, and actin spectrum, respectively. a 0- a The value of 5 satisfies the following formula: S311. Random noise is superimposed on the simulated tissue spectrum S using the following formula to obtain the generated spectrum. S m : In the above formula, n It is random noise; S312. A model is constructed using the convolutional neural network AlexNet to generate the spectrum. S m The input data for the convolutional neural network model is the input data, and the output data is the coefficients of each component in the spectrum. In S32, the fused neural network model is based on the Boosting model to fuse the convolutional neural network and the long short-term memory network, and the output F of the fused neural network model is obtained by the following formula: Among them, O CNN and O LSTM These are the outputs of the CNN model and the LSTM model, respectively. Indicates the corrected linear unit, and .

2. The Raman spectroscopy system based on ICCD and fiber-optic gating as described in claim 1, characterized in that, In S2, the method for acquiring the Raman spectral signal of the target tissue is as follows: within a predetermined area of ​​the tissue to be acquired, 10 points are randomly selected to acquire Raman spectra, and the acquisition time for each point is at least 10 seconds.

3. The Raman spectroscopy system based on ICCD and fiber-optic gating as described in claim 1, characterized in that, In S2, the gating unit is used to control the exposure time of the ICCD to ensure that the shutter speed of the ICCD is 500 ps and the gate width is 5 ns.

4. The Raman spectroscopy system based on ICCD and fiber-optic gating as described in claim 1, characterized in that, In S30, the preprocessing includes: noise reduction, baseline correction, normalization, and spectral distortion correction operations on the acquired Raman spectra; The feature extraction method includes: S301. Extract effective feature information from multiple dimensions in the preprocessed Raman spectrum signal image; S302. The spectral peak feature extraction algorithm based on continuous wavelet transform extracts key information that is biologically significant for tumor diagnosis from effective feature information, and then obtains spectral peak feature data. The effective feature information refers to feature peak fragments, which are used to characterize the attribution of biologically significant peaks for tumor diagnosis. The key information includes the peak position, peak intensity, and peak width of the spectral peaks in the characteristic spectral peak segment.

5. The Raman spectroscopy system based on ICCD and fiber-optic gating as described in claim 1, characterized in that, In S301, the attribution of biologically significant spectral peaks for tumor diagnosis is characterized by the following characteristic peak fragments: Through 616-626cm -1 Characteristic peak fragments characterize amide IV; Passing through 637-647cm -1 Characteristic spectral peak fragments characterize CS stretching vibrations; Passing 785-795cm -1 Characteristic peak fragments characterize DNA; Through 855-865cm -1 Characteristic peak fragments characterize lipids; Passing 998-1008cm -1 Characteristic peak fragments characterize phenylalanine; Through 1075-1085cm -1 Characteristic peak fragments characterize DNA; Passing through 1235-1245cm -1 Characteristic peak fragments characterize amide III; Through 1318-1328cm -1 Characteristic spectral peak fragments characterize CH deformation vibrations; Through 1444-1454cm -1 Characteristic spectral peak fragments characterize CH2 bending vibrations; Through 1547-1557cm -1 Characteristic peak fragments characterize amide II; Through 1611-1621cm -1 Characteristic peak fragments characterize tryptophan; Through 1664-1674cm -1 Characteristic peak fragments are used to characterize amine I.

Images