A raman spectrum consensus sampling method for biological tissue section classification

Abstract

The application discloses a Raman spectrum consensus sampling method for biological tissue section classification. The method comprises the following steps: collecting Raman spectrum of a tissue section sample, and then performing pretreatment on the Raman spectrum; then, a binary classification model is established and analyzed on the tissue section sample to determine whether the tissue section is a metastatic tumor; secondly, after confirming that the obtained tissue section sample is a metastatic tumor tissue section, a multi-classification model is trained and predicted; finally, the stability of the prediction model and the minimum sampling number are verified by using a consensus sampling method, multiple random sampling Raman spectrum data from the same sample are used, a single sample is independently predicted by using a support vector machine model, and finally, based on the consensus sampling concept, the prediction results of multiple spectra are integrated to obtain the final result of the tissue sample classification.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of tumor recognition, and particularly relates to a Raman spectrum consensus sampling method for biological tissue section classification. BACKGROUND

[0002] Cancer of Unknown Primary (CUP) is a metastatic tumor with an indeterminate primary site. Although CUP accounts for only 3-5% of all cancers worldwide, it is still one of the six major causes of cancer-related deaths. The classification confirmation of CUP helps to guide the selection of treatment methods for patients, thereby significantly affecting the prognosis and survival rate of patients.

[0003] At present, morphological, histopathological and emerging molecular methods are still commonly used methods for diagnosing metastatic tumor classification. However, existing studies have found that morphological differences have no direct relationship with tumor metastasis and metastatic tumor classification; histopathological methods have factors such as limited availability of tissues, limited available immunohistochemical antibodies and subjective cognitive differences of doctors, which cannot produce accurate diagnostic results; emerging molecular methods such as DNA / RNA sequencing have strict requirements for the acquisition and storage of biological samples, high sequencing cost, and low penetration rate in second- and third-tier cities.

[0004] Chinese patent CN117116469A discloses a deep learning-based oral cancer Raman spectrum multi-classification method and system. The oral cancer Raman spectrum multi-classification method uses a multi-scale feature fusion deep residual neural network model to classify the collected oral Raman spectrum data. The multi-scale feature fusion deep residual neural network model includes a data mapping layer, a main feature network, a feature fusion network and a classification network connected in turn. The data mapping layer maps the oral Raman spectrum data into mapped data. The main feature network extracts features from the mapped data to obtain an output feature map. The feature fusion network fuses the output feature map to obtain a fused output feature map. The classification network extracts features from the fused output feature map and calculates to obtain an oral cancer multi-classification result. However, the design does not fully consider the actual situation of tumor heterogeneity and metastatic tumor redifferentiation. Moreover, after obtaining the classification result using the prior classification method, the method lacks a scientific model robustness evaluation method, and cannot feedback the required minimum spectrum acquisition quantity based on the model classification effect, but only consumes excessive sampling time and computer resources during spectrum acquisition.

[0005] Therefore, there is an urgent need to design a method based on unlabeled Raman spectrum to realize rapid, robust and accurate metastatic tumor metastatic tumor tissue section classification. SUMMARY

[0006] In order to solve the above problems, the application designs a Raman spectrum consensus sampling method for biological tissue section classification. Multiple random sampling Raman spectrum data from the same sample are used to independently predict a single sample using a support vector machine model. Based on the consensus sampling concept, the prediction results of multiple spectra are integrated to obtain the final result of tissue sample classification.

[0007] A Raman spectrum consensus sampling method for biological tissue section classification, comprising the following steps:

[0008] Step S1, collecting Raman spectrum of tissue section sample;

[0009] Step S2, pre-processing Raman spectrum;

[0010] Step S3, establishing and analyzing a two-classification model for the tissue section sample to determine whether the tissue section is a metastatic tumor;

[0011] Step S4, after confirming that the obtained tissue section sample is a metastatic tumor tissue section, training and predicting a multi-classification model;

[0012] Step S5, using a consensus sampling method to verify the stability and minimum sampling number of the prediction model.

[0013] Preferably, in step S1, collecting Raman spectrum specifically comprises the following steps:

[0014] Step S11, observing the macroscopic morphological characteristics of the tissue section under a microscope of a confocal Raman spectrometer at a low magnification;

[0015] Step S12, for each tissue section sample, three different regions of interest are individually selected, without obvious cell necrosis in the region, and the selected regions are distributed as much as possible in different positions of the tissue section, and the area of each region is 2*2mm 2 ;

[0016] Step S13, in each region, more than 160 Raman spectra are collected, and the step length during spectrum collection is 50μm.

[0017] Preferably, in step S2, pre-processing Raman spectrum comprises the following steps:

[0018] Step S21, determining the cutoff point of Raman shift using Raman spectrum, and using the spectrum of 786-1740cm-1 fingerprint region for analysis;

[0019] Step S22, using Z-score outlier detection method to identify and delete cosmic ray spikes in Raman spectrum data;

[0020] Step S23, baseline correction is performed using adaptive iteratively reweighted penalized least squares method;

[0021] Step S24, normalization and smoothing of the spectrum are performed.

[0022] Preferably, the step S3 of establishing and analyzing the tissue slice sample by the binary classification model to determine whether the tissue slice is a metastatic tumor comprises the following steps:

[0023] Step S31, the Raman spectra of multiple in situ tumor tissue slices and the Raman spectra of multiple metastatic tumor tissue slices are randomly divided into a training set and a test set according to a ratio of 7:3; all spectra collected from the same patient sample are regarded as a whole during classification, and the training set and the test set are divided;

[0024] Step S32, hyperparameter optimization of the support vector machine model is performed using a grid search method; wherein the input data of the support vector machine model is the Raman spectrum data of the tumor tissue, and the prediction output of the support vector machine model is a binary classification label for the tumor tissue, i.e., "in situ tumor" or "metastatic tumor";

[0025] Step S33, cross-validation of the support vector machine model is performed using a leave-one-out method to evaluate the classification accuracy of the support vector machine model;

[0026] In each round, the sample collected from one patient is taken as the test set, and the samples collected from other patients are taken as the training set; during cross-validation, the classification result of the test set is taken as the basis for judging the model, and it is determined whether the tumor located in the liver is an in situ tumor or a metastatic tumor; if it is an in situ tumor, the spectrum prediction process is terminated at this step; if it is a metastatic tumor, the multi-classification model is inputted for the next step of prediction.

[0027] Preferably, the step S4 of training and predicting the multi-classification model after confirming that the obtained tissue slice sample is a metastatic tumor tissue slice comprises the following steps:

[0028] Step S41, the Raman spectra of multiple tumor tissue samples from the abdominal cavity are randomly divided into a training set and a test set according to a ratio of 7:3;

[0029] Step S42, hyperparameter optimization of the support vector machine model is performed using a grid search method; wherein the input data of the support vector machine model is the Raman spectrum data of the metastatic tumor tissue, and the prediction output of the model is a determination of the primary source location of the metastatic tissue, and the result is a multi-classification label, such as "liver", "colorectal", "lung", and other organs;

[0030] Step S43, cross-validation of the model is then performed using a leave-one-out method to evaluate the classification accuracy of the model;

[0031] Step S44, in the verification method described in step S43, the above operation is repeated for each tissue slice, that is, under the total sample number of the tissue slice, the test accuracy obtained when the target sample is taken as the test set one by one; the set of test accuracies is used as the basis for evaluating the classification accuracy of the model;

[0032] Step S45, the Raman spectrum data of the metastatic tumor tissue slice is introduced one by one, and the multi-classification model trained in the above description is applied to classify and predict the Raman spectrum of the metastatic tumor, and the primary source position of the metastatic tissue is determined.

[0033] Preferably, in step S5, the stability and minimum sampling number of the prediction model are verified using the consensus sampling method, including the following steps:

[0034] Step S51, based on the consensus sampling principle, the complete Raman spectrum set of each metastatic tumor patient is regarded as a unified entity, the original data set is resampled, and a plurality of virtual data sets are generated; by adjusting the sampling ratio, the Raman spectrum samples in the set are extracted as a new data set;

[0035] Step S52, 150 sampling iterations are performed for each sampling ratio;

[0036] Step S53, as the sampling ratio decreases, the prediction accuracy gradually starts to fluctuate, and the variance becomes larger, the variance of the 150 sampling prediction results of each sampling ratio is calculated, the step-down graph of each metastatic tumor is obtained, and the inflection point position is quantified as the sampling ratio; through the sampling ratio of the inflection point position, the minimum sampling number of the Raman spectrum that can maintain the robustness of a single metastatic tumor sample under the premise of using the prediction result of the classification model can be determined.

[0037] Preferably, in step S32, the hyperparameter optimization of the support vector machine model using the grid search method includes:

[0038] Step S321, use "tuned_parameters" to define the hyperparameter network to be adjusted, and these parameters include the kernel function type, the hyperparameter gamma of the kernel function, and the penalty coefficient C;

[0039] Step S322, according to the listed multiple possible hyperparameters gamma and penalty coefficient C, the performance of each set of parameter combination is evaluated by cross-validation using the Grid SearchCV function; cross-validation divides the dataset into 5 parts according to the description in step S31, according to the source of the tissue section of the spectrum, and uses 4 parts for training and 1 part for validation in turn, to ensure the adaptability and generalization ability of the model to the data; in the optimization process, the Grid SearchCV function searches for the best parameter combination in the parameter grid according to the cross-validation result, and uses this best parameter as the parameter used by the next classification model.

[0040] Preferably, in step S51, the adjustment method of the sampling ratio is: from 0.9 to 0.1, decreasing by 0.1 each time, and then from 0.09 to 0.01, decreasing by 0.01 each time.

[0041] The advantages and effects of the present application are as follows:

[0042] The present application designs a Raman spectrum consensus sampling method for biological tissue section classification. The Raman spectrum of the tissue section sample is collected, and then the Raman spectrum is pretreated. Then, a binary classification model is established and analyzed for the tissue section sample to determine whether the tissue section is a metastatic tumor. Next, after confirming that the obtained tissue section sample is a metastatic tumor tissue section, a multi-classification model is trained and predicted. Finally, the stability of the prediction model and the minimum sampling number method are verified using the consensus sampling method. Multiple random sampling Raman spectrum data from the same sample are used to independently predict the single sample using a support vector machine model, and based on the consensus sampling concept, the prediction results of multiple spectra are integrated to obtain the final result of the tissue sample classification.

[0043] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the contents of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following will be described in detail with the preferred embodiments of the present application and with the help of the accompanying drawings.

[0044] According to the detailed description of the specific embodiments of the present application in the following text combined with the accompanying drawings, those skilled in the art will be more clear about the above and other purposes, advantages and features of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative effort. In all the drawings, similar elements or parts are generally identified by similar reference signs. In the drawings, the elements or parts are not necessarily drawn according to the actual proportions.

[0046] Figure 1 A flow chart of a Raman spectrum consensus sampling method for biological tissue section classification designed by the present application. DETAILED DESCRIPTION

[0047] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. In the following description, specific details such as specific configurations and components are provided only to help the overall understanding of the embodiments of the present application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, in order to be clear and concise, the description of known functions and structures is omitted in the embodiments.

[0048] It should be understood that the "one embodiment" or "the embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "one embodiment" or "the embodiment" appearing throughout the specification does not necessarily mean the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner.

[0049] In addition, the reference numerals and / or letters can be repeated in different examples in the present application. Such repetition is for the purpose of simplification and clarity, and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0050] The term "and / or" herein is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, B alone and A and B together. The term "and" herein is a description of another association relationship of the associated objects, which means that there can be two relationships, for example, A and B, which means that there are two cases of A alone and A and B together. In addition, the character " / " herein generally means that the associated objects before and after are in an "or" relationship.

[0051] The term "at least one", as used herein, is merely descriptive Borel's relationship between the associated objects, indicating that there may be three relationships, for example, at least one of A and B, which can represent: A exists alone, A and B exist simultaneously, and B exists alone.

[0052] It should also be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion.

[0053] Embodiment 1

[0054] This embodiment mainly introduces a Raman spectrum consensus sampling method for biological tissue section classification, which includes the following steps:

[0055] Step S1, collect the Raman spectrum of the tissue section sample; Specifically, the following steps are included:

[0056] Step S11, under the microscope of the confocal Raman spectrometer, observe the macroscopic morphological characteristics of the tissue section at low magnification;

[0057] Step S12, for each tissue section sample, three different regions of interest are selected individually, which require no obvious cell necrosis in the region, and the selected regions are distributed as much as possible in different positions of the tissue section, and the area of each region is 2x2mm 2 ;

[0058] Step S13, in each region, collect more than 160 Raman spectra, and the step length when collecting the spectrum is 50μm.

[0059] Further, the magnification is 5x.

[0060] Further, the macroscopic morphological characteristics include size, shape, boundary, whether there is a necrotic area caused by sample preparation problems, etc.

[0061] Step S2, pretreat the Raman spectrum; Specifically, the following steps are included:

[0062] Step S21, determine the cutoff point of the Raman shift of the Raman spectrum, analyze the spectrum in the fingerprint region using 786-1740cm -1 ;

[0063] Step S22, identify and delete cosmic ray spikes in the Raman spectrum data using Z-score outlier detection method;

[0064] Step S23, baseline correction is performed using adaptive iteratively reweighted penalized least squares (airPLS);

[0065] Step S24, normalization and smoothing of the spectra are performed for further analysis.

[0066] Step S3, establishment and analysis of a binary classification model for the tissue slice sample to determine whether the tissue slice is a metastatic tumor; specifically comprising the following steps:

[0067] Step S31, Raman spectra of multiple liver in situ tumor tissue slices and Raman spectra of multiple liver metastasis tissue slices are randomly divided into a training set and a test set according to a ratio of 7:3. When classifying, all spectra collected from the same patient sample are considered as a whole, and the training and test sets are divided. That is, assuming that the total number of tissue slices is 100, the Raman spectra from 70 tissue slices are divided into a training set, and the Raman spectra from 30 tissue slices are divided into a test set;

[0068] Step S32, hyperparameter optimization of the support vector machine model is performed using a grid search method. First, "tuned_parameters" is used to define the hyperparameter network to be adjusted, including the kernel function type (kernel), the hyperparameter gamma of the kernel function, and the penalty coefficient C. A plurality of possible gamma and C values are listed here. Then, "GridSearchCV" is used to evaluate the performance of each parameter combination through cross-validation (cv=5). Cross-validation divides the data set into 5 parts according to the description in step S31, according to the source of the tissue slices of the spectra, and uses 4 parts for training and 1 part for validation in turn, to ensure the adaptability and generalization ability of the model. During the optimization process, "GridSearchCV" searches for the best parameter combination in the parameter grid according to the cross-validation results, and uses this best parameter as the parameter used by the classification model in the next step.

[0069] Step S33, model cross-validation is performed using the leave-one-out method to evaluate the classification accuracy of the support vector machine model.

[0070] That is, assuming that the total number of tissue slices is 100, the Raman spectra from 99 tissue slices are divided into a training set, and the Raman spectra from 1 tissue slice are divided into a test set. When cross-validating, the classification result of the test set is used as the basis for judging the pros and cons of the model, and it is determined whether the tumor located in the liver is an in situ tumor or a metastatic tumor. If it is an in situ tumor, the spectrum prediction process terminates at this step, and if it is a metastatic tumor, it is input into the multi-classification model for the next step of prediction.

[0071] The support vector machine model described in this step has tumor tissue Raman spectrum data as input data, and the model prediction output is the binary classification label judgment of the tumor tissue: "in situ tumor" or "metastatic tumor".

[0072] Step S4, after confirming that the obtained tissue section sample is a metastatic tumor tissue section, the training and prediction of the multi-classification model are performed; specifically including the following steps:

[0073] Step S41, the Raman spectra of multiple tumor tissue samples from the abdominal cavity are randomly divided into a training set and a test set according to a ratio of 7:3; that is, assuming that the total number of tissue sections is 100, the Raman spectra from 70 of the tissue sections are divided into a training set, and the Raman spectra from 30 of the tissue sections are divided into a test set;

[0074] Step S42, hyperparameter optimization of the support vector machine model is performed using the grid search method; wherein the support vector machine model has metastatic tumor tissue Raman spectrum data as input data, and the model prediction output is the judgment of the primary source location of the metastatic tissue, with the result being a multi-classification label: such as "liver", "colorectal", "lung", and other organs; the grid search method is the same as in step S32, that is, first, "tuned_parameters" is used to define the hyperparameter network to be adjusted, which includes the kernel function type (kernel), the hyperparameter gamma of the kernel function, and the penalty coefficient C. A plurality of possible gamma and C values are listed here. Then, "GridSearchCV" is used to evaluate the performance of each parameter combination through cross-validation (cv=5). Cross-validation divides the data set into 5 parts according to the tissue section source of the spectrum as described in 3.1.1, and uses 4 parts for training and 1 part for validation in rotation to ensure the model's adaptability and generalization ability. During the optimization process, "GridSearchCV" searches for the best parameter combination in the parameter grid according to the cross-validation results, and uses this best parameter as the parameter for the next classification model.

[0075] Step S43, then the leave-one-out method is used for model cross-validation to evaluate the classification accuracy of the model; that is, assuming that the total number of tissue sections is 100, the Raman spectra from 99 of the tissue sections are divided into a training set, and the Raman spectra from 1 of the tissue sections are divided into a test set;

[0076] Step S44, in the verification method described in step S43, the above operations are repeated for each tissue section, that is, the test accuracy obtained when the target sample is used as the test set under the total number of tissue section samples; the set of test accuracies is used as the basis for evaluating the classification accuracy of the model; in this embodiment, a total accuracy of 97.4% can be obtained.

[0077] Step S45, introduce the Raman spectrum data of the metastatic tumor tissue slice one by one, apply the multi-classification model trained in the above description to the Raman spectrum of the metastatic tumor for classification prediction, and collect more than 480 Raman spectra for each tumor metastasis tissue slice. In the above process, multiple Raman spectra from the same tissue slice are independently classified and predicted. For example, a total of 480 Raman spectra are collected for the first sample, of which 445 Raman spectra are classified as tumor type 1, 35 Raman spectra are classified as tumor type 2, and 1 Raman spectrum is classified as tumor type 3. According to the consensus sampling concept, the sample is finally correctly classified as tumor type 1. In the classification task, the more the number of correctly classified spectra, the higher the accuracy, and the stronger the classification ability of the model. The classification accuracy of a certain tissue sample is calculated according to the formula:

[0078]

[0079] The support vector machine model described in this step has input data of Raman spectrum data of metastatic tumor tissue, and the model prediction output is the determination of the primary source location of the metastatic tissue, and the result is a multi-classification label: such as "liver", "colorectal", "lung", etc.

[0080] Step S5, use the consensus sampling method to verify the stability and minimum sampling number of the prediction model. Specifically, the following steps are included:

[0081] Step S51, based on the consensus sampling principle, the complete Raman spectrum set of each metastatic tumor patient is regarded as a unified entity, the original data set is resampled with replacement to generate multiple virtual data sets; by adjusting the sampling ratio (from 0.9 to 0.1, decreasing by 0.1 each time, then from 0.09 to 0.01, decreasing by 0.01 each time), extract the Raman spectrum samples in the set as a new data set;

[0082] Step S52, for each sampling ratio, perform 150 sampling iterations;

[0083] Step S53, as the sampling ratio decreases, the prediction accuracy gradually starts to fluctuate, and the variance becomes larger. Calculate the variance of the 150 sampling prediction results of each sampling ratio to obtain the step-down graph of each metastatic tumor, and quantize the inflection point position as the sampling ratio. Through the sampling ratio of the inflection point position, the minimum sampling number of Raman spectrum that can maintain the robustness of a single metastatic tumor sample under the premise of using the prediction results of the classification model can be determined. In this embodiment, the minimum number of collected Raman spectra is 24.

[0084] Please refer to Figure 1 , Figure 1A flow chart of a Raman spectrum consensus sampling method for biological tissue section classification designed for the present application; wherein T1, T2, T3, T4, T5, X1, X2, X3 are the number of different tumor tissue samples collected (T1 represents sampling from No. 1 position in situ tumor, X1 represents sampling from No. 1 position unknown tumor, X1(M2) represents tumor transferred from No. 2 position to No. 1 position (actually sampled from No. 1 position in clinical), and subscript 1, 2, 3... represent different source positions of tumor types), N is the number of spectra collected for a single sample, and D is the data point dimension in a single spectrum. (The above figure takes the prediction process of tumor X1(M2) transferred from No. 2 position to No. 1 position as an example).

[0085] The Raman spectrum consensus sampling method for biological tissue section classification designed by the present application collects the Raman spectrum of the tissue section sample, and then pre-processes the Raman spectrum. Then, the establishment and analysis of the two-classification model of the tissue section sample are carried out to determine whether the tissue section is a metastatic tumor. Next, after confirming that the obtained tissue section sample is a metastatic tumor tissue section, the training and prediction of the multi-classification model are carried out. Finally, the stability of the prediction model and the minimum sampling number are verified by using the consensus sampling method. Multiple random sampling Raman spectrum data from the same sample are used to independently predict a single sample using a support vector machine model. Based on the consensus sampling concept, the final result of the tissue sample classification is obtained by integrating the prediction results of multiple spectra. The present application considers the tumor heterogeneity in the tissue sample and the redifferentiation when tumor metastasis occurs. The consensus sampling concept is used for the Raman spectrum collected in multiple regions to ensure the reliability and robustness of tumor classification. And through the variance sudden drop chart obtained by consensus sampling, the minimum number of spectra required for each sample is calculated, which speeds up the spectrum collection process and reduces the time required for tissue section sample classification.

[0086] The above only describes the preferred embodiments of the present application and does not limit the protection scope of the present application. For those skilled in the art, the present application can have various changes and variations. Any changes, modifications, replacements, integrations and parameter changes to these embodiments within the spirit and principles of the present application, which can realize the same functions without departing from the principles and spirit of the present application, fall within the protection scope of the present application.

Claims

1. A Raman spectroscopy consensus sampling method for biological tissue section classification, characterized in that: The following steps are involved: Step S1, collecting Raman spectra of tissue slice samples; Step S2, preprocessing the Raman spectrum; Step S3: establishing and analyzing a binary classification model for the tissue section sample to determine whether the tissue section is a metastatic tumor; Step S4: After confirming that the obtained tissue slice sample is a tissue slice of a metastatic tumor, training and prediction of a multi-classification model are performed; Step S5: using the consensus sampling method to verify the stability of the prediction model and the minimum number of samples; In step S3, establishing and analyzing a binary classification model for the tissue section sample to determine whether the tissue section is a metastatic tumor includes the following steps: Step S31: randomly dividing the Raman spectra of multiple in situ tumor tissue sections and the Raman spectra of multiple metastatic lesion tissue sections into a training set and a test set according to a ratio of 7:3; during classification, all spectra collected from the same patient sample are considered as a whole and the training and test sets are divided; Step S32: Optimizing hyperparameters of the support vector machine model using a grid search method; wherein the input data of the support vector machine model is the Raman spectral data of the tumor tissue, and the prediction output of the support vector machine model is a binary classification label determination of the tumor tissue: "in situ tumor" or "metastatic tumor"; Step S33: cross-validating the support vector machine model using the leave-one-out method to evaluate the classification accuracy of the support vector machine model; In each round, samples collected from one patient are used as the test set, and samples collected from other patients are used as the training set. During cross-validation, the classification results of the test set are used as the basis for judging the quality of the model. In addition, it is determined whether the tumor located on the liver is an in situ tumor or a metastatic tumor. If it is an in situ tumor, the spectral prediction process ends at this step. If it is a metastatic lesion, it is input into the multi-classification model for the next prediction.

2. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 1, characterized in that: In step S1, collecting the Raman spectrum specifically includes the following steps: Step S11, observing the macroscopic morphological features of the tissue section at low magnification under a microscope of a confocal Raman spectrometer; Step S12: For each tissue section sample, three different regions of interest are individually selected, requiring no obvious cell necrosis within the regions, and the selected regions are distributed as much as possible in different locations of the tissue section; Step S13: In each region, more than 160 Raman spectra are collected, and the step length of collecting the spectra is 50 μm.

3. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 1, characterized in that: In step S2, pre-processing the Raman spectrum includes the following steps: Step S21: determine the cutoff point of the Raman shift using the Raman spectrum, using 786~1740cm -1 The spectrum of the fingerprint area is analyzed; Step S22: using the Z-score outlier detection method to identify and delete cosmic ray peaks in the Raman spectrum data; Step S23: performing baseline correction using an adaptive iterative reweighted penalty least squares method; Step S24: normalize and smooth the spectrum.

4. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 1, characterized in that: In step S4, after confirming that the obtained tissue slice sample is a tissue slice of a metastatic tumor, training and prediction of a multi-classification model are performed, including the following steps: Step S41, randomly dividing the Raman spectra of various tumor tissue samples from the abdominal cavity into a training set and a test set according to a ratio of 7:3; Step S42: Optimizing hyperparameters of a support vector machine model using a grid search method; wherein the support vector machine model receives Raman spectral data of metastatic tumor tissue as input, and the model predicts and outputs a multi-classification label for the primary source location of the metastatic tissue: "liver," "colorectal," "lung," and other organs; Step S43: Then, a leave-one-out method is used to perform model cross-validation to evaluate the classification accuracy of the model; Step S44: Repeat steps S41 to S42 for each tissue slice in the verification method described in step S43, i.e., obtain the test accuracy obtained when the target samples are used as the test set one by one under the total number of tissue slice samples; the above set of test accuracies is used as the basis for evaluating the classification accuracy of the model; Step S45 , importing the Raman spectrum data of metastatic tumor tissue slices one by one, applying the trained multi-classification model to classify and predict the Raman spectrum of the metastatic tumor, and determining the primary source location of the metastatic tissue.

5. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 1, characterized in that: In step S5, the consensus sampling method is used to verify the stability of the prediction model and the minimum sampling number, including the following steps: Step S51: Based on the consensus sampling principle, the complete Raman spectrum set of each metastatic tumor patient is regarded as a unified entity, and the original dataset is resampled with replacement to generate multiple virtual datasets; by adjusting the sampling ratio, the Raman spectrum samples in the set are extracted as new datasets; Step S52: perform 150 sampling iterations for each sampling ratio; Step S53: As the sampling ratio decreases, the prediction accuracy gradually begins to fluctuate and the variance increases. The variance of the prediction results of 150 samplings for each sampling ratio is calculated to obtain a sudden drop-off diagram for each metastatic tumor, and the inflection point position is quantitatively identified as the sampling ratio. The sampling ratio of the inflection point position can be used to determine the minimum number of Raman spectra samples required to maintain the robustness of the prediction results for a single metastatic tumor sample using the multi-classification model.

6. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 1, characterized in that: In step S32, using the grid search method to optimize the hyperparameters of the support vector machine model includes: Step S321: Use "tuned_parameters" to define the hyperparameter network to be adjusted. These parameters include the kernel function type, the kernel function hyperparameter gamma, and the penalty coefficient C. Step S322: Based on the listed multiple possible hyperparameters gamma and penalty coefficients C, use the Grid SearchCV function to evaluate the performance of each set of parameter combinations through cross-validation; the cross-validation divides the data set into 5 parts according to the description in step S31 and the tissue section source of the spectrum, and uses 4 of them for training and 1 for validation in turn to ensure the model's adaptability and generalization ability to the data; during the optimization process, the Grid SearchCV function searches for the best parameter combination in the parameter grid based on the cross-validation results, and uses the best parameters as the parameters for the next classification model.

7. The Raman spectroscopy consensus sampling method for biological tissue section classification according to claim 5, characterized in that: In step S51 , the sampling ratio is adjusted from 0.9 to 0.1, decreasing by 0.1 each time, and then from 0.09 to 0.01, decreasing by 0.01 each time.

Images