Menstrual blood and peripheral blood classification identification model and construction method and application thereof

Abstract

The invention provides a menstrual blood and peripheral blood classification identification model and a construction method and application thereof, and relates to the field of surgical identification. According to the method, volatile organic compounds in menstrual blood and peripheral blood are detected based on an electronic nose method, and obtained response intensity data of the menstrual blood and the peripheral blood in an electronic nose sensor matrix are subjected to rank sum detection, principal coordinate analysis and inter-group similarity analysis; and then a classification identification model of the menstrual blood and the peripheral blood is established through a random forest model, so that the menstrual blood and the peripheral blood are identified without damage, simply, quickly and accurately.

Description

Technical Field

[0001] The present invention belongs to the field of surgical identification, relates to the classification and identification of menstrual blood and peripheral blood, and specifically relates to a classification and identification model of menstrual blood and peripheral blood, and a construction method and application thereof. Background Art

[0002] Identification of the source of body fluids and tissues can be used to infer possible criminal activities at the crime scene, provide clues for the reconstruction of the crime scene, and provide strong evidence support for case detection. As the most common body fluid sample in forensic medicine, the accurate identification of blood is of great significance. Further identification of samples identified as blood to determine whether they are peripheral blood or menstrual blood can provide important directional clues for the characterization of the case. At present, the identification of peripheral blood and menstrual blood, two common body fluids in forensic medicine, is still a difficult point in the identification of the source of body fluids and tissues in forensic medicine.

[0003] In forensic practice, traditional body fluid identification methods such as chemical tests, spectral analysis and microscopy have problems with insufficient specificity and sensitivity. With the development of molecular biology technology, a variety of new body fluid identification methods have gradually emerged, such as RNA analysis, DNA methylation and microbial detection. Among them, RNA analysis (including messenger RNA and microRNA, etc.) technology is relatively mature, but these methods are usually performed in the laboratory and require a series of destructive steps, such as DNA or RNA extraction and amplification, and finally tested on a qPCR platform or sequencing instrument. This has certain limitations for the precious forensic samples.

[0004] In recent years, gas detection technology has developed rapidly, especially the electronic nose technology represented by metal oxide semiconductor sensors. Electronic nose technology is welcomed by gas detection technicians due to its low cost, low power consumption, fast detection speed and portability. Intelligent electronic nose systems have application value and research significance in industrial applications or academic research. In the field of food safety, intelligent electronic nose systems can be used to detect whether food has changed taste or deteriorated. In the field of environmental monitoring, intelligent electronic nose systems can be used in mines or homes to detect whether there are flammable, explosive or toxic gas leaks. In the chemical industry, most chemical raw materials are toxic or flammable gases, and the human membrane sensory system cannot be relied on to detect industrial gas leaks. Therefore, the electronic nose system will be a good toxic and harmful gas monitoring system. In the medical field, the intelligent electronic nose system can detect volatile organic compounds exhaled by the human body to screen for human diseases. Therefore, the intelligent electronic nose system has the significance of in-depth research and a wide range of application scenarios.

[0005] Similar to the human membrane sensory system, the electronic nose system consists of three parts: gas sensor array, signal data processing and pattern recognition. The components of the electronic nose system correspond to the human olfactory system: the gas sensor array corresponds to the sensory receptor part of the human olfactory system, the signal data processing part corresponds to the membrane glomerulus processing electrical signal part of the human membrane sensory system, and the pattern recognition part imitates the function of the cerebral cortex part of the human olfactory system. The target gas interacts with the gas sensor to produce a chemical reaction, converting the chemical signal of the target gas into the electrical signal of the gas sensor, and then undergoes a series of signal preprocessing processes such as signal integration and data normalization, and extracts the main features of the gas sensor response data, and then sends the extracted feature information to the pattern recognition algorithm, and finally completes the qualitative or quantitative analysis of the target gas.

[0006] The function of the gas sensor array is to accurately convert the chemical signal of the gas into a digital signal. Before the digital signal is transmitted to the pattern recognition module, it needs to undergo data preprocessing. Data preprocessing is a data processing process that filters, standardizes, and extracts features from the gas response data collected by the data acquisition board. The data processing process is an important and indispensable processing link of the electronic nose system, and it is also an important method to improve the recognition accuracy and stability of the electronic nose system. Signal filtering is to remove part of the interference noise of the response signal and reduce the interference of other impurity gases, thereby improving the signal-to-noise ratio of the signal. Data standardization is to scale the gas response data to [0, 1] or [-1, 1] to eliminate the order of magnitude difference of signals of different dimensions and balance the influence of different response data on the pattern recognition algorithm. Feature extraction is to condense the information of multi-dimensional data into low-dimensional data and present the information of high-dimensional data in a low-dimensional form. Feature extraction can reduce the amount of data that the pattern recognition algorithm needs to process and can also filter out some unimportant features, thereby improving the efficiency of the algorithm in identifying gases.

[0007] The pattern recognition algorithm of the intelligent electronic nose system analyzes the gas composition or gas concentration based on sample characteristics, and is also a decisive part that affects the gas recognition performance. In actual classification tasks or regression tasks, the original input data usually needs to be preprocessed. The pattern recognition algorithm is used to convert the preprocessed data into a variable space that is easy to analyze and process, making complex classification and regression tasks easier to solve. Classic pattern recognition algorithms include machine learning algorithms such as K nearest neighbors and random forests. In some simple classification or regression tasks, classic machine learning algorithms can achieve satisfactory recognition results.

[0008] Random forest is widely used in practice for classification and regression problems. It has good generalization ability, resistance to overfitting, and robustness to noise. It performs well in processing large-scale data sets and high-dimensional features, and is generally considered to be an efficient and effective machine learning algorithm. In the problem of electronic nose gas classification, the random forest algorithm is widely used. Its advantages include effective processing of high-dimensional feature data, strong resistance to overfitting, ability to adapt to complex data relationships, and easy to explain and understand. By adopting the bagging ensemble method and random feature selection, random forest is able to integrate multiple decision trees, thereby improving classification performance and avoiding the situation where the model overfits the training data.

[0009] Li Chunbao et al. (Li Siyi, et al. Rapid detection of odor in fresh pork based on electronic nose. Food Industry Science and Technology. 2023) studied a method for rapid identification of odorous raw pork. The electronic nose was used to determine the volatile compounds of normal pork and odorous pork in two parts (pork and hind leg meat), and the samples were classified and identified by combining principal component analysis (PCA), linear discriminant analysis (LDA), and random forest (RF), and the headspace gas chromatography-ion mobility chromatography technology was used for auxiliary verification. The results showed that the PCA, LDA, and RF models after electronic nose detection can effectively distinguish normal pork from odorous pork; the hind leg meat test set showed better classification accuracy than the pork, which was 91.00% and 81.00%, respectively. Summary of the invention

[0010] In view of the problems that the prior art requires sample pre-processing (extraction of DNA or RNA, etc.) and damages the original sample, the present invention provides a menstrual blood and peripheral blood classification and identification model and its construction method and application. In order to achieve non-destructive testing, the electronic nose detection system provides a solution. The electronic nose used in the present invention is equipped with 10 different metal oxide sensors to form a sensor array, which can analyze volatile organic compounds in samples. By analyzing the volatile organic compounds in different samples by the electronic nose, the response intensity data of menstrual blood and peripheral blood in the sensor matrix can be obtained. After these response data are statistically analyzed and a random forest model is constructed, a menstrual blood and peripheral blood classification and identification model can be established for classifying and identifying menstrual blood and peripheral blood.

[0011] To achieve the above purpose, the technical solution adopted by the present invention is as follows: In one aspect, the present invention provides a method for constructing a menstrual blood and peripheral blood classification and identification model, comprising the following steps: The volatile organic compounds in menstrual blood and peripheral blood samples were determined by electronic nose analysis method, and the response intensity data of menstrual blood and peripheral blood samples in the electronic nose sensor matrix were obtained. Statistical analysis and machine learning were performed on the response intensity data of the above menstrual blood and peripheral blood samples to establish a classification and identification model for menstrual blood and peripheral blood.

[0012] Preferably, the electronic nose uses the portable PEN3 of the German AIRSENSE company, whose sensor matrix consists of 10 sensors, and the 10 sensors are W1S, W1W, W6S, W1C, W5S, W2S, W3S, W5C, W3C and W2W.

[0013] Preferably, the electronic nose adopts direct headspace aspiration method for sample measurement, and the measurement conditions of the direct headspace aspiration method are: sampling time is 1s / group; sensor self-cleaning time is 80s; sensor zeroing time is 5s; sample preparation time is 5s; injection flow rate is 400ml / min; analysis sampling time is 80s.

[0014] Preferably, the menstrual blood and peripheral blood samples are sampled through a carrier, and the carrier includes a toilet paper carrier and a cotton swab carrier.

[0015] Preferably, the statistical analysis method includes rank sum test, principal coordinate analysis and inter-group similarity analysis.

[0016] Preferably, the rank sum test method is to use the "wilcox.test" function of the R language platform to perform the Wilcoxon rank sum test. The rank sum test is a non-parametric test method, which is mainly used to compare the differences between two independent samples. Its principle is based on whether there is a significant difference between the two sets of data. The rank sum test does not depend on the specific form of the overall distribution. It arranges the ranks in order by the size of the numbers and calculates the sum of the ranks to perform hypothesis testing. It is not limited by the overall distribution and has a wide range of applications; it is suitable for graded data and data without definite values ​​at both ends; it is easy to understand and easy to calculate. When two samples come from two independent but non-normal or unclear populations, and the sample size is small (such as less than 10), the rank sum test is used to compare whether the difference between the two samples is significant. The rank sum test is a simple and effective non-parametric test method, which is suitable for comparing the differences between two independent samples.

[0017] Preferably, the principal coordinate analysis method is to use the "pcoa" function of "ape" in the R language platform to perform principal coordinate analysis. Principal Coordinate Analysis (PCoA) is mainly used to analyze the similarity between samples, and performs nonlinear dimensionality reduction based on the distance or similarity matrix between samples. Distance-based methods such as single-link clustering and complete-link clustering are usually used to map high-dimensional data to low-dimensional space in order to better visualize and analyze the relationship between samples. The coordinates in the principal coordinate analysis results reflect the similarity or distance between samples, and are usually used for clustering and classification analysis to help identify the community structure or classification information of samples.

[0018] Preferably, the inter-group similarity analysis method is to use the "anosim" function in the "vegan" package of the R language platform to perform inter-group similarity analysis. Anosim analysis is a non-parametric test method used to evaluate the overall similarity and similar significance of two or more groups of experimental data. The P value measures whether the difference between groups is statistically significant. If it is less than 0.05 or other common levels, the difference between groups is significant; the R value (-1 to 1) reflects the difference between groups and within groups. 0 is equivalent, and the difference between groups is greater than 0. It mainly solves the problem that the difference between groups is not significant in high-dimensional data analysis. In high-dimensional data analysis, such as microbial community structure and gene expression profile, PCA (principal component analysis), PCoA (principal coordinate analysis), NMDS (non-metric multidimensional scaling analysis) and other methods are often used for dimensionality reduction, but these methods often do not provide indicators of the significance of differences between groups. At this time, Anosim analysis becomes an effective supplementary means.

[0019] Preferably, the machine learning method is to establish a random forest model. Random Forest is a machine learning method that performs classification or regression by constructing multiple decision trees. It combines the ideas of bagging integration and random feature selection to improve the generalization ability and accuracy of the model. The main features of random forest include: 1. The basic classifier is a decision tree: random forest consists of multiple decision trees, each of which is a basic classifier; 2. Bagging integration: Through the Bootstrap sampling method, samples are randomly extracted from the original data set for training each decision tree. In this way, multiple slightly different training sets can be generated, which increases the diversity of the model; 3. Random feature selection: In the process of building each decision tree, a part of the features are randomly selected for node splitting. This can reduce the correlation between features and increase the diversity of the model; 4. Integration of decision trees: random forest votes or averages the classification results of all decision trees to determine the final classification result. The training process of random forest is as follows: 1. Randomly select samples: Bootstrap sampling is performed from the original data set to generate multiple slightly different training sets; 2. Randomly select features: At each node, a part of the features is randomly selected from all the features for node splitting; 3. Construct a decision tree: According to the selected training set and features, a decision tree is constructed; 4. Repeat steps 2 and 3: Repeat multiple times to construct multiple decision trees; 5. Ensemble decision trees: Vote or average the classification results of all decision trees to determine the final classification result.

[0020] Preferably, the random forest model is established by using “randomForest” of the R language platform.

[0021] Specifically, the random seed "set.seed" in the "randomForest" is set to 321, the "srswor" method in the "sampling" package is used to divide the training set and the test set samples, "ntree" is 1000, and the "hellinger" method in the experimental "decostand" function is used to standardize the data.

[0022] Specifically, during the establishment of the random forest model, the samples that have been identified and classified need to be randomly divided into a training set and a test set, and the number of samples in the training set is greater than that in the test set. After the model is initially established using the training set samples, the training set and test set data are used for inspection. If the test set identification accuracy is ≥90%, it means that the random forest model has a good prediction effect and a high accuracy rate. If the test set identification accuracy is <90%, the model prediction and classification effect is poor.

[0023] On the other hand, the present invention provides a menstrual blood and peripheral blood classification and identification model constructed by the above-mentioned construction method.

[0024] On the other hand, the present invention provides an application of the above-mentioned construction method or the above-mentioned menstrual blood and peripheral blood classification and identification model in distinguishing menstrual blood from peripheral blood.

[0025] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention uses 10 sensors of the electronic nose to detect volatile organic compounds, establishes a classification and identification model for menstrual blood and peripheral blood on common carriers, and the prediction accuracy reaches 100%.

[0026] 2. The present invention is the first to use electronic nose technology to identify menstrual blood and peripheral blood. It is portable, non-destructive, does not require pre-treatment, and can simply, quickly and accurately identify menstrual blood and peripheral blood samples, filling the gap in forensic non-destructive identification of peripheral blood and menstrual blood. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 Analysis of electronic nose response values ​​and PCoA analysis of different carriers (Swab: cotton swab carrier; TP: toilet paper carrier).

[0028] A shows the significant differences in the response value data of different sensors; B shows the PCoA analysis results of different carriers.

[0029] Figure 2 Radar chart of the composition and distribution of menstrual blood and peripheral blood in cotton swab carriers. (Blood: peripheral blood; MB: menstrual blood.) Figure 3 PCoA analysis results of cotton swab carrier. (Blood: peripheral blood; MB: menstrual blood.) Figure 4 Random forest prediction model for cotton swab carriers. (The left picture is the training set, the right picture is the test set; Blood: peripheral blood; MB: menstrual blood.) Figure 5 Radar chart of the composition and distribution of menstrual blood and peripheral blood in toilet paper carrier. (Blood: peripheral blood; MB: menstrual blood.) Figure 6 PCoA analysis results of toilet paper carrier. (Blood: peripheral blood; MB: menstrual blood.) Figure 7 Random forest prediction model for toilet paper carrier. (The left picture is the training set, the right picture is the test set; Blood: peripheral blood; MB: menstrual blood.) DETAILED DESCRIPTION It is worth noting that the raw materials used in the present invention are all common commercially available products, and their sources are not specifically limited.

[0030] Example 1: Electronic nose analysis data collection 1) Sample collection In this embodiment, samples of unrelated individuals from adults (aged between 20 and 30 years old) in China were collected, including 30 menstrual blood samples (15 each for toilet paper carriers and cotton swab carriers) and 30 peripheral blood samples (15 each for toilet paper carriers and cotton swab carriers); during the collection of menstrual blood samples, the samples of toilet paper carriers were padded with clean toilet paper on the second day of the menstrual cycle for 1-2 hours; the cotton swab carriers were collected with medical cotton swabs, also on the second day of the menstrual cycle; during the collection of peripheral blood samples, the venous site was selected for disinfection and puncture, and peripheral blood was collected with negative pressure anticoagulation blood collection tubes, and then 200 microliters of blood in the anticoagulation tube was drawn with a pipette and dropped on the head of the medical cotton swab and toilet paper. This study was ethically approved by the Institutional Review Committee of Shanxi Medical University (approval number: 2021GLL049) in accordance with the guidelines of the World Medical Association and the Declaration of Helsinki. All donors voluntarily participated in this study on the basis of informed consent. All collection methods were in compliance with relevant guidelines and regulations. After preparation, the samples were stored in a -80°C refrigerator for future use.

[0031] 2) Sample pretreatment After taking out the sample, place it in a clean 100mL beaker, seal it with a double layer of plastic wrap, and let it stand at room temperature for 2 hours before testing on the machine.

[0032] 3) Instruments and equipment The electronic nose system was purchased from AIRSENSE, Germany, model PEN3. The electronic nose contains 10 different metal oxide sensors, forming a sensor array, and the specific description is shown in Table 1.

[0033] Table 1 Sensor performance description

[0034] 4) How the electronic nose works The direct headspace aspiration method was used for determination. The electronic nose instrument started testing 2 minutes after it was turned on. After taking out the sample, place it in a clean 100mL beaker, seal it with a double layer of plastic wrap, and let it stand at room temperature for 2 hours before testing on the machine. Insert the injection needle directly into the headspace of the 100mL headspace injection bottle, and insert an activated carbon filter gas replenishment device to ensure the pressure balance in the sample bottle. Use the electronic nose to monitor the effect of continuous volatilization of the sample gas. Determination conditions: sampling time is 1s / group; sensor self-cleaning time is 80s; sensor zeroing time is 5s; sample preparation time is 5s; injection flow rate is 400ml / min; analysis sampling time is 80s.

[0035] 5) Results collation The three response values ​​(G / G0) corresponding to 70-72s in the electronic nose detection data of each sample were selected, and then the average value was calculated. The average value corresponding to the cotton swab carrier sample is shown in Table 2, and the average value corresponding to the toilet paper carrier sample is shown in Table 3 (blood represents peripheral blood samples, and MB represents menstrual blood samples) for subsequent data analysis.

[0036] Table 2 Average response values ​​corresponding to cotton swab carrier samples

[0037] Table 3 Average response values ​​corresponding to toilet paper carrier samples

[0038] Example 2: Statistical analysis and establishment of random forest model 1) Analytical methods The response value data obtained in Example 1 were subjected to statistical analysis and machine learning using R language platform software (v3.6.0; http: / / www.r-project.org / ): the Wilcoxon rank sum test was performed using the default parameters of the "wilcox.test" function; principal coordinate analysis was performed using the "pcoa" function of the R package "ape" based on the Bray-Curtis distance to visualize the differences between samples at different time points, and the data were standardized using the "hellinger" method in the decostand function, while "bray" was selected to calculate the distance between samples; Anosim analysis was performed using the "anosim" function in the "vegan" package in the R language; a random forest model was established using the "randomForest" function in the R package, with the random seed set.seed set to 321, the "srswor" method in the "sampling" package was used to divide the samples into training and test sets, ntree was 1000, and the data were standardized using the "hellinger" method in the experimental "decostand" function. When constructing the random forest model, the toilet paper carrier samples in Example 1 were randomly divided into a training set and a test set, with 18 samples in the training set (9 samples each for menstrual blood and peripheral blood) and 12 samples in the test set (6 samples each for menstrual blood and peripheral blood); the cotton swab carrier samples in Example 1 were randomly divided into a training set and a test set, with 18 samples in the training set (9 samples each for menstrual blood and peripheral blood) and 12 samples in the test set (6 samples each for menstrual blood and peripheral blood).

[0039] 2) Analyze the results 2.1 Analysis of electronic nose response values ​​and PCoA analysis of different carriers The odor detected on the two blood carriers (cotton swab carrier and toilet paper carrier) is different. Figure 1 A in the figure shows that during the detection of two carrier samples, the detection response values ​​of the electronic nose sensors W1S, W1W, W6S, W1C, W5S and W2S were significantly different (P<0.05), and the detection response values ​​of the electronic nose sensors W3S, W5C, W3C and W2W were different, but not significantly (rank sum test). Figure 1 Figure B shows the results of principal coordinate analysis. The first axis (horizontal axis) of the PCoA graph explains 95.76% of the variance, and the second axis (vertical axis) explains 4.41% of the variance. The samples were clustered according to their types. Further anosim analysis found that R=0.716 and P=0.008, indicating that the two groups of carrier samples have little similarity and large differences.

[0040] 2.2 Radar chart, PCoA and random forest prediction model of cotton swab carrier Figure 2 The radar chart in the middle shows the distribution of the odor of menstrual blood and peripheral blood on the cotton swab carrier: the sensor S7 has the highest response value, which is sensitive to sulfides, many terpenes and organic sulfides; followed by the sensor S9, which is sensitive to aromatic components and organic sulfides. The response values ​​of peripheral blood samples at S2, S6, and S8 sensors are higher than those of menstrual blood samples.

[0041] PCoA analysis Figure 3 As shown in the figure, the volatile substances in peripheral blood and menstrual blood were clustered according to their types. The first coordinate explained 72.32% of the variance; the second coordinate explained 28.64% of the variance. Anosim analysis showed R=0.1446, P=0.003, indicating that there were significant differences between peripheral blood and menstrual blood on the cotton swab carrier.

[0042] Next, the machine learning algorithm was used to classify and analyze the peripheral blood and menstrual blood on the cotton swab carrier, and a random forest model was established through the training set. The results are as follows: Figure 4 As shown, the accuracy of predicting peripheral blood and menstrual blood in the training set (9 cases each of peripheral blood and menstrual blood) and the test set (6 cases each of peripheral blood and menstrual blood) reached 100%.

[0043] 2.3 Construction of toilet paper carrier radar chart, PCoA and random forest model Figure 5 The middle radar chart shows the distribution of the odor of menstrual blood and peripheral blood on the cotton swab carrier: on the toilet paper carrier, similar to the cotton swab carrier, the sensor S7 has the highest response value, followed by sensor S9. Next are sensors S6 and S2. For the toilet paper carrier as a whole, the response value of menstrual blood on the toilet paper carrier is higher than that on the cotton swab carrier.

[0044] PCoA analysis Figure 6As shown in the figure, the volatile substances in peripheral blood and menstrual blood were clustered according to their types. The first coordinate explained 92.73% of the variance; the second coordinate explained 5.01% of the variance. Anosim analysis showed R value = 0.9006, P < 0.001, indicating that there were significant differences between peripheral blood and menstrual blood on the toilet paper carrier.

[0045] Next, a machine learning algorithm was used to classify and analyze the peripheral blood and menstrual blood on the toilet paper carrier, and a random forest model was established using the training set. The results are as follows: Figure 7 As shown, the accuracy of predicting peripheral blood and menstrual blood in the training set (9 cases each of peripheral blood and menstrual blood) and the test set (6 cases each of peripheral blood and menstrual blood) reached 100%.

[0046] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A method for constructing a classification and identification model for menstrual blood and peripheral blood, characterized in that: The following steps are involved: The volatile organic compounds in menstrual blood and peripheral blood samples were determined by electronic nose analysis method, and the response intensity data of menstrual blood and peripheral blood samples in the electronic nose sensor matrix were obtained. Statistical analysis and machine learning were performed on the response intensity data of the above menstrual blood and peripheral blood samples to establish a classification and identification model for menstrual blood and peripheral blood.

2. The construction method according to claim 1, characterized in that: The sensor matrix of the electronic nose consists of sensors W1S, W1W, W6S, W1C, W5S, W2S, W3S, W5C, W3C and W2W.

3. The construction method according to claim 1, characterized in that: The electronic nose adopts direct headspace aspiration method for sample measurement, and the measurement conditions of the direct headspace aspiration method are: sampling time is 1s / group; sensor self-cleaning time is 80s; sensor zeroing time is 5s; sample preparation time is 5s; injection flow rate is 400ml / min; and analysis sampling time is 80s.

4. The construction method according to claim 1, characterized in that: The menstrual blood and peripheral blood samples are sampled through a carrier, and the carrier includes a toilet paper carrier and a cotton swab carrier.

5. The construction method according to claim 1, characterized in that: The statistical analysis methods include rank sum test, principal coordinate analysis and inter-group similarity analysis.

6. The construction method according to claim 5, characterized in that: The rank sum test method is to use the "wilcox.test" function of the R language platform to perform the rank sum test, the principal coordinate analysis method is to use the "pcoa" function in the "ape" package in the R language platform to perform the principal coordinate analysis, and the inter-group similarity analysis method is to use the "anosim" function in the "vegan" package in the R language platform to perform the inter-group similarity analysis.

7. The construction method according to claim 1, characterized in that: The machine learning method is to establish a random forest model.

8. The construction method according to claim 7, characterized in that: The random forest model was established using the "randomForest" tool in the R language platform.

9. The menstrual blood and peripheral blood classification and identification model constructed by the construction method according to any one of claims 1 to 8.

10. Use of the construction method according to any one of claims 1 to 8 or the menstrual blood and peripheral blood classification and identification model according to claim 9 in identifying menstrual blood and peripheral blood.

Images