Tongue coating microorganism-based tumor prediction system, method and applications thereof

Abstract

This invention relates to the fields of oncology diagnosis, prediction, and assessment, and particularly to a tumor prediction system, method, and application based on tongue coating microorganisms. The system includes: a microbial information acquisition module configured to acquire tongue coating microbial information of a test sample; and a data processing module configured to obtain the probability that the test sample is tumor-positive by predicting the probability of a positive result based on discriminative features of the tongue coating microbial information obtained through automatic learning. The system automatically predicts the probability of different test samples being tumor-positive based on the abundance of different species and genera in the tongue coating microorganisms, thus providing an economical, non-invasive, efficient, and accurate early tumor screening strategy.

Description

[0001] This application is a divisional application of Chinese invention patent application filed on July 22, 2022, with application number 202210868365.0 and invention title "Tumor Prediction System, Method and Application Based on Tongue Coating Microorganisms". Technical Field

[0002] This invention relates to the field of oncology diagnosis, prediction, and assessment technology, and more specifically, to a tumor prediction system, method, and application based on tongue coating microorganisms. By analyzing the correlation between tongue coating microorganisms and oncology, an economical, non-invasive, and highly accurate tumor prediction can be achieved. Background Technology

[0003] According to the latest data, gastric cancer (GC) is the third leading cause of cancer-related deaths worldwide, with 1.09 million new GC cases and 770,000 deaths in 2020 alone. Of these, China accounted for 480,000 new cases and 370,000 deaths, roughly half of the world's cases. China has a high incidence and mortality rate of gastric cancer. Early detection, early diagnosis, and early treatment are crucial to reducing the mortality rate. However, the current early diagnosis rate of gastric cancer in China is still less than 20%. The diagnosis and screening of GC still rely on gastroscopy, but its application is greatly limited due to its invasiveness, high cost, and the need for specialized endoscopists. Furthermore, because early gastric cancer lacks specific symptoms, the specificity and sensitivity of clinical disease markers are poor, and more than 60% of patients have local or distant metastases at the time of diagnosis. The 5-year survival rate for patients with locally early GC exceeds 60%, while the 5-year survival rates for patients with local and distant metastases drop significantly to 30% and 5%, respectively. Therefore, new diagnostic or screening methods for GC are urgently needed to improve the early diagnosis rate and prognosis in this population.

[0004] Traditional Chinese medicine (TCM) is a medical science and cultural heritage that has been applied and preserved by the Chinese people over thousands of years. Tongue diagnosis is one of the important bases for TCM disease diagnosis. TCM theory holds that changes in the tongue (the color, size, and shape of the tongue, and the color, thickness, and moisture content of the tongue coating) can reflect the health status of the body, especially closely related to stomach diseases. In the diagnosis of stomach diseases in TCM, the specific manifestations of stomach diseases are often determined based on tongue information using experience or dialectical methods. Furthermore, research has shown a close relationship between the oral cavity and the tongue coating microbiota.

[0005] Artificial intelligence (AI) can be used to screen, diagnose, and treat various diseases. A paper by Cheung CY et al. (Cheung CY, Xu D, Cheng CY, et al. A deep-learning system for the assessment of cardiovascular disease risk via the measurement of retinal-vessel calibre. Nature biomedical engineering 2021;5(6):498-508. doi: 10.1038 / s41551-020-00626-4 [published Online First: 2020 / 10 / 14]) discloses a deep learning system that assesses the risk of cardiovascular disease by measuring the diameter of retinal vessels, and can effectively predict the risk of cardiovascular disease. A paper by Takenaka K et al. (Takenaka K, Ohtsuka K, Fujii T, et al. Development and Validation of a Deep Neural Network for Accurate Evaluation of Endoscopic Images From Patients With Ulcerative Colitis. Gastroenterology 2020;158(8):2150-57. doi: 10.1053 / j.gastro.2020.02.012 [published Online First: 2020 / 02 / 16]) discloses a deep neural network (see references) for evaluating endoscopic images of patients with ulcerative colitis. The network identifies patients with endoscopic remission and histological remission with an accuracy of 90.1% and an accuracy of 92.9%.

[0006] Fuzhou Data Technology Research Institute Co., Ltd.'s patent CN110251084A provides an artificial intelligence-based tongue image detection and recognition method to solve the real-time detection, shooting, storage, and uploading of tongue images during the tongue image acquisition process, while recognizing the tongue image's color, shape, coating texture, and color. The solution mainly involves tongue image acquisition and recognition technology, with tongue image recognition focusing more on extracting characteristics such as tongue image color, texture, tongue coating area, or tongue coating thickness. However, these works do not establish a correspondence between tongue image and tongue coating information and a specific gastric disease such as gastric cancer.

[0007] Shenyang Zhilang Technology Co., Ltd.'s patent CN111710394A proposes an artificial intelligence-assisted early gastric cancer screening system to solve the problem of the large workload in confirming positive gastric cancer by automating the analysis of gastroscopy slice images instead of manual analysis. However, this strategy based on gastroscopy image analysis still requires a large number of gastroscopy images collected by professional instruments for model learning. In the testing phase, decisions still need to be made based on the gastroscopy images of each test subject. However, obtaining gastroscopy images still has drawbacks such as high time consumption, high material costs, and high standards for the test population, making it difficult to achieve nationwide screening.

[0008] Jiangsu Tianrui Precision Medical Technology Co., Ltd.'s patent CN112133427A provides an artificial intelligence-based auxiliary diagnostic system for gastric cancer, including: a diagnostic selection module, a data acquisition module, a preprocessing module, a diagnostic module, and a display output module. This system can provide personalized diagnostic results based on the collected patient data. The data used in this diagnostic system includes the patient's basic information, lifestyle and diet, infection history, disease history, family history, clinical symptoms, and laboratory test results. Among these, collecting data such as clinical symptoms and laboratory test results is relatively difficult, and relying solely on basic information, lifestyle and diet, infection history, disease history, and family history can affect the effectiveness of initial screening and diagnosis.

[0009] Shanghai Rendong Medical Laboratory Co., Ltd.'s patent CN114203256A provides a method for constructing a MIBC (muscle-invasive bladder cancer) subtyping and prognostic prediction model based on microbial abundance. This method primarily analyzes MIBC transcriptome RNA-seq data from the TCGA database to obtain microbial data from MIBC patients. Then, it performs NMF clustering based on microbial abundance profiles to establish a molecular subtyping of MIBC at the microbial level. This allows for in-depth analysis of the correlation between microorganisms and MIBC at the tumor tissue microbial level, establishing a MIBC prognostic prediction model. This model helps to accurately predict the 1-5 year survival rate of MIBC patients. Therefore, this invention mainly aims to establish a molecular subtyping of MIBC at the microbial level, with the goal of accurately predicting patient prognostic survival rates.

[0010] Unfortunately, no studies have yet confirmed a correlation between changes in the tongue coating microbiota and tumors, or the value of changes in the tongue coating microbiota in tumor diagnosis and screening.

[0011] The present invention aims to address these and other unresolved needs in the art. Summary of the Invention

[0012] To address at least one of the technical problems mentioned in the background section, this invention aims to design a deep learning-based tumor screening system using computer-aided methods. This system automatically predicts the probability that different test samples are tumor-positive based on the abundance of different species and genera in tongue coating microorganisms, thereby serving as an economical, non-invasive, efficient, and accurate early tumor screening strategy.

[0013] A tumor prediction system based on tongue coating microbes includes: The microbial information acquisition module is configured to acquire the tongue coating microbial information of the test sample; The data processing module is configured to obtain the probability that a test sample is tumor-positive by performing the following operations: Based on the discriminative features of the tongue coating microbial information obtained through automatic learning, the probability of a test sample being positive is predicted.

[0014] In one specific embodiment, the tumor is at least one of gastric cancer, breast cancer, colorectal cancer, esophageal cancer, hepatobiliary and pancreatic cancer, lung cancer, prostate cancer, thyroid cancer, ovarian cancer, neuroblastoma, trophoblastic tumor, or head and neck squamous cell carcinoma.

[0015] In one specific embodiment, the tumor is at least one of gastric cancer, breast cancer, colorectal cancer, esophageal cancer, hepatobiliary and pancreatic cancer, or lung cancer.

[0016] In one specific embodiment, the tumor is gastric cancer.

[0017] In one embodiment, the system further includes an output module configured to output prediction results.

[0018] In one specific embodiment, the output module is configured to output tongue coating microbial information and prediction results.

[0019] In one specific embodiment, the output module outputs in at least one mode: electronic display, sound broadcast, printing, or network transmission.

[0020] In one specific embodiment, the tongue coating microbial information includes the genus and species abundance of tongue coating microorganisms.

[0021] In one specific embodiment, the discriminability features are derived from the genus and species abundance of the microorganisms.

[0022] In one specific embodiment, the discriminative features are derived from the high-dimensional features of the genus and species abundance of microorganisms. The aim is to fully compare, analyze, and learn the commonalities and differences in tongue coating microbial information between positive and negative patients. By deeply discriminating the discriminative features of the genus and species abundance of microorganisms in the test sample's tongue coating, the differences between positive and negative patients can be obtained, thus determining the probability that the test sample is tumor-positive. Therefore, through the comparison, analysis, and learning of tongue coating microbial information, the diagnosis and prediction of tumors in the test sample can be achieved, providing a non-invasive, non-human tissue-derived, cost-effective, and highly accurate tumor diagnosis and prediction system.

[0023] In one specific embodiment, the data processing module obtains the probability that the test sample is tumor-positive by performing the following operations: After training, the neural network extracts high-dimensional features from the abundance of microbial genera and species in the input and then predicts the probability that the test sample is positive.

[0024] In one specific embodiment, the neural network is a multilayer perceptron (MLP).

[0025] In one specific embodiment, the neural network is trained using the following steps: 1) The abundance of tongue microbial genera and / or species collected from tumor-positive patients and / or tumor-negative individuals is used as an input vector and input into the input layer of the model; 2) The hidden layer of the model extracts high-dimensional features of microbial species and / or genus abundance; 3) Output the probability distribution of positive and negative genera and / or species of tongue coating microorganisms through the softmax classifier of the output layer.

[0026] In one specific embodiment, the length of the input vector is 706 / 1339, representing the genus / species of microorganism, respectively.

[0027] In one specific embodiment, each element of the input vector represents the abundance of microorganisms in a particular genus or species.

[0028] In one specific embodiment, the input layer corresponds to the length of the input vector.

[0029] In one specific embodiment, the number of neurons in the hidden layer is set to 512. Since the number of neurons in the hidden layer is greater than the number of neurons in the input layer, during the forward operation, the abundance features of the input layer are nonlinearly mapped to a high-dimensional space to form high-dimensional features. After extracting these high-dimensional features, the probability of the essential oil input layer outputting either positive or negative can be determined.

[0030] In one specific embodiment, each layer in the neural network, except for the output layer, is accompanied by an activation function and normalization.

[0031] In one specific embodiment, the output layer includes two neurons: a tumor-positive neuron and a tumor-negative neuron.

[0032] In one specific embodiment, the neurons in each layer of the neural network are interconnected, and the cross-entropy objective function is minimized using the probability distribution of the output:

[0033] Where y i These are elements in the actual one-hot annotation vector corresponding to the test sample. The model predicts the class y. i The probability is given by 0, where 0 represents the negative category and 1 represents the positive category. One-hot annotation is a vector of 0 and 1. For example, if there are two categories, the one-hot annotations corresponding to categories 0 and 1 are (1,0,0) and (0,1,0).

[0034] Validated by a large number of real patients, the application of tongue microbiota as a non-invasive diagnostic and screening method for tumors is significantly superior to conventional blood tumor markers. The tumor prediction system based on tongue microbiota exhibits superior genus / species sensitivity (0.914 / 0.929 vs 0.283-0.566, 0.362-0.539, 0.947 / 0.947 vs 0.688-0.976, 0.759-0.938), specificity (0.947 / 0.947 vs 0.688-0.976, 0.759-0.938), and accuracy (0.929 / 0.937 vs 0.603-0.622, 0.645-0.662) compared to conventional AI models based on blood tumor markers. Its AUC value is also higher (0.945 / 0.975 vs 0.945 / 0.975). (0.682-0.715, 0.694-0.760). Considering the enormous burden of cancer detection in China and globally, it is believed that the widespread use of tongue microbiota combined with artificial intelligence deep learning methods is the most economical, non-invasive, and acceptable method for screening and predicting early-stage cancers, which will also bring about huge socio-economic impacts.

[0035] Increasing the number of hidden layers or adjusting the number of hidden layer units can achieve similar recognition accuracy. In other words, neural networks composed of fully connected layers with different hyperparameters are suitable for tumor positivity discrimination tasks based on microbial abundance. Through deep learning technology, the probability of tumor positivity can be automatically determined to screen out high-incidence groups of tumors.

[0036] Tumor prediction methods based on tongue coating microbiota include: Obtain the tongue coating microbial information of the test sample; The microbial information of the tongue coating of the test sample is input into the aforementioned system to obtain the tumor positivity probability of the test sample.

[0037] The information on tongue coating microorganisms includes the genus and species abundance of tongue coating microorganisms.

[0038] The aforementioned application of the tumor prediction system and / or method based on tongue coating microbes includes: The system and / or method described herein are used to predict tumors in test specimens.

[0039] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined to obtain specific implementation methods.

[0040] The beneficial effects of this invention are as follows: This invention aims to provide a tumor screening system based on deep learning. By fully comparing, analyzing, and learning the commonalities and differences in tongue coating microbial information between positive and negative patients, the system obtains the differences between positive and negative patients. By deeply discriminating the discriminability between the genera and species abundance of tongue coating microorganisms in the test sample, the probability of the test sample being tumor-positive can be determined. Thus, through the analysis and learning of tongue coating microbial information, the system can achieve the diagnosis and prediction of tumors in the test sample. Verification shows that the genera / species sensitivity for predicting gastric cancer using tongue coating microorganisms reaches 0.914 / 0.929, the specificity reaches 0.947, the accuracy reaches 0.929 / 0.937, and the AUC value reaches 0.945 / 0.975. The predictive effect is superior to conventional artificial intelligence models based on blood tumor markers. This system can serve as an economical, non-invasive, efficient, and accurate early tumor screening strategy, which will undoubtedly bring significant socio-economic benefits.

[0041] The present invention adopts the above-mentioned technical solution to achieve the above objectives, which makes up for the shortcomings of the prior art, is reasonably designed, and is easy to operate. Attached Figure Description

[0042] Some of the accompanying drawings are provided to enable those skilled in the art to more quickly and clearly understand the above and / or other objects, features, advantages and examples of this application. It should be noted that the drawings, illustrative embodiments and their descriptions constituting this application are used to provide a further understanding of this application and do not constitute an improper limitation of this application.

[0043] Figure 1 This is a schematic diagram of a multicenter clinical study and its patient distribution; Figure 2 This is the overall algorithm structure of the neural network in this application; Figure 3 The ROC and AUC are based on internal and external validation of three models (SVM, DT, KNN) for eight hematological malignancies. Figure 4 This is a principal coordinate analysis diagram of microorganisms in the GCs and NGCs groups; Figure 5 This is a heatmap of the top 30 microorganisms in the GCs and NGCs groups; Figure 6 It is the Shannon Diversity Index; Figure 7 The ROC and AUC of the tumor prediction system based on tongue coating microbes on the validation set; Figure 8 This is a distribution map of microbial species in the GCs and NGCs groups (where E shows the distribution of microbial species at the genus level, and F shows the distribution of microbial species at the species level). Detailed Implementation

[0044] Those skilled in the art can refer to the content of this document and appropriately replace and / or modify the process parameters to achieve the desired results. However, it should be particularly noted that all similar replacements and / or modifications are obvious to those skilled in the art and are considered to be included in this invention. The content of this invention has been described through preferred examples, and those skilled in the art can obviously make modifications or appropriate changes and combinations to the content described herein without departing from the content, spirit and scope of this invention to realize and apply the technology of this invention.

[0045] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the technical solutions of this application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0047] In this invention, the microbial information of the tongue coating specifically includes the genus, species, and abundance of microorganisms. Any method that can obtain the genus, species, and abundance of microorganisms in the sample tongue coating can be used to obtain the relevant information. This application adopts the following method to obtain the microbial information: Before eating breakfast and drinking water, the subjects (including the training set and the test set) rinsed their mouths three times with sterile water. Professional operators collected their tongue samples using tongue swabs. Specifically, the swab was rolled while scraping the tongue from the base to the tip 30 times (5 rolls per swab, 6 swabs in total). After sampling, the swabs were immediately placed in a freezer tube and then the samples were transferred to a freezer at -80°C. Microbial DNA was extracted using the EZNA Tissue DNA Extraction Kit (D3396-01; Omega, Norcross, Georgia, USA) according to the manufacturer's instructions. PCR products were isolated, extracted, and purified using the AxyPrep PCR Purification Kit (AP-PCR-500G; Corning, NY, USA), and quantified using the Quant-iT PicoGreen dsDNA reagent (P7581, Thermo Scientific, Waltham, MA, USA). Finally, 2×250bp end sequencing was performed using a Novaseq sequencer from LC-BioCo., Ltd.

[0048] The present invention will now be described in detail.

[0049] <Clinical Specimens> A nationwide multicenter clinical study was conducted to eliminate the influence of regional, dietary, and center differences on the study, including 11 centers in 8 cities, located in Hangzhou, Wenzhou, and Shanghai in the east, Fuzhou in the south, Chengdu in the west, Liaoning and Heilongjiang in the north, and Taiyuan in the central region.

[0050] like Figure 1 As shown, from January 2020 to October 2021, 1111 patients with gastric cancer (GC) were recruited from 8 centers, and 1519 patients without gastric cancer (NGC) were recruited from 3 centers, including 169 healthy controls (HCs), 648 patients with superficial gastritis (SGs), and 702 patients with atrophic gastritis (AGs). 865 patients from the GC group and 1287 patients from the NGC group were randomly selected to train and validate the aforementioned system. These included 448 patients with early-stage GC (TNM1+II), 417 patients with late-stage GC (TNM3+IV), 141 healthy controls (HCs), 547 patients with superficial gastritis (SGs), and 599 patients with atrophic gastritis (AGs). Approximately 80% of the cases were used as the training dataset, and approximately 20% were used as the internal validation dataset. In addition, 246 cases of GC and 232 cases of NGC from three centers were used as independent external validation datasets, including 162 cases of early GC, 84 cases of late GC, 28 cases of HC, 101 cases of SG, and 103 cases of AG. These gastric cancer (GC) patients were all newly diagnosed with gastric cancer, had not previously received treatment for their disease, and had not undergone surgery, chemotherapy, radiotherapy, targeted therapy, or biotherapy. No single tumor was found in any of the GC patients; patients with two or more malignancies were excluded. HCs, SGs, and AGs were confirmed by gastroscopy.

[0051] Clinical information was collected from all participants, including age, sex, height, weight, family history, smoking, alcohol consumption, TNM stage, and blood tumor markers. Pathological staging was based on the American Joint Committee on Cancer 8th edition, stage 23. Table 1 shows general patient information between the GC and NGC groups, such as age, sex, BMI, smoking, and alcohol consumption, which showed a very good match in the training, internal validation, and independent external validation datasets.

[0052] Table 1. Clinical information of GC and NGC participants

[0053] <Statistical Analysis> All statistical analyses were performed using SPSS 23.0 software (SPSS Inc., Chicago, IL, USA). Results are expressed as mean ± SD or mean ± SEM. Parametric or nonparametric tests were used depending on whether the data were orthogonally distributed. Chi-square tests were used for count data. P < 0.05 was considered statistically significant.

[0054] <Ethical Approval> This application was approved by the centralized ethics committees of 11 participating centers (IRB-2019-56), including the Research Ethics Committee of Zhejiang Cancer Hospital, the First Affiliated Hospital of Wenzhou Medical University, Liaoning Cancer Hospital, Renji Hospital Affiliated to Shanghai Jiao Tong University, Fujian Cancer Hospital, Harbin Medical University Cancer Hospital, Sichuan Cancer Hospital, Shanxi Cancer Hospital, Zhejiang Tongde Hospital, Zhejiang Provincial Hospital of Traditional Chinese Medicine, and Yuhang People's Hospital.

[0055] <Clinical Validation> Example 1: Using computer-aided methods, a tumor prediction system based on deep learning was designed. This system automatically predicts the probability that different test samples are tumor-positive based on the abundance of microorganisms in different genera and species.

[0056] We obtained data on the abundance of genus and species of tongue coating microorganisms from positive and negative samples from gastric cancer patients and non-gastric cancer individuals, respectively.

[0057] Based on the above abundance data of tongue coating microorganisms by genera and species, a neural network based on multilayer perceptron (MLP) was designed. By taking the abundance of tongue coating microorganisms in different genera and species as input, the probability of the test sample being positive is finally predicted based on the extracted high-dimensional features.

[0058] The overall algorithm structure is as follows Figure 2As shown, the input to the entire model is a vector of length 706 or 1339, representing the number of microbial genera or species, respectively. Each element of the input vector represents the abundance of microorganisms in a specific genus or species. The input layer of the multilayer perceptron corresponds to the length of the input vector. High-dimensional features of microbial abundance are extracted by adding hidden layers. The output layer contains two neurons and a softmax classifier, outputting the probabilities of the input sample being positive and negative for gastric cancer, respectively. The neurons in each layer are interconnected, and the cross-entropy objective function is minimized using the probability distribution of the output.

[0059] Where y i These are elements in the true one-hot annotation vector corresponding to the test sample. The model predicts the class y. i The probability is given by 0, where 0 represents the negative category and 1 represents the positive category. One-hot annotation is a vector of 0 and 1. For example, if there are two categories, the one-hot annotations corresponding to categories 0 and 1 are (1,0,0) and (0,1,0).

[0060] Tongue microbiome data were collected from 328 GC (gastric cancer) and 304 NGC (non-gastric cancer, including 155 HCs (healthy) and 149 AGs (atrophic gastritis)) participants from Zhejiang Cancer Hospital and the First Affiliated Hospital of Zhejiang University of Traditional Chinese Medicine. The clinical information of participants with gastric cancer and those without gastric cancer is shown in Table 2. It can be seen that the two groups were well matched in terms of general clinical information such as age, gender, smoking and drinking.

[0061] Table 2. Clinical information of participants regarding tongue coating

[0062] Data from participants with gastric cancer and those without gastric cancer were divided into training and test sets in an 8:2 ratio. Sensitivity, specificity, and accuracy on the test set are shown in Tables 3 and 4.

[0063] Table 3. Test Results (Genus)

[0064] Table 4. Test Results (Species)

[0065] Tables 3 and 4 show that the tongue coating microbiome exhibits the same high specificity for gastric cancer diagnosis and prediction at both the genus and species levels. The similarity in data may be due to limited sample size. It should be noted that the sensitivity and accuracy of the microbiome at the species level are higher than at the genus level, demonstrating high sensitivity, specificity, and accuracy for gastric cancer diagnosis and prediction. For comparison, the tumor detection system based on tongue coating microbes is compared with conventionally used blood tumor markers. A combination of several classic blood tumor markers is used to verify their predictive value for tumors. Available blood tumor markers include alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), cancer antigen 125 (CA125), cancer antigen 15-3 (CA15-3), cancer antigen 199 (CA199), cancer antigen 72-4 (CA72-4), and cancer antigen 242 (CA242). The tumor markers include at least one of the following: CA50, non-small cell lung cancer-associated antigen (CYFRA21-1), small cell lung cancer-associated antigen (neuron-specific enolase, NSE), squamous cell carcinoma antigen (SCC), total prostate-specific antigen (TPSA), free prostate-specific antigen (FPSA), α-L-fucosidase (AFU), EBV-VCA antibody, tumor-associated substance (TSGF), ferritin, β2-microglobulin (β2-MG), pancreatic embryonic antigen (POA), or progastrin-releasing peptide (PROGRP), particularly at least one of CEA, CA242, CA72-4, CA125, CA199, CA50, AFP, or ferritin, and more particularly a combination of the above eight hematologic tumor markers. The predictive method based on the above-mentioned hematologic tumor markers includes the following steps: 1) Data Preprocessing: Since all cases had varying degrees of missing serum markers, and the training data needed to be complete, the data needed to be completed before model training. This application uses the K-nearest neighbor missing value imputation method to complete the data; specifically, the imputed value for missing serum markers is the average of the values ​​of the two nearest neighbors. 2) Model Training: This invention employs three machine learning classification methods: Support Vector Machine (SVM), Decision Tree (DT), and K-Nearest Neighbors (KNN). Specifically, the eight blood tumor markers of the case (CEA, CA242, CA72-4, CA125, CA199, CA50, AFP, and Ferritin) correspond to sample features, and the positive or negative diagnosis of the case corresponds to the sample label. All completed samples are fed into the three classifiers for fitting. 3) Model Evaluation: This application evaluated the model using internal and external validation. Internal validation used data from the same hospital but different cases as the training data, while external validation used case data from a different hospital than the training data. Three metrics, including sensitivity, specificity, and accuracy, were used to evaluate the model's predictions.

[0066] The clinical information of blood tumor markers in GC patients is shown in Table 5. It can be seen that compared with NGC patients, the concentrations of blood tumor markers such as CEA, CA424, CA724, CA125, CA199, CA50, AFP and Ferritin were significantly increased in GC patients.

[0067] Table 5. Clinical information on hematological tumor markers in GC patients

[0068] Approximately 80% of the cases were used as the training dataset, and approximately 20% were used as the internal validation dataset. 246 cases of GC and 232 cases of NGC from three centers were used as independent external validation datasets. The sensitivity, specificity, and accuracy validation results of the machine learning classification method based on hematological tumor markers for GC diagnosis are shown in Table 6. The internal validation sensitivity of the three models based on hematological tumor markers (SVM, DT, and KNN) ranged from 0.283 to 0.566, specificity from 0.688 to 0.976, and accuracy from 0.603 to 0.622. The external validation sensitivity ranged from 0.362 to 0.539, specificity from 0.759 to 0.938, and accuracy from 0.645 to 0.662. The ROC and AUC of the machine learning classification method based on hematological tumor markers for GC diagnosis in internal and external validation are shown in [reference needed]. Figure 3 The AUC values ​​for internal validation range from 0.682 to 0.715, while those for external validation range from 0.694 to 0.760. This indicates that the SVM algorithm achieves specificity exceeding 90% for both internal and external validation, demonstrating its valuable information for gastric cancer diagnosis. In contrast, DT and KNN show decreased specificity but improved sensitivity and accuracy to varying degrees, providing multifaceted information for gastric cancer diagnosis.

[0069] Table 6. Sensitivity, specificity, and accuracy of models based on blood tumor markers for GC diagnosis.

[0070] It should be clarified that the comparative scheme described in this application uses eight serum indicators, including CEA, CA242, CA72-4, CA125, CA199, CA50, AFP, and Ferritin. Adding, reducing, or replacing several serum indicators can predict the positivity or seroprevalence of tumors, especially gastric cancer. The comparative scheme employs three machine learning classifiers: SVM, DT, and KNN. Other machine learning classifier methods, such as logistic regression and random forest, can also achieve the same purpose.

[0071] Comparative analysis of Tables 3-4 and 6 shows that the tumor prediction system based on tongue coating microorganisms provided in this application has significantly better sensitivity and accuracy than the above-mentioned blood tumor marker-based models, and its specificity is better than the DT and KNN models. The proposed solution is forward-looking, cost-effective, non-invasive, and has higher sensitivity, specificity and accuracy.

[0072] Principal Coordinate Analysis (PcoA) is a visualization method that reduces the dimensionality of multidimensional data to study data similarity or difference. It uses a distance matrix to find the principal coordinates. After sorting a series of eigenvalues ​​and eigenvectors, it selects the top-ranking eigenvalues ​​to effectively identify the most "primary" elements and structures in the data, describing the relationships between samples. First, random sampling is used to calculate the one-sided distance between each sample. Then, a two-dimensional PCoA plot is drawn based on the distance matrix. Therefore, the closer the samples are, i.e., the more similar the abundance and composition of microorganisms, the closer they will be in the PCoA plot. The principal coordinate analysis plots of microorganisms in the GCs and NGCs groups are shown below. Figure 4 As shown, the heatmaps of the top 30 microorganisms in the GCs and NGCs groups are as follows: Figure 5 As shown, analysis Figure 4 and Figure 5 It can be seen that there are significant differences between the microbiomes of the GCs group and the NGCs group.

[0073] The Shannon diversity index is an index used to investigate territorial diversity (α-diversity) of a community. Its value is positively correlated with species diversity. The Shannon diversity index in this application is as follows: Figure 6 As shown, in terms of α-diversity, GCs have significantly higher species richness than NGCs, indicating that GCs have a greater number of species in their tongue coating.

[0074] In addition, we used 80% of the samples as the training set and 20% as the validation set to build the MLP model. Figure 7The ROC (Receiver Operating Characteristic) and AUC (Area Under ROC Curve) of the validation set are shown. It can be seen that the ROC curve of the system in this application is relatively far away from the (0,0)-(1,1) line. It is observed that the AUC value of genus is 0.945 and that of species is 0.975. The AUC values ​​of species / genus are significantly higher than the internal validation AUC values ​​(0.682-0.715) and external validation AUC values ​​(0.694-0.760) of the SVM, DT and KNN models of eight blood tumor markers. It can be seen that the tumor prediction system based on tongue coating microorganisms in this application is a well-performing prediction model, and its diagnostic value for GC is significantly better than the model that simply uses the combination of eight blood tumor markers.

[0075] Comparing the species distribution of the GCs and NGCs groups, the genus-level distribution is as follows: Figure 8 As shown in E, the display at this level is as follows Figure 8 As shown in F, the tumor prediction system based on tongue coating microorganisms in this application can clearly distinguish tongue coating microorganisms at the genus and species level to make targeted diagnostic predictions for GCs, providing a prospective, economical, non-invasive, and effective screening and diagnostic prediction method for tumors.

[0076] The conventional techniques described in the above embodiments are existing technologies known to those skilled in the art, and therefore will not be described in detail here.

[0077] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0078] Although the present invention has been described in detail and specific embodiments have been cited, it will be apparent to those skilled in the art that various changes or modifications can be made without departing from the spirit and scope of the invention.

[0079] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0080] All matters not covered in this invention are common knowledge.

Claims

1. A tongue coating microorganism-based tumor prediction system, characterized by, include: The microbial information acquisition module is used to acquire tongue coating samples from subjects and to acquire abundance information of microorganisms in the tongue coating samples at the genus and / or species levels through gene technology. The data processing module includes a pre-trained multilayer perceptron neural network model, configured to: receive abundance information of tongue coating microorganisms (genus and / or species) acquired by the microbial information acquisition module as an input vector, wherein the length of the input vector is determined based on the number of actually detected microbial genera and / or species, and each vector element represents the relative or absolute abundance of a specific microbial taxonomic unit genus or species; perform nonlinear mapping on the input microbial abundance information through at least one hidden layer to extract high-dimensional discriminative features that can distinguish between positive and negative gastric cancer samples; and output the probability value that the subject sample is predicted to be positive for gastric cancer based on the high-dimensional discriminative features through a Softmax classifier in the output layer. The neural network model is obtained through the following steps: 1) The abundance of tongue microbial genera and / or species collected from tumor-positive patients and / or tumor-negative individuals is used as an input vector and input into the input layer of the model; 2) The hidden layer of the model extracts high-dimensional features of microbial species and / or genus abundance; 3) Output the probability distribution of positive and negative genus and / or species of tongue coating microorganisms through the softmax classifier of the output layer, and minimize the cross-entropy loss function between the predicted probability and the true label through the backpropagation algorithm, and iteratively update the weight parameters of the neural network until the model converges. The system is specifically designed for the auxiliary diagnosis or early screening of gastric cancer, and the system has an accuracy of at least 0.929 in predicting gastric cancer on an independent validation set.

2. The tongue-coating microorganism-based tumor prediction system according to claim 1, characterized by, The method for collecting the tongue coating sample is as follows: the subject is fasting and a sterile tongue swab is used to scrape the tongue coating sample from the root of the tongue to the tip of the tongue.

3. The tongue-coating microorganism-based tumor prediction system according to claim 1 or 2, characterized by, The neurons in each layer of the neural network are interconnected, and the cross-entropy objective function is minimized using the probability distribution of the output: where y i is an element in the true one-hot label vector corresponding to the test example, is the probability that the model predicts class y i , where 0 represents the negative class and 1 represents the positive class.

4. The tongue-coating microorganism-based tumor prediction system according to claim 1, characterized by, The number of hidden layer neurons in the multilayer perceptron neural network model is greater than the number of input layer neurons, and there is at least one hidden layer, with the number of hidden layer neurons configured to be 512.

5. A method for predicting a tumor based on tongue coating microorganisms, characterized by, Includes the following steps: Tongue samples were collected from the subjects, and the abundance information of microorganisms in the tongue samples at the genus and / or species levels was obtained by high-throughput sequencing technology. The input vector formed by the microbial abundance information obtained in the aforementioned steps is input into the data processing module of the pre-trained tumor prediction system as described in any one of claims 1-4. The multilayer perceptron neural network in the data processing module performs high-dimensional feature extraction and calculation on the input vector, and outputs the predicted probability that the subject sample is positive for gastric cancer through the Softmax classifier in its output layer.

6. The tongue-coating microorganism-based tumor prediction method according to claim 5, characterized by, It also includes determining, based on the predicted probability and a preset diagnostic threshold, whether the subject needs further clinical examination for gastric cancer.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tumor prediction method based on tongue coating microorganisms as described in claim 5 or 6.

8. The application of the tumor prediction system based on tongue coating microorganisms as described in any one of claims 1-4 or the tumor prediction method based on tongue coating microorganisms as described in any one of claims 5-6 in the preparation of products for auxiliary diagnosis or early screening of gastric cancer.

Images