Model for predicting possibility of submucosal deep infiltration of colorectal tumor based on molecular characteristics of biopsy sample under endoscope

Abstract

The invention relates to the field of biomedicine, and discloses a method, a system and a kit for constructing a model for predicting the submucosal deep infiltration possibility of colorectal tumor based on molecular characteristics of a biopsy sample under an endoscope, electronic equipment applying the construction method and a computer readable medium storing computer program codes. The construction method comprises the steps of obtaining a plurality of independent variables, carrying out screening and data processing on the independent variables, further screening effective independent variables in a regression model training process, reducing the number of the independent variables, obtaining a relationship among the independent variables through correlation analysis, carrying out replacement and combination by adopting the independent variables with a replacement relationship, and obtaining a regression model. And obtaining a plurality of prediction models. According to the prediction model, a good prediction result can be obtained by using fewer effective independent variables, the prediction efficiency is high, the application range is wide, and the cost is low.

Description

Technical Field

[0001] The present application mainly relates to the fields of biomedicine, biological research and bioinformatics research technology, colorectal cancer, etc., and relates to a model for predicting the possibility of deep submucosal invasion of colorectal tumors based on the molecular characteristics of endoscopic biopsy samples, and specifically to a method for constructing a model for evaluating the possibility of deep submucosal invasion of a subject, a system, a kit, an electronic device using the construction method, and a computer-readable medium storing computer program code. Background Art

[0002] Colorectal cancer (CRC), as one of the most common malignant tumors in my country, its influence and harm cannot be ignored. According to the 2022 China Malignant Tumor Disease Burden Report, there are 517,000 new cases of colorectal cancer in my country, and 240,000 deaths. Its incidence rate ranks second among all malignant tumors in my country, and its mortality rate ranks fourth. This severe situation highlights the huge threat that colorectal cancer poses to the health of Chinese residents, and it is urgent to attract widespread attention and in-depth research from all sectors of society to promote the innovation and development of prevention and control strategies and treatment technologies for colorectal cancer.

[0003] In the early screening of colorectal tumors, colonoscopy plays a vital role, which can detect intestinal tumors early, thus allowing timely therapeutic intervention. For the initial diagnosis of intestinal polyps, accurate judgment of the cancerous state and degree of infiltration of the lesions has a decisive influence on the formulation of subsequent treatment plans. However, it is worth noting that there are often significant differences between the pathological results of endoscopic biopsy and the pathological results after resection, which increases the complexity of diagnosis to a certain extent.

[0004] For colorectal adenomas, endoscopic treatment is highly favored due to its advantages such as less trauma, rapid recovery, and relatively low cost. For the treatment options of early colorectal adenocarcinoma, the depth of submucosal invasion becomes a key decision factor. Specifically, when the depth of submucosal invasion does not exceed 1000μm, it is considered to be shallow submucosal invasion, in which case endoscopic treatment is usually an appropriate choice. However, when the depth of submucosal invasion exceeds 1000μm, that is, it enters the deep invasion stage, comprehensive considerations are required, and surgical intervention may be required to expand the range of resection to ensure the treatment effect.

[0005] Currently, biopsy samples are limited by the limitations of sampling, which makes it difficult for biopsy pathological diagnosis to accurately determine the depth of submucosal invasion. According to research data, the sensitivity of pathological biopsy for deep submucosal invasion is only 3.8%, and this defect is particularly prominent in clinical practice. After many patients receive endoscopic treatment (such as snare resection, endoscopic mucosal resection and endoscopic submucosal dissection), surgical pathological diagnosis reveals deep submucosal invasion, which requires surgical treatment again.

[0006] Therefore, accurate prediction of tumor infiltration depth before surgery is particularly important in the selection of treatment strategies, which can not only reduce the number of unnecessary surgeries and bring substantial benefits to patients, but also contribute to the rational use of medical resources. In view of the need for such clinical treatment decisions, this patent aims to use biopsy samples to establish an efficient and accurate molecular typing method to predict the depth of submucosal invasion. Through this method, physicians can develop more reasonable treatment plans and make wise choices between endoscopic treatment and surgical operations, thereby protecting the interests of patients while also achieving effective savings in medical resources. Summary of the invention

[0007] To solve the above technical problems, the present application provides a model for predicting the possibility of deep submucosal invasion of colorectal tumors based on the molecular characteristics of endoscopic biopsy samples. More specifically, it provides a method, system, and kit for constructing a model for evaluating the possibility of deep submucosal invasion of a subject, as well as an electronic device using the construction method and a computer-readable medium storing computer program code.

[0008] The present application provides, on one hand, a method for constructing a model for evaluating the possibility of deep submucosal infiltration in a subject, comprising:

[0009] Step S10: obtaining multiple independent variables for the model, the multiple independent variables including: the expression levels of microRNA in plasma hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, hsa-miR-1228, and performing structured processing on the multiple independent variables;

[0010] Step S20: Divide the subjects into SM2 group and non-SM2 group, and encode them as 1 and 0 respectively to form the dependent variable for the model;

[0011] Step S30: for the multiple independent variables, calculating the correlation coefficient between each independent variable and the dependent variable, and taking the independent variable with a correlation coefficient greater than a second threshold as a valid independent variable;

[0012] Step S40: performing correlation analysis on the multiple independent variables to determine the relationship between the multiple independent variables;

[0013] Step S50: dividing the data into a training set and a test set, constructing a regression model including the independent variables according to the training set, and testing the performance of the regression model according to the test set, wherein the data includes the multiple independent variables and the dependent variable;

[0014] Step S60: further screening the valid independent variables according to any of the p-value, correlation coefficient and AUC value of the regression model on the training set and the test set of each of the independent variables in the regression analysis result to obtain the screened valid independent variables, wherein when the p-value of the independent variable is less than 0.05, the independent variable whose correlation coefficient is greater than the first threshold is taken as the valid independent variable;

[0015] Step S70: Repeat step S60 until the AUC value of the regression model on the training set decreases to about 0.8, the number of the effective independent variables after screening is greater than or equal to 2, and the effective independent variables after screening include at least one microRNA, to obtain a plurality of first models;

[0016] Step S80: According to the relationship between the multiple independent variables obtained in step S40, the independent variables with replacement relationship in the several first models are replaced and combined, and a regression model is constructed using the replaced and combined independent variables to obtain several second models, wherein the model includes any one of the first model and the second model.

[0017] In a second aspect, the present application provides a system for assessing the possibility of deep submucosal infiltration in a subject, characterized by comprising:

[0018] A data acquisition module, used for obtaining sample data of a subject, wherein the sample data includes any one of the expression amounts of microRNAs in plasma: hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228;

[0019] A data preprocessing module, used to preprocess the sample data, including removing duplicate samples, unifying units, and dividing the sample data into different analysis groups according to indicator types and quantities;

[0020] A model recommendation module, for recommending, for different analysis groups, a plurality of prediction models with larger AUC values ​​from the prediction models established by the construction method described in the first aspect according to the types and numbers of the indicators included therein, according to the sizes of the AUC values;

[0021] A model selection module, used to provide a selection function for the user to output one or more prediction models from the prediction models recommended by the model recommendation module for risk assessment and calculation;

[0022] The risk assessment module is used to calculate the corresponding prediction value according to one or more prediction models output by the model selection module, and to give a suitable Logit threshold or risk score threshold according to the department or population information corresponding to the sample data, and to divide the risk level into any one of high possibility, medium possibility and low possibility according to the Logit threshold or risk score threshold.

[0023] In a third aspect, the present application proposes a kit for evaluating the possibility of deep submucosal infiltration in a subject, comprising a prediction model established using the construction method described above.

[0024] The kit further comprises one or more substances selected from the following: PCR buffer, polymerase, dNTP, restriction endonuclease, enzyme digestion buffer, fluorescent dye, fluorescence quencher, fluorescent reporter, exonuclease, alkaline phosphatase, primer probe, internal standard, and control.

[0025] In a fourth aspect, the present application provides an electronic device, comprising:

[0026] a memory for storing instructions executable by a processor;

[0027] A processor is used to execute the instructions to implement the construction method as described above.

[0028] In a fifth aspect, the present application proposes a computer-readable medium storing a computer program code, wherein the computer program code implements the construction method as described above when executed by a processor.

[0029] Compared with the prior art, this application has the following beneficial effects:

[0030] The prediction model for the possibility of deep submucosal invasion (SM2) established according to the construction method of the present application is based on the data of 198 biopsy tissue samples, and 6 logistic regression prediction models with AUC values ​​between 0.79 and 0.90 on the test set are derived. The detection sensitivity of pathological biopsy for SM2 is only 3.8%, while the detection sensitivity of these models constructed in the present application for SM2 can reach 60-90%. The construction method of the present application screens the independent variables, and in the training process of the regression model, the independent variables are gradually screened, so that the model can obtain better prediction results using fewer effective independent variables, thereby improving the prediction efficiency and saving costs, so that the formed models, test kits, systems, and electronic equipment can be applied to a wider range of scenarios and populations, and have extremely high social value. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings are included to provide a further understanding of the present application. They are included and constitute a part of the present application. The accompanying drawings illustrate embodiments of the present application and together with the present specification serve to explain the principles of the present application. In the accompanying drawings:

[0032] Figure 1 is an exemplary flow chart of a method for constructing a model for evaluating the possibility of deep submucosal infiltration in a subject according to an embodiment of the present application;

[0033] Figure 2 It is a correlation coefficient matrix diagram between various indicators involved in the embodiments of the present application;

[0034] Figure 3 is the ROC curve and the corresponding AUC result of the prediction model 1 of the embodiment of the present application on the training set;

[0035] Figure 4 is the ROC curve and corresponding AUC result of the cross-validation of the prediction model 1 of the embodiment of the present application on the training set;

[0036] Figure 5 is the ROC curve and the corresponding AUC result of the prediction model 1 of the embodiment of the present application on the test set;

[0037] Figure 6 is a block diagram of a system for evaluating the possibility of deep submucosal infiltration in a subject according to an embodiment of the present application;

[0038] Figure 7 It is a system block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0040] In addition, it should be noted that the use of words such as "first" and "second" to define components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above words have no special meaning and cannot be understood as limiting the scope of protection of this application. In addition, although the terms used in this application are selected from well-known and commonly used terms, some terms mentioned in the specification of this application may be selected by the applicant at his or her discretion, and their detailed meanings are explained in the relevant parts of the description of this article. In addition, it is required to understand this application not only by the actual terms used, but also by the meaning implied by each term.

[0041] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously. At the same time, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0042] As used herein, the term "cancer" refers to the presence of cells having typical characteristics of cancer cells, such as uncontrolled proliferation, immortality, metastatic potential, rapid growth and proliferation rate, and certain characteristic morphological features known in the art. In one or more embodiments, "cancer" can be intestinal cancer or intestinal cancer. In one or more embodiments, "cancer" can include pre-malignant cancers as well as malignant cancers.

[0043] In one or more embodiments, as understood by those skilled in the art, the method as described herein does not involve steps performed by a doctor / physician. Therefore, before the final diagnosis performed by a physician can be provided to the subject, the results obtained by the method as described herein need to be combined with clinical data and other clinical manifestations. The final diagnosis of whether the subject suffers from colon cancer is the scope of the physician and is not considered to be part of this disclosure. Therefore, the terms "determine", "detection" and "diagnosis" as used herein refer to the probability or possibility of identifying a subject suffering from a disease (such as colon cancer) at any stage of development or determining the susceptibility of the subject to developing the disease. In one or more embodiments, "diagnosis", "determination" and "detection" are performed before symptoms are manifested. In one or more embodiments, "diagnosis", "determination" and "detection" allow clinicians / physicians (in combination with other clinical manifestations) to confirm colon cancer in subjects suspected of suffering from colon cancer.

[0044] As used herein, the term "sample" means a sample collected from a subject for detecting the type and amount of colorectal cancer markers therein. The sample is from tissue, cells or body fluids, such as colorectal intestinal tissue, colorectal polyps, adenomas, blood, serum or plasma, preferably colorectal polyps. In one or more embodiments, the sample is a colorectal polyp biopsy, preferably a fine needle aspiration biopsy.

[0045] In one or more embodiments, the method further comprises: extraction, quality inspection and / or storage of sample RNA before step S10.

[0046] Some terms used in the embodiments of the present invention are listed below. Within the scope of the present specification and claims, the relevant terms are defined as follows. Other terms not listed are defined as commonly used in the art, and their meanings are well known to those skilled in the art.

[0047] Submucosal invasion: When cancer invades the submucosal layer (SM), it is graded according to different standards, including relative value grading and absolute value grading. The relative value grading method divides the submucosal layer into three equal parts: SM1 means that the invasion depth of the cancer is limited to the upper 1 / 3 of the submucosal layer, SM2 means that the invasion depth of the cancer reaches the middle 1 / 3 of the submucosal layer, and SM3 means that the invasion depth of the cancer reaches the lower 1 / 3 of the submucosal layer.

[0048] Ct value: threshold cycle, an important parameter in real-time fluorescence quantitative PCR detection, which represents the number of cycles required for the PCR reaction to reach the set threshold. Ct refers to the number of PCR cycles at which the fluorescence signal reaches the set threshold. The main factors affecting the Ct value are: 1. The initial copy number of the target DNA: the higher the copy number, the smaller the Ct value. 2. PCR reaction efficiency: the higher the reaction efficiency, the smaller the Ct value. 3. The threshold set for the fluorescence signal: the lower the threshold setting, the larger the Ct value. The main applications of the Ct value are: 1. Quantitative detection of the relative amount of target DNA. 2. Evaluate the quality of the PCR reaction. 3. Quantitative analysis of the relative changes in gene expression. The smaller the Ct value, the higher the number of starting templates of the target DNA. By analyzing the Ct value, experiments such as gene expression quantification, microbial quantification, and residual DNA detection can be performed.

[0049] Correlation Analysis: Correlation Analysis in English is a method used in statistics to study whether there is a relationship between two or more random variables. Its main purpose is to determine whether there is a statistical correlation or dependence between two or more variables, and to try to quantify the degree and form of this correlation. The basic methods of correlation analysis include: linear correlation, rank correlation, distance correlation, etc.

[0050] Correlation coefficient: It is a statistic used to quantify the closeness of the relationship between variables in correlation analysis. It reflects the strength of the linear correlation between two variables. The larger the absolute value, the stronger the linear correlation between the two variables; the closer to 0, the weaker the correlation. Commonly used ones include Pearson correlation coefficient and Spearman correlation coefficient. The correlation coefficient referred to in this study refers to the Pearson correlation coefficient.

[0051] p-value: The full name is Probability Value, which indicates the probability of the observed data appearing in the hypothesis space. Specifically, p-value represents: when the null hypothesis of the hypothesis test is true, the probability of obtaining data equal to or more extreme than the observed result. Generally speaking, if the p-value is very small, such as less than 0.01, it means that the result is extremely unlikely to be a random event under the null hypothesis, so the null hypothesis is rejected, that is, the result is statistically significant. If the p-value is very large, such as greater than 0.05, the null hypothesis cannot be rejected, that is, the result is not statistically significant. The smaller the p-value, the higher the statistical significance of the result. Commonly used significance judgment thresholds are 0.05 and 0.01. Therefore, p-value reflects the probability of observing the current result under the premise that the null hypothesis is true, and is an important basis for judging whether the hypothesis test result is significant. The smaller the p-value, the more significant the result.

[0052] t-test: t-test in English is a statistical method to test whether there is a significant difference between the means of two samples. The basic idea of ​​t-test is to construct a hypothesis, calculate the observed statistic t value, determine the p-value based on the t distribution, and finally determine whether the original hypothesis is established based on the p-value.

[0053] Multicollinearity: Multicollinearity refers to the existence of a strong linear correlation between independent variables. Its main problems are: 1. It affects the accuracy of the least squares estimate and increases the variance of the regression coefficient; 2. It is impossible to accurately estimate the marginal effect of the independent variable on the dependent variable; 3. It will cause the regression coefficients of some independent variables to be insignificant; 4. It will reduce the predictive ability of the regression equation. The methods for judging multicollinearity are: 1. Correlation matrix method: observe the correlation coefficient between independent variables; 2. Variance inflation factor method: If the value of VIF is too large, it means that there is multicollinearity; 3. Condition number method: If the condition number is too large, there is multicollinearity. Methods for dealing with multicollinearity: 1. Increase the sample size; 2. Remove independent variables with multicollinearity; 3. Combine independent variables or use principal components; 4. Regularization methods such as ridge regression. Therefore, multicollinearity will have an adverse effect on the results of regression analysis and needs to be identified and handled during the modeling process.

[0054] Logistic regression: Logistic regression is a commonly used classification model that establishes the relationship between independent variables and categorical dependent variables. The main features of the logistic regression model are: 1. The predicted dependent variable is a binary or multi-class discrete variable; 2. The value of the linear combination of independent variables is converted into a probability between 0 and 1 using the logistic function; 3. It can handle categorical variables and continuous variables; 4. Parameter estimation usually uses the maximum likelihood method; 5. It can explain the influencing factors of classification and the weights of each dependent variable; 6. It can calculate classification probabilities and make category predictions. The main steps to establish a logistic regression model are: (1) Collect data and handle missing values, etc. (2) Select input variables and handle categorical variables. (3) Establish a logistic regression equation. (4) Estimate parameters by maximum likelihood. (5) Evaluate the overall effect of the model. (6) Perform statistical tests to evaluate the impact of each variable. (7) Establish classification rules through probability. (8) Make predictions for new data. The logistic regression model can not only quantitatively analyze the effects of variables, but also be used for classification prediction. It is a very useful classification analysis method.

[0055] ROC curve and AUC value: Both are used to measure the quality of classifiers. The ROC (Receiver Operating Characteristic) curve is the receiver operating characteristic curve. The horizontal axis is the false positive rate and the vertical axis is the true positive rate. The curve drawn should be above the y=x line. The area under the ROC curve is the AUC value. The larger the AUC, the better the classification performance of the classifier (such as the logistic regression model).

[0056] The present application is further described below with reference to the embodiments and drawings.

[0057] Example 1: Method for constructing a model for assessing the likelihood of deep submucosal invasion in a subject

[0058] Figure 1 FIG. 1 is an exemplary flow chart of a method for constructing a model for evaluating the possibility of deep submucosal infiltration in a subject according to an embodiment of the present application. Figure 1 The construction method of this embodiment includes the following steps:

[0059] Step S10: obtaining multiple independent variables for the model, the multiple independent variables including: the expression levels of microRNA in plasma hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, hsa-miR-1228, and performing structured processing on the multiple independent variables;

[0060] Step S20: Divide the subjects into SM2 group and non-SM2 group, and encode them as 1 and 0 respectively to form the dependent variable for the model;

[0061] Step S30: for the multiple independent variables, calculating the correlation coefficient between each independent variable and the dependent variable, and taking the independent variable with a correlation coefficient greater than a second threshold as a valid independent variable;

[0062] Step S40: performing correlation analysis on the multiple independent variables to determine the relationship between the multiple independent variables;

[0063] Step S50: dividing the data into a training set and a test set, constructing a regression model including the independent variables according to the training set, and testing the performance of the regression model according to the test set, wherein the data includes the multiple independent variables and the dependent variable;

[0064] Step S60: further screening the valid independent variables according to any of the p-value, correlation coefficient and AUC value of the regression model on the training set and the test set of each of the independent variables in the regression analysis result to obtain the screened valid independent variables, wherein when the p-value of the independent variable is less than 0.05, the independent variable whose correlation coefficient is greater than the first threshold is taken as the valid independent variable;

[0065] Step S70: Repeat step S60 until the AUC value of the regression model on the training set decreases to about 0.8, the number of the effective independent variables after screening is greater than or equal to 2, and the effective independent variables after screening include at least one microRNA, to obtain a plurality of first models;

[0066] Step S80: According to the relationship between the multiple independent variables obtained in step S40, the independent variables with replacement relationship in the first models are replaced and combined, and regression models are constructed using the replaced and combined independent variables to obtain a plurality of second models, wherein the models include any of the first model and the second model.

[0067] The arrangement order and numbering of the above steps S10 to S80 are not used to limit the execution order of each step. For example, the execution order of the above steps S10 and S20, and steps S30 to S50 can be arbitrary.

[0068] The inventors of the present application have studied the screening of existing colorectal cancer biomarkers and the simplification of models, and found that if the existing colorectal cancer biomarkers are structured to form independent variables, and then the samples are grouped according to clinical diagnosis and encoded to form dependent variables, then the variables can be screened through correlation analysis and statistical tests between the independent variables and the dependent variables. The ability of an independent variable or a colorectal cancer biomarker to distinguish different clinical groups is judged by the size of the correlation coefficient between the independent variable and the dependent variable. The larger the correlation coefficient with the dependent variable, the stronger the ability of the independent variable or the colorectal cancer biomarker to distinguish different clinical groups, and the more it needs to be given priority in constructing a colorectal cancer prediction model. Whether there is a significant difference in the numerical distribution of the independent variable between the two clinical groups and the size of the p-value can also be used to judge the ability of the independent variable / colorectal cancer biomarker to distinguish the dependent variable or different clinical groups. At the same time, the inventors also found that through the correlation analysis between independent variables or colorectal cancer biomarkers, the correlation between independent variables or colorectal cancer biomarkers can be found. Two strongly correlated independent variables or colorectal cancer biomarkers can replace each other in model construction or practical applications, and only one of them needs to be selected in the model construction, so as to achieve the effect of streamlining the model. In addition, the inventors also found that if a logistic regression model of all variables including valid variables is constructed on the training set, and the performance of the model is tested on the test set, the p-value, coefficient value and AUC value of the model as a whole of each variable in the logistic regression result can be combined with the above-mentioned correlation analysis results and statistical test results. The variables can be further screened and the set of valid variables can be narrowed, thereby obtaining extremely excellent colorectal cancer marker screening effects and model streamlining effects. Based on the above findings, the present application proposes a method for constructing a model for evaluating the possibility of deep submucosal infiltration in subjects.

[0069] The data source for the construction method of this application is the following information from 198 patients admitted to hospitals in two different regions:

[0070] 1) microRNA expression results: expression levels of 13 microRNAs in plasma, including hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, and hsa-miR-1228;

[0071] 2) Clinical diagnosis results.

[0072] After obtaining the above data, the data was preprocessed, including:

[0073] 1) Eliminate samples with incomplete information, duplicate samples, etc.;

[0074] 2) Variable structured processing, i.e., structured processing of multiple independent variables in step S10. For example, samples are divided into SM2 group and non-SM2 group according to clinical diagnosis, and coded as 1 and 0 respectively to form dependent variables, such as step S20; finally, the processed independent variables and dependent variables are organized into a two-dimensional table or data matrix, where each column represents a variable and each row represents a sample, so as to facilitate the subsequent use of this information to establish a prediction model.

[0075] The objects processed in steps S30 to S50 are all the multiple independent variables. The method for screening the multiple independent variables in step S30 to obtain valid independent variables may include the following method 1 and / or method 2:

[0076] Method 1: Calculate the correlation coefficient between each independent variable and the dependent variable, and take the independent variable with a correlation coefficient greater than the second threshold as a valid independent variable;

[0077] Method 2: Perform t-test statistical analysis on each independent variable between the SM2 group and the non-SM2 group, and take the independent variable whose p-value is less than the third threshold as the valid independent variable.

[0078] The present application does not limit the values ​​of the second threshold, the third threshold, and the first threshold in step S70, which can be determined according to actual data distribution. For example, the first threshold = 0.8, the second threshold = 0.8, and the third threshold = 0.05.

[0079] In step S40, correlation analysis is performed on all the independent variables in step S10, i.e., hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, and hsa-miR-1228, as well as the clinical sample groups. The results are shown in FIG. Figure 2 shown.

[0080] Figure 2It is a correlation coefficient matrix diagram between the various indicators involved in the embodiments of the present application. Among them, the independent variables of the diagonal of the matrix from the upper left corner to the lower right corner are: hsa-miR-150 (illustrated as miR.150), cel-miR-39 (illustrated as cel.miR.39), Class.Y, hsa-miR-129 (illustrated as miR.129), hsa-miR-1228 (illustrated as miR.1228), hsa-miR-378a (illustrated as miR.378a), hsa-miR-195 (illustrated as miR.195), hsa-miR-214 (illustrated as miR.214), hsa-miR-193a (illustrated as miR.193a), hsa-miR-584 (illustrated as miR.584), U6 snRNA (U6snRNA in the figure), hsa-miR-345 (miR.345 in the figure), hsa-miR-16 (miR.16 in the figure), hsa-miR-21 (miR.21 in the figure). Among them, Class.Y represents the dependent variable or clinical sample grouping. The value in each small square in the figure represents the correlation coefficient between the two indicators corresponding to its horizontal and vertical coordinates. The closer the absolute value of the correlation coefficient is to 1, the higher the correlation is. The correlation coefficients on the diagonal are all 1. For example, the number in the small square in the first column of the second row is 0.19, indicating that the correlation coefficient between hsa-miR-150 and cel-miR-39 is 0.19, and the correlation between the two is very low. Figure 2 The color bar on the right represents the correlation coefficient corresponding to the small squares in the figure. The white in the middle is 0, indicating no correlation; the blue at the upper end is 1, indicating a complete positive correlation; the red at the lower end is -1, indicating a complete negative correlation.

[0081] like Figure 2As shown, the correlation between hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, hsa-miR-1228 and clinical diagnosis group (Class.Y) is as follows from high to low: hsa-miR-150=hsa-miR-378a>cel-miR-39>hsa-miR-195>hsa-miR-1228>hsa-miR-16>hsa-miR-584>hsa-miR-345>hsa-miR-193a=hsa-miR-21>U6 snRNA>hsa-miR-129>hsa-miR-214. The top five microRNAs most relevant to clinical diagnosis grouping are hsa-miR-150, hsa-miR-378a, cel-miR-39, hsa-miR-195 and hsa-miR-1228, which are particularly highly correlated, several times or dozens of times that of other microRNAs. This result suggests that hsa-miR-150, hsa-miR-378a, cel-miR-39, hsa-miR-195 and hsa-miR-1228 should be given priority in subsequent modeling. At the same time, hsa-miR-129 and hsa-miR-214, which have the weakest correlation with clinical diagnosis grouping, are excluded first in the subsequent marker streamlining.

[0082] like Figure 2As shown, the correlation between different microRNAs can also be obtained. The correlation coefficients between hsa-miR-16, hsa-miR-21 and hsa-miR-345 are greater than 0.9, so there is a strong correlation between the three microRNAs and they can replace each other; the correlation coefficients between hsa-miR-378a, hsa-miR-195 and hsa-miR-214 are at least 0.72, and hsa-miR-16, hsa-miR-21 and hsa-miR-345 are related to hsa-miR-214, hsa-miR-193a, hsa-miR-584 and U6 There is a strong correlation between snRNAs, and there is a certain degree of substitution relationship; the correlation coefficient between hsa-miR-1228 and other microRNAs is at most 0.60, so there is a low or moderate correlation with other microRNAs, but no substitution relationship; the correlation coefficient between hsa-miR-129 and other microRNAs is at most 0.59, so there is at most a moderate correlation with other microRNAs, but no substitution relationship; there is a low or moderate correlation between hsa-miR-150 and other microRNAs, but no substitution relationship; there is a low or moderate correlation between cel-miR-39 and other microRNAs, but no substitution relationship.

[0083] According to the above correlation analysis, firstly, we can find out which microRNAs are the most effective or the least effective in distinguishing different clinical diagnostic groups; secondly, we can understand whether there is a strong correlation or substitution relationship between the key microRNAs. Among the microRNAs with a strong correlation or substitution relationship, we can select one of them in the modeling process, so as to further simplify the model so that the model can achieve the greatest differentiation effect with the least parameters.

[0084] Similarly, by analyzing the correlation between the independent variables and the clinical diagnostic groups (Class.Y), we can find out which hematological indicators are most effective or ineffective in distinguishing different clinical diagnostic groups, so as to further streamline the model so that the model can achieve the greatest differentiation effect with the least parameters.

[0085] In step S30, method 2 is adopted, that is, a t-test statistical analysis is performed on each independent variable between the SM2 group and the non-SM2 group, and the independent variable whose p-value of the analysis result is less than the third threshold is regarded as a valid independent variable. The results are as follows:

[0086] The geom_signif() function in R language was used to perform t-test statistical analysis on the expression levels of different microRNAs in the SM2 group and the non-SM2 group. The specific steps include: 1. Input the expression data of a certain microRNA in the SM2 group and the non-SM2 group as two samples of the t-test; 2. Calculate the mean and standard deviation of the two samples; 3. Assuming that the two samples are independent, calculate the degrees of freedom; 4. Calculate the t statistic; 5. Find the corresponding P value according to the t distribution table; 6. If the P value is less than the significance level (default 0.05), the difference is considered to be statistically significant, and the smaller the p-value of the statistical test result, the greater the effect of the independent variable on the sample classification. The final results of the t.test statistical test for all microRNAs are shown in Table 1. From the numerical value of p-value, except for the exogenous reference cel-miR-39, hsa-miR-129, U6snRNA and hsa-miR-214 are the three microRNAs with the weakest ability to distinguish between the SM2 group and the non-SM2 group, while hsa-miR-150, hsa-miR-1228 and hsa-miR-378a are the three microRNAs with the strongest ability to distinguish between the SM2 group and the non-SM2 group.

[0087] Table 1 t-test results of different microRNAs between SM2 group (SM2) and non-SM2 group (non-SM2)

[0088]

[0089]

[0090] In step S70, when using multinomial logistic regression modeling, it is necessary to avoid multicollinearity between variables, otherwise it will affect the stability of the model. According to the above analysis results and conclusions: There is a strong correlation between the three microRNAs hsa-miR-16, hsa-miR-21 and hsa-miR-345, so these three variables are not suitable for modeling together. In order to more comprehensively evaluate the distinguishing effect of each independent variable on the dependent variable, although there are different degrees of correlation between other independent variables, they are no longer considered here.

[0091] Take hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 The variables snRNA, cel-miR-39, hsa-miR-16, hsa-miR-21, and hsa-miR-1228 were used as independent variables and the dependent variables to construct a multivariate logistic regression model 0-1. The OR value, p-value, and 95% confidence interval of the OR value of each variable were sorted in Table 2. The table content was sorted from small to large according to the p-value. The more the variables were ranked, the more they needed to be given priority in subsequent modeling. The top five variables were hsa-miR-1228, hsa-miR-195, hsa-miR-214, hsa-miR-378a, and U6 snRNA according to the p-value.

[0092] Table 2 Significance analysis results of different variables in the multivariate logistic regression model

[0093]

[0094]

[0095] According to the above step S70, 5 first models are obtained. On this basis, according to the relationship between the multiple independent variables obtained in step S40, the independent variables with replacement relationships in the first model are replaced and combined, and the regression model is constructed using the replaced and combined independent variables to obtain multiple second models. The present application provides one of the second models. Therefore, the present application obtains a total of 6 models, and the following is combined with Examples 2 to 7 to illustrate the 6 models one by one. It should be noted that the training set and test set used in each model are the same. For each model, the receiver operating characteristic (ROC) curve and the area under the curve (AUC) are used as evaluation indicators, and these indicators are calculated on the training set, validation set, and test set. The ROC curve can intuitively reflect the classification effect of the model by comprehensively judging the recognition ability of the model for positive and negative samples; and the AUC value can quantify the ROC curve and further judge the sensitivity and specificity of the model prediction. Performing ROC and AUC analysis can comprehensively evaluate the effect of the prediction model we constructed, and provide a basis for subsequent model optimization and clinical application.

[0096] Example 2: Establishment and verification of logistic regression prediction model 1

[0097] On the training set, a logistic regression model including two independent variables, hsa-miR-378a and hsa-miR-1228, was constructed to predict the possibility of deep submucosal infiltration of the sample. First, based on the expression data of hsa-miR-378a and hsa-miR-1228 and the sample labels, the prediction model 1 was obtained by the logistic regression algorithm. The mathematical expression of the prediction model 1 is: Logit(P) = β1×hsa-miR-378a+β2×hsa-miR-1228+β0, where Logit(P) represents the logarithmic probability of the sample being classified as a positive example, hsa-miR-378a and hsa-miR-1228 represent the expression levels of the corresponding microRNAs, β1~β2 are the regression coefficients corresponding to these microRNAs, and β0 is the intercept term. After model training, we obtained the value range of coefficient β1 as [0.47, 1.50]; the value range of coefficient β2 as [-2.51, -0.86]; the value range of coefficient β0 as [-2.14, 40.10]. Preferably, β1 is 0.98, β2 is -1.68, and β0 is 18.98, which can obtain the optimal prediction result. The established logistic regression model can be used to predict the possibility of deep submucosal infiltration of new samples based on the expression results of hsa-miR-378a and hsa-miR-1228.

[0098] The ROC curve and corresponding AUC results of prediction model 1 on the training set are as follows: Figure 3 As shown in the figure, the ROC curve of the model on the training set and the corresponding AUC result show that the ROC curve of the model shows a good overall effect and a high AUC value of 0.80, which indicates that the model has good predictive performance in terms of differentiation.

[0099] In order to fully verify the stability and generalization ability of the model, the 5-fold cross-validation method was used for internal cross-validation of the model. Specifically, the training set data was divided into 5 mutually exclusive subsets, 4 of which were used as training sets each time, and the other subset was used as the validation set, for a total of 5 rounds of training and validation. The 5-fold cross-validation can measure the stability of the model on different data subsets. The ROC curve and corresponding AUC results of the 5-fold cross-validation of the model on the training set are shown in Figure 2. Figure 4 As shown in the figure, it can be seen from the ROC curve of the 5-fold cross-validation and the corresponding AUC result that after 5-fold cross-validation, the final model obtained an average AUC of 0.79 and a standard deviation of 0.09 on the validation set, achieving the set performance target. It also proves that the model is stable and effective, providing a guarantee for the practical application of the model.

[0100] After obtaining satisfactory training set results, prediction model 1 is applied to an independent test set for prediction, and the various evaluation indicators of the model on the test set are calculated. This process can verify the generalization ability of the model from the training set to the test set, and check whether the model has overfitting problems on unseen samples, so as to further evaluate the generalization performance of the model. The ROC curve and corresponding AUC results of the prediction model 1 on the test set are shown in Figure 2. Figure 5 As shown in the figure, the ROC curve of prediction model 1 on the test set and the corresponding AUC result show that the ROC curve of prediction model 1 shows that the overall effect of the model on the test set is good, and the corresponding AUC value is also high, which is 0.79. On the one hand, this shows that prediction model 1 still maintains good discrimination ability on the test set. On the other hand, the AUC of the test set is comparable to the AUC of the training set, indicating that the model is not overfitting and has good generalization ability.

[0101] After processing the data using the caret package in R, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training and test sets at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index at each cutoff and selected the cutoff value that made the specificity reach more than 60% and the Youden index reach the maximum as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 70.0%, the sensitivity was 75.0%, and the specificity was 68.75%.

[0102] Similarly, on the test set data, the prediction accuracy was 66.0%, the sensitivity was 84.9%, and the specificity was 61.6%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 1 on the test set is good and can be clinically verified and applied. According to prediction model 1, only the two indicators of hsa-miR-378a and hsa-miR-1228 of the subject need to be obtained to accurately predict the possibility of deep submucosal infiltration, which is low cost and high efficiency.

[0103] In Examples 3 to 7, the same method as in Example 2 was used to obtain the ROC curve and the corresponding AUC result on the training set, the ROC curve and the corresponding AUC result of the 5-fold cross validation, the ROC curve and the corresponding AUC result on the test set, and the related ROC curve graphs, which were used to evaluate the prediction performance of prediction models 2 to 6. The following will only describe these results without combining the accompanying drawings, and the same content will not be repeated.

[0104] Example 3: Establishment and verification of logistic regression prediction model 2

[0105] On the training set, a logistic regression model including three independent variables, hsa-miR-150, hsa-miR-378a, and hsa-miR-1228, was constructed to predict the possibility of deep submucosal invasion of the samples. First, based on the expression data of hsa-miR-150, hsa-miR-378a, and hsa-miR-1228 and the labels of the samples, prediction model 2 was obtained by the logistic regression algorithm. The mathematical expression of prediction model 2 is: Logit(P) = β1×hsa-miR-150+β2×hsa-miR-378a+β3×hsa-miR-1228+β0, where Logit(P) represents the logarithmic probability of the sample being classified as a positive example, hsa-miR-150, hsa-miR-378a and hsa-miR-1228 represent the expression levels of the corresponding microRNAs, β1 to β3 are the regression coefficients corresponding to these microRNAs, and β0 is the intercept term. After model training, we obtained that the value range of coefficient β1 is [-1.43, 0.32]; the value range of coefficient β2 is [0.27, 1.42]; the value range of coefficient β3 is [-2.31, -0.44]; the value range of coefficient β0 is [1.31, 62.80]. Preferably, β1 is -0.55, β2 is 0.84, β3 is -1.38, and β0 is 32.06, and the optimal prediction result can be obtained. The established logistic regression model can be used to predict the possibility of deep submucosal infiltration of new samples based on the expression results of hsa-miR-150, hsa-miR-378a and hsa-miR-1228 in new samples.

[0106] The ROC curve of prediction model 2 on the training set showed good overall effect, and the AUC value was also high, which was 0.80, indicating that the model had good predictive performance in terms of differentiation. The ROC curve of the 5-fold cross-validation of prediction model 2 on the training set had an average AUC of 0.79 and a standard deviation of 0.07, which achieved the set performance target and proved that the model was stable and effective, providing a guarantee for the practical application of the model. The ROC curve of prediction model 2 on the test set had a good overall effect, and the corresponding AUC value was also high, which was 0.79. On the one hand, it showed that the model still maintained a good ability to distinguish SM2 samples from non-SM2 samples on the test set. On the other hand, the AUC of the test set was comparable to the AUC of the training set, indicating that the model was not overfitted and had good generalization ability. In summary, the model had good predictive performance on the test set and could be clinically verified and applied.

[0107] After processing the data using the caret package in R, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training set and test set at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index under each cutoff and selected the cutoff value that made the specificity reach more than 60% and the Youden index reach the maximum as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 67.0%, the sensitivity was 84.85%, and the specificity was 62.4%. Similarly, on the test set data, the prediction accuracy was 82.5%, the sensitivity was 75.0%, and the specificity was 84.38%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 2 on the test set is good and can be clinically verified and applied. According to prediction model 2, only three indicators, hsa-miR-150, hsa-miR-378a and hsa-miR-1228, are needed to accurately predict the possibility of deep submucosal infiltration, which is low-cost and high-efficiency.

[0108] Example 4: Establishment and verification of logistic regression prediction model 3

[0109] On the training set, a logistic regression model including four independent variables, hsa-miR-150, hsa-miR-195, hsa-miR-378a, and hsa-miR-1228, was constructed to predict the possibility of deep submucosal invasion of the samples. First, based on the expression data of hsa-miR-150, hsa-miR-195, hsa-miR-378a, and hsa-miR-1228 and the labels of the samples, the prediction model 3 was obtained by the logistic regression algorithm. The mathematical expression of prediction model 3 is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-1228+β0, where Logit(P) represents the logarithmic probability of the sample being classified as a positive example, hsa-miR-150, hsa-miR-195, hsa-miR-378a and hsa-miR-1228 represent the expression levels of the corresponding microRNAs, β1~β4 are the regression coefficients corresponding to these microRNAs, and β0 is the intercept term. After model training, we obtained the value range of coefficient β1 as [-1.42, 0.33]; the value range of coefficient β2 as [-0.41, 0.54]; the value range of coefficient β3 as [0.09, 1.50]; the value range of coefficient β4 as [-2.33, -0.45]; the value range of coefficient β0 as [0.79, 62.57]. Preferably, β1 is -0.55, β2 is 0.06, β3 is 0.79, β4 is -1.39, and β0 is 31.68, which can obtain the optimal prediction result. The established logistic regression model can be used to predict the expression of hsa-miR-150, hsa-miR-195, hsa-miR-378a and hsa-miR-1228 in new samples, and judge the possibility of deep submucosal infiltration of new samples.

[0110] The ROC curve of prediction model 3 on the training set showed good overall effect, and the AUC value was also high, which was 0.79, indicating that the model had good predictive performance in terms of differentiation. The ROC curve of the 5-fold cross-validation of prediction model 3 on the training set had an average AUC of 0.79 and a standard deviation of 0.10, which achieved the set performance target and proved that the model was stable and effective, providing a guarantee for the practical application of the model. The ROC curve of prediction model 3 on the test set had a good overall effect, and the corresponding AUC value was also high, which was 0.80. On the one hand, it showed that the model still maintained a good ability to distinguish SM2 samples from non-SM2 samples on the test set. On the other hand, the AUC of the test set was comparable to the AUC of the training set, indicating that the model was not overfitted and had good generalization ability. In summary, the model had good predictive performance on the test set and could be clinically verified and applied.

[0111] After processing the data using the caret package in R language, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training set and test set at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index under each cutoff and selected the cutoff value that made the specificity reach more than 60% and the Youden index reach the maximum as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 68.0%, the sensitivity was 81.82%, and the specificity was 64%. Similarly, on the test set data, the prediction accuracy was 82.5%, the sensitivity was 75.0%, and the specificity was 84.38%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 3 on the test set is good and can be clinically verified and applied. According to prediction model 3, only four indicators, namely hsa-miR-150, hsa-miR-195, hsa-miR-378a and hsa-miR-1228, are needed to accurately predict the possibility of deep submucosal infiltration, which is low-cost and high-efficiency.

[0112] Example 5: Establishment and verification of logistic regression prediction model 4

[0113] On the training set, a logistic regression model including five independent variables, hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16, and hsa-miR-1228, was constructed to predict the possibility of deep submucosal invasion of the samples.First, based on the expression data of hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16, and hsa-miR-1228 and the labels of the samples, the prediction model 4 was obtained by the logistic regression algorithm. The mathematical expression of prediction model 4 is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-16+β5×hsa-miR-1228+β0, where Logit(P) represents the logarithmic probability of the sample being classified as a positive example, hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16 and hsa-miR-1228 represent the expression levels of the corresponding microRNAs, β1~β5 are the regression coefficients corresponding to these microRNAs, and β0 is the intercept term. After model training, we get the value range of β1 is [-1.43, 0.34]; the value range of coefficient β2 is [-0.43, 0.56]; the value range of coefficient β3 is [-0.06, 1.66]; the value range of coefficient β4 is [-0.64, 0.61]; the value range of coefficient β5 is [-2.34, -0.44]; the value range of coefficient β0 is [-1.83, 64.62]. Preferably, β1 is -0.55, β2 is 0.07, β3 is 0.80, β4 is -0.01, β5 is -1.39, and β0 is 31.40, which can obtain the optimal prediction result. The established logistic regression model can be used to predict the possibility of deep submucosal infiltration of new samples based on the expression results of hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16 and hsa-miR-1228.

[0114] The ROC curve of prediction model 4 on the training set shows good overall effect, and the AUC value is also high, which is 0.80, indicating that the model has good predictive performance in terms of differentiation. The ROC curve of the 5-fold cross-validation of prediction model 4 on the training set has an average AUC of 0.78 and a standard deviation of 0.07, which achieves the set performance target. It also proves that the model is stable and effective, providing a guarantee for the practical application of the model. The ROC curve of prediction model 4 on the test set has a good overall effect, and the corresponding AUC value is also high, which is 0.80. On the one hand, it shows that the model still maintains a good ability to distinguish SM2 samples from non-SM2 samples on the test set. On the other hand, the AUC of the test set is comparable to the AUC of the training set, indicating that the model is not overfitting and has good generalization ability. In summary, the model has good predictive performance on the test set and can be clinically verified and applied.

[0115] After processing the data using the caret package in R, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training set and test set at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index under each cutoff and selected the cutoff value that made the specificity reach more than 60% and the Youden index reach the maximum as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 73.4%, the sensitivity was 72.7%, and the specificity was 73.6%. Similarly, on the test set data, the prediction accuracy was 87.5%, the sensitivity was 75.0%, and the specificity was 90.6%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 4 on the test set is good and can be clinically verified and applied. According to prediction model 4, only five indicators, namely hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16 and hsa-miR-1228, are needed to accurately predict the possibility of deep submucosal infiltration, which is low-cost and highly efficient.

[0116] Example 6: Establishment and verification of logistic regression prediction model 5

[0117] On the training set, a logistic regression model including five independent variables, hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21, and hsa-miR-1228, was constructed to predict the possibility of deep submucosal invasion of the samples.First, based on the expression data of hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21, and hsa-miR-1228 and the labels of the samples, the prediction model 5 was obtained by the logistic regression algorithm. The mathematical expression of prediction model 5 is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-21+β5×hsa-miR-1228+β0, where Logit(P) represents the logarithmic probability of the sample being classified as a positive example, hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21 and hsa-miR-1228 represent the expression levels of the corresponding microRNAs, β1~β5 are the regression coefficients corresponding to these microRNAs, and β0 is the intercept term. After model training, we get the value range of coefficient β1 is [-1.45, 0.30]; the value range of coefficient β2 is [-0.43, 0.53]; the value range of coefficient β3 is [-0.07, 1.47]; the value range of coefficient β4 is [-0.32, 0.58]; the value range of coefficient β5 is [-2.47, -0.47]; the value range of coefficient β0 is [2.16, 67.30]. Preferably, β1 is -0.58, β2 is 0.05, β3 is 0.70, β4 is 0.13, β5 is -1.47, and β0 is 34.73, which can obtain the optimal prediction result. The established logistic regression model can be used to predict the possibility of deep submucosal infiltration of new samples based on the expression results of hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21 and hsa-miR-1228.

[0118] The ROC curve of prediction model 5 on the training set shows a good overall effect, and the AUC value is also high, which is 0.80, indicating that the model has good predictive performance in terms of differentiation. The ROC curve of the 5-fold cross-validation of prediction model 5 on the training set corresponds to an average AUC of 0.80 and a standard deviation of 0.10, which achieves the set performance target. It also proves that the model is stable and effective, providing a guarantee for the practical application of the model. The ROC curve of prediction model 5 on the test set has a good overall effect, and the corresponding AUC value is also high, which is 0.80. On the one hand, it shows that the model still maintains a good ability to distinguish SM2 samples from non-SM2 samples on the test set. On the other hand, the AUC of the test set is comparable to the AUC of the training set, indicating that the model is not overfitting and has good generalization ability. In summary, the model has good predictive performance on the test set and can be clinically verified and applied.

[0119] After processing the data using the caret package in R, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training set and test set at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index under each cutoff and selected the cutoff value that made the specificity reach more than 60% and the Youden index reach the maximum as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 66.5%, the sensitivity was 90.9%, and the specificity was 60.0%. Similarly, on the test set data, the prediction accuracy was 80.0%, the sensitivity was 75.0%, and the specificity was 81.3%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 5 on the test set is good and can be clinically verified and applied. According to prediction model 5, only five indicators, namely hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21 and hsa-miR-1228, are needed to accurately predict the possibility of deep submucosal infiltration, which is low-cost and high-efficiency.

[0120] Example 7: Establishment and verification of logistic regression prediction model 6

[0121] Constructed from hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 The logistic regression prediction model composed of 12 microRNAs, namely, snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228, the mathematical expression of prediction model 6 is: Logit(P)=β1×hsa-miR-129+β2×hsa-miR-150+β3×hsa-miR-193a+β4×hsa-miR-195+β5×hsa-miR-214+β6×hsa-miR-345+β7×hsa-miR-378a+β8×hsa-miR-584+β9×U6 snRNA+β10×hsa-miR-16+β11×hsa-miR-21+β12×hsa-miR-1228+β0, where the value range of coefficient β1 is [-1.08, 0.58]; the value range of coefficient β2 is [-1.55, 0.57]; the value range of coefficient β3 is [-0.38, 1.05]; the value range of coefficient β4 is [-0.06, 1.29]; the value range of coefficient β5 is [-1.69, 0.16]; the value range of coefficient β6 is The value range of coefficient β7 is [-0.27, 1.74]; the value range of coefficient β8 is [-0.49, 1.35]; the value range of coefficient β9 is [-1.20, 0.23]; the value range of coefficient β10 is [-2.30, 0.99]; the value range of coefficient β11 is [-0.76, 1.89]; the value range of coefficient β12 is [-2.93, -0.25]; the value range of coefficient β0 is [-8.21, 71.15]. Preferably, β1 is -0.25, β2 is -0.49, β3 is 0.34, β4 is 0.61, β5 is -0.77, β6 is 0.32, β7 is 0.74, β8 is 0.43, β9 is -0.49, β10 is -0.65, β11 is 0.57, β12 is -1.59, and β0 is 31.47, which can obtain the best prediction result. The established logistic regression model can be used to predict the expression results of all microRNAs in new samples and judge the possibility of deep submucosal infiltration of new samples.

[0122] The ROC curve of prediction model 6 on the training set showed good overall effect, and the AUC value was also high, which was 0.84, indicating that the model had good predictive performance in terms of differentiation. The ROC curve of the 5-fold cross-validation of prediction model 6 on the training set had an average AUC of 0.83 and a standard deviation of 0.10, which achieved the set performance target and proved that the model was stable and effective, providing a guarantee for the practical application of the model. The ROC curve of prediction model 6 on the test set had a good overall effect, and the corresponding AUC value was also high, which was 0.80. On the one hand, it showed that the model still maintained a good ability to distinguish SM2 samples from non-SM2 samples on the test set. On the other hand, the AUC of the test set was comparable to the AUC of the training set, indicating that the model was not overfitted and had good generalization ability. In summary, the model had good predictive performance on the test set and could be clinically verified and applied.

[0123] After processing the data using the caret package in R language, we obtained the prediction accuracy, sensitivity, and specificity of the model for the training set and test set at different cutoff values ​​(from 0.1 to 1.0). In order to determine the optimal cutoff value, we calculated the Youden index under each cutoff and selected the cutoff value that maximized the Youden index as the optimal setting. After determining the optimal cutoff value, we evaluated the training set data, and the results showed that the prediction accuracy was 83.0%, the sensitivity was 81.8%, and the specificity was 83.2%. Similarly, on the test set data, the prediction accuracy was 85.0%, the sensitivity was 62.5%, and the specificity was 90.6%. These indicators show that under the selected optimal cutoff, the model showed good performance on both training and test data. In summary, the prediction performance of prediction model 6 on the test set is good and can be clinically verified and applied. According to prediction model 6, only 12 indicators, including hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228, are needed to accurately predict the possibility of deep submucosal infiltration, which is low-cost and high-efficiency.

[0124] The above six models involve different indicators and can be applied to multiple different classification groups obtained according to indicators and quantities. Preferably, prediction model 1 uses at least two indicators to obtain accurate prediction results, which is simple and low-cost.

[0125] In some embodiments, Figure 1After step S80 shown, the process also includes adjusting the Logit threshold, risk score threshold or cutoff value of the model prediction result to ensure the detection accuracy of the model in different populations.

[0126] Figure 6 is a block diagram of a system for evaluating the possibility of deep submucosal infiltration in a subject according to an embodiment of the present application. Figure 6 As shown, the system of this embodiment includes a data collection module 610 , a data preprocessing module 620 , a model recommendation module 630 , a model selection module 640 and a risk assessment module 650 .

[0127] Among them, the data acquisition module 610 is used to obtain sample data of the subject, and the sample data includes the expression level of microRNA in plasma hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228.

[0128] The data preprocessing module 620 is used to preprocess the sample data, including removing duplicate samples, unifying units, and dividing the sample data into different analysis groups according to indicator types and quantities.

[0129] The model recommendation module 630 is used to recommend multiple prediction models with larger AUC values ​​from the six prediction models established by the construction method described above according to the types and numbers of indicators included in different analysis groups. The model formula with the largest AUC value is recommended first, and finally displayed in the form of a model formula list on the user operation interface, sorted from large to small according to the AUC value of the model.

[0130] The model selection module 640 is used to provide the user with a selection function to output one or more prediction models for risk assessment and calculation from the prediction models recommended by the model recommendation module 630. The user does not need to make a selection, and the system will select the model with the largest AUC value by default.

[0131] The risk assessment module 650 is used to calculate the corresponding predicted value according to one or more prediction models output by the model selection module 640, and to give a suitable Logit threshold or risk score threshold according to the department or population information corresponding to the sample data, and to divide the risk level into any one of high possibility, medium possibility, and low possibility according to the Logit threshold or risk score threshold. In some embodiments, the Logit threshold or risk score threshold includes 1 threshold for binary division, that is, dividing into high possibility and low possibility. In some embodiments, the Logit threshold or risk score threshold includes 2 thresholds for tridivision, that is, dividing into high possibility, medium possibility, and low possibility. In other embodiments, further subdivisions can be made, and this application does not limit this.

[0132] System 600 can quantitatively obtain biomarker information in the subject's sample and input it into the model for SM2 risk assessment, avoiding errors in subjective judgment, and achieving effective assessment and risk prediction of the possibility of deep submucosal infiltration in different populations. Unlike existing colorectal cancer screening systems, this system 600 can not only quantitatively, automatically, and continuously give SM2 risk levels, improving the accuracy and efficiency of assessment; it also provides a rich model formula containing different types of indicators. Considering that some samples in actual applications cannot obtain the most complete indicators, the system 600 can also provide the most appropriate prediction model for this situation; in addition, the system 600 also provides a model recommendation module 630 and a model selection module 640, making the system more flexible and meeting the diverse application needs of customers; in addition, the system 600 can use the most appropriate threshold for risk assessment based on the department or population from which the sample comes.

[0133] The present application also proposes a kit for evaluating the possibility of deep submucosal infiltration of a subject, which includes a prediction model established by the model construction method described above. The prediction model can be specifically integrated into an integrated circuit in the kit. Taking prediction model 1 as an example, the kit can be used to receive, for example, hsa-miR-378a and hsa-miR-1228, and calculate the prediction results based on the built-in prediction model 1 to give the possibility of deep submucosal infiltration of the subject. The kit can include the 6 different prediction models described above at the same time to suit different populations, and has a very wide range of applicability.

[0134] The present application also includes an electronic device, including a memory and a processor, wherein the memory is used to store instructions executable by the processor; and the processor is used to execute the instructions to implement the aforementioned method for constructing a model for assessing the possibility of deep submucosal infiltration in a subject.

[0135] Figure 7 is a system block diagram of an electronic device according to an embodiment of the present application. Figure 7 As shown, the electronic device 700 may include an internal communication bus 701, a processor 702, a read-only memory (ROM) 703, a random access memory (RAM) 704, and a communication port 705. When applied to a personal computer, the electronic device 700 may also include a hard disk 706. The internal communication bus 701 can realize data communication between components of the electronic device 700. The processor 702 can make judgments and issue prompts. In some embodiments, the processor 702 can be composed of one or more processors. The communication port 705 can realize data communication between the electronic device 700 and the outside. In some embodiments, the electronic device 700 can send and receive information and data from the network through the communication port 705. The electronic device 700 may also include different forms of program storage units and data storage units, such as a hard disk 706, a read-only memory (ROM) 703 and a random access memory (RAM) 704, which can store various data files used for computer processing and / or communication, and possible program instructions executed by the processor 702. The processor executes these instructions to implement the main part of the method. The processing result of the processor is transmitted to the user device through the communication port and displayed on the user interface.

[0136] The above-mentioned model construction method can be implemented as a computer program, stored in the hard disk 706, and loaded into the processor 702 for execution to implement the model construction method of the present application.

[0137] The present application also includes a computer-readable medium storing a computer program code, which, when executed by a processor, implements the model construction method described above.

[0138] When the construction method of the model for assessing the possibility of deep submucosal infiltration of a subject is implemented as a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described herein can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0139] It should be understood that the embodiments described above are only illustrative. The embodiments described herein may be implemented in hardware, software, firmware, middleware, microcode, or any combination thereof. For hardware implementation, the processor may be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and / or other electronic units designed to perform the functions described herein, or a combination thereof.

[0140] Some aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may be referred to as "data blocks", "modules", "engines", "units", "components" or "systems". The processor may be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DAPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, or combinations thereof. In addition, various aspects of the present application may be expressed as computer products located in one or more computer-readable media, which include computer-readable program codes. For example, computer-readable media may include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, tapes ...), optical disks (e.g., compact disks CDs, digital versatile disks DVDs ...), smart cards, and flash memory devices (e.g., cards, sticks, key drives ...).

[0141] A computer-readable medium may include a propagated data signal containing computer program code, such as in baseband or as part of a carrier wave. The propagated signal may have a variety of manifestations, including electromagnetic, optical, etc., or a suitable combination. A computer-readable medium may be any computer-readable medium other than a computer-readable storage medium, which may be connected to an instruction execution system, device or apparatus to communicate, propagate or transmit a program for use. The program code on the computer-readable medium may be propagated via any suitable medium, including radio, cable, fiber optic cable, radio frequency signal, or similar medium, or any combination of the above mediums.

[0142] The basic concepts have been described above. Obviously, for those skilled in the art, the above invention disclosure is only used as an example and does not constitute a limitation of the present application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements and amendments to the present application. Such modifications, improvements and amendments are suggested in the present application, so such modifications, improvements and amendments still belong to the spirit and scope of the exemplary embodiments of the present application.

[0143] At the same time, the present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0144] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining digits. Although the numerical domains and parameters used to confirm the breadth of their range in some embodiments of the present application are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

Claims

1. A method for constructing a model for assessing the possibility of deep submucosal infiltration in a subject, characterized in that: include: Step S10: obtaining multiple independent variables for the model, the multiple independent variables including: the expression levels of microRNA in plasma hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21, hsa-miR-1228, and performing structured processing on the multiple independent variables; Step S20: Divide the subjects into SM2 group and non-SM2 group, and encode them as 1 and 0 respectively to form the dependent variable for the model; Step S30: for the multiple independent variables, calculating the correlation coefficient between each independent variable and the dependent variable, and taking the independent variable with a correlation coefficient greater than a second threshold as a valid independent variable; Step S40: performing correlation analysis on the multiple independent variables to determine the relationship between the multiple independent variables; Step S50: dividing the data into a training set and a test set, constructing a regression model including the independent variables according to the training set, and testing the performance of the regression model according to the test set, wherein the data includes the multiple independent variables and the dependent variable; Step S60: further screening the valid independent variables according to any of the p-value, correlation coefficient and AUC value of the regression model on the training set and the test set of each of the independent variables in the regression analysis result to obtain the screened valid independent variables, wherein when the p-value of the independent variable is less than 0.05, the independent variable whose correlation coefficient is greater than the first threshold is taken as the valid independent variable; Step S70: Repeat step S60 until the AUC value of the regression model on the training set is reduced to 0.78, the number of the effective independent variables after screening is greater than or equal to 2, and the effective independent variables after screening include at least one microRNA, to obtain a plurality of first models; Step S80: According to the relationship between the multiple independent variables obtained in step S40, the independent variables with replacement relationship in the several first models are replaced and combined, and a regression model is constructed using the replaced and combined independent variables to obtain several second models, wherein the model includes any one of the first model and the second model.

2. The construction method according to claim 1, characterized in that: The relationship between the multiple independent variables determined according to step S40 includes: hsa-miR-16, hsa-miR-21, and hsa-miR-345 have a substitution relationship; There is a certain degree of substitution relationship between the three microRNAs hsa-miR-378a, hsa-miR-195 and hsa-miR-214; hsa-miR-16, hsa-miR-21, and hsa-miR-345, these three microRNAs have a certain degree of substitution relationship with hsa-miR-214, hsa-miR-193a, hsa-miR-584, and U6 snRNA; There is no substitution relationship between hsa-miR-150 and other microRNAs, no substitution relationship between hsa-miR-129 and other microRNAs, no substitution relationship between cel-miR-39 and other microRNAs, and no substitution relationship between hsa-miR-1228 and other microRNAs.

3. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-378a and hsa-miR-1228, and the prediction formula of the model is: Logit(P)=β1×hsa-miR-378a+β2×hsa-miR-1228+β0, wherein the value range of coefficient β1 is [0.47,1.50]; the value range of coefficient β2 is [-2.51,-0.86]; the value range of coefficient β0 is [-2.14,40.10].

4. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-150, hsa-miR-378a and hsa-miR-1228, and the prediction formula of the model is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-378a+β3×hsa-miR-1228+β0, wherein the value range of coefficient β1 is [-1.43, 0.32]; the value range of coefficient β2 is [0.27, 1.42]; the value range of coefficient β3 is [-2.31, -0.44]; the value range of coefficient β0 is [1.31, 62.80].

5. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-150, hsa-miR-195, hsa-miR-378a and hsa-miR-1228, and the prediction formula of the model is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-1228+β0, wherein the value range of coefficient β1 is [-1.42, 0.33]; the value range of coefficient β2 is [-0.41, 0.54]; the value range of coefficient β3 is [0.09, 1.50]; the value range of coefficient β4 is [-2.33, -0.45]; the value range of coefficient β0 is [0.79, 62.57].

6. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-16 and hsa-miR-1228, and the prediction formula of the model is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-16+β5×hs a-miR-1228+β0, where the value range of coefficient β1 is [-1.43, 0.34]; the value range of coefficient β2 is [-0.43, 0.56]; the value range of coefficient β3 is [-0.06, 1.66]; the value range of coefficient β4 is [-0.64, 0.61]; the value range of coefficient β5 is [-2.34, -0.44]; the value range of coefficient β0 is [-1.83, 64.62].

7. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-150, hsa-miR-195, hsa-miR-378a, hsa-miR-21 and hsa-miR-1228, and the prediction formula of the model is: Logit(P)=β1×hsa-miR-150+β2×hsa-miR-195+β3×hsa-miR-378a+β4×hsa-miR-21+β5×h sa-miR-1228+β0, among which, the value range of coefficient β1 is [-1.45, 0.30]; the value range of coefficient β2 is [-0.43, 0.53]; the value range of coefficient β3 is [-0.07, 1.47]; the value range of coefficient β4 is [-0.32, 0.58]; the value range of coefficient β5 is [-2.47, -0.47]; the value range of coefficient β0 is [2.16, 67.30].

8. The construction method according to claim 1, characterized in that: The effective independent variables of the model are hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228, the prediction formula of the model is: Logit(P)=β1×hsa-miR-129+β2×hsa-miR-150+β3×hsa-miR-193a+β4×hsa-miR-195+β5×hsa-miR-214+β6×hsa-miR-345+β7×hsa-miR-378a+β8×hsa-miR-584+β9×U6snRNA+β10×hsa-miR-16+β11×hsa-miR-21+β12×hsa-miR-1228+β0, wherein the value range of coefficient β1 is [-1.08,0.58]; the value range of coefficient β2 is [-1.08,0.58]; The value range is [-1.55, 0.57]; the value range of coefficient β3 is [-0.38, 1.05]; the value range of coefficient β4 is [-0.06, 1.29]; the value range of coefficient β5 is [-1.69, 0.16]; the value range of coefficient β6 is [-1.08, 1.72]; the value range of coefficient β7 is [-0.27, 1.74]; the value range of coefficient β8 ... The value range is [-0.49, 1.35]; the value range of coefficient β9 is [-1.20, 0.23]; the value range of coefficient β10 is [-2.30, 0.99]; the value range of coefficient β11 is [-0.76, 1.89]; the value range of coefficient β12 is [-2.93, -0.25]; the value range of coefficient β0 is [-8.21, 71.15].

9. A system for assessing the possibility of deep submucosal invasion in a subject, characterized in that: include: A data acquisition module, used for obtaining sample data of a subject, wherein the sample data includes any one of the expression amounts of microRNAs in plasma: hsa-miR-129, hsa-miR-150, hsa-miR-193a, hsa-miR-195, hsa-miR-214, hsa-miR-345, hsa-miR-378a, hsa-miR-584, U6 snRNA, hsa-miR-16, hsa-miR-21 and hsa-miR-1228; A data preprocessing module, used to preprocess the sample data, including removing duplicate samples, unifying units, and dividing the sample data into different analysis groups according to indicator types and quantities; A model recommendation module, for recommending, for different analysis groups, according to the types and numbers of indicators contained therein, a plurality of prediction models with larger AUC values ​​from the six prediction models established by the construction method according to any one of claims 1 to 8, according to the sizes of AUC values; A model selection module, used to provide a selection function for the user to output one or more prediction models from the prediction models recommended by the model recommendation module for risk assessment and calculation; The risk assessment module is used to calculate the corresponding prediction value according to one or more prediction models output by the model selection module, and to give a suitable Logit threshold or risk score threshold according to the department or population information corresponding to the sample data, and to divide the risk level into any one of high possibility, medium possibility and low possibility according to the Logit threshold or risk score threshold.

10. A kit for evaluating the possibility of deep submucosal invasion in a subject, characterized in that: The method comprises a prediction model established by the construction method according to any one of claims 1 to 8.

11. An electronic device, characterized in that: include: a memory for storing instructions executable by a processor; A processor, configured to execute the instructions to implement the construction method according to any one of claims 1 to 8.

12. A computer readable medium storing computer program code, characterized in that: When the computer program code is executed by a processor, the construction method according to any one of claims 1 to 8 is implemented.

Images