Method for constructing a prediction model for the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma

Abstract

The present invention provides a method for constructing a radiotherapy and chemotherapy efficacy prediction model for patients with esophageal squamous cell carcinoma, including: step S1, obtaining the original RNAseq sequencing data of the patient, normalizing the expression matrix of the transcriptome data to obtain FPKM; step S2, inputting the processed expression matrix into the constructed LASSO model, and screening out highly correlated features; step S3, constructing an XGBoost prediction model for classification fitting, optimizing model parameters, and accurately realizing the prediction of complete pathological remission and incomplete pathological remission of the patient. Based on the transcriptome data of ESCC patients, after LASSO regression feature screening, the extracted typical features are classified and fitted using XGBoost, combined with the patient's efficacy and survival data, effectively realizing the high-risk and low-risk stratification of clinical treatment of patients, compared with other efficacy prediction modeling, the performance of prediction accuracy and the like is significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer-aided medicine, and in particular to a method for constructing a radiotherapy and chemotherapy efficacy prediction model for patients with esophageal squamous cell carcinoma. Background Art

[0002] Esophageal cancer (EC) is the sixth leading cause of cancer-related death worldwide and the seventh leading cause of morbidity. The incidence and mortality of esophageal cancer in China account for about half of the global total, with nearly 90% of cases attributed to esophageal squamous cell carcinoma (ESCC). Preoperative neoadjuvant chemoradiotherapy (nCRT) as a guideline recommendation can increase the R0 resection rate, improve local control, and prolong the survival of ESCC patients. However, in ESCC, the pathological complete response (pCR) rate after nCRT is only about 28%, and a certain proportion of patients do not benefit from nCRT but have toxicity-related morbidity risks. If these ESCC patients can be divided into subgroups with corresponding good or bad prognoses, doctors can be guided to perform therapeutic interventions or plan different diagnosis and treatment plans to avoid missing the best treatment opportunity. Therefore, accurate prediction of the efficacy of neoadjuvant chemoradiotherapy is crucial for optimizing treatment strategies and improving patient prognosis.

[0003] At present, there are relatively few studies or methods for predicting the efficacy of chemoradiotherapy for esophageal squamous cell carcinoma. Existing studies have used radiomics features or pathological features from CT, MRI, and H&E images of pathologically confirmed ESCC patients to construct a machine learning-based efficacy prediction model, and achieved good classification performance on the validation set. In addition, some researchers have extracted deep learning features of ESCC patients from pre-trained convolutional neural networks and developed a model based on deep learning methods to predict the therapeutic response of ESCC patients to nCRT. However, there is a lack of research on the use of transcriptomics data to predict the efficacy of chemoradiotherapy for ESCC patients. In terms of methods, machine learning (ML) and deep learning (DL) methods have their own advantages. In view of the limited sample size in the patient cohort, the number of deep learning parameters is large, and the high-dimensional deep features extracted from small sample data sets often lead to overfitting of the model, resulting in weak generalization ability and low prediction accuracy of the model. The shallow features extracted by feature engineering combined with traditional machine learning are more conducive to model learning and clinical deployment, while taking into account significant classification performance to achieve ideal prediction results.

[0004] The above-mentioned record of background technical knowledge is intended to help ordinary technicians in this field understand the existing technologies that are relatively close to the present invention, and at the same time facilitate the understanding of the inventive concept and technical solution of the present invention. It should be clear that in the absence of clear evidence that the above-mentioned content has been disclosed before the filing date of this patent application, the above-mentioned background technology should not be used to evaluate the novelty of the technical solution of this application. Summary of the invention

[0005] Technical issues

[0006] In order to solve the above problems, the purpose of the present invention is to provide a method for constructing a radiotherapy and chemotherapy efficacy prediction model for patients with esophageal squamous cell carcinoma. Based on the transcriptome data of ESCC patients, after LASSO regression feature screening, the extracted typical features are classified and fitted using XGBoost, and combined with the patient's efficacy and survival data, the high-risk and low-risk stratification of clinical treatment of patients is effectively achieved. The leave-one-out cross-validation method is used during prediction to alleviate overfitting and provide accurate performance estimation. Compared with other efficacy prediction modeling, the performance in prediction accuracy and other performance is significantly improved.

[0007] Technical Solution

[0008] That is, the present invention includes the following partial aspects.

[0009] The first part is the construction method of the prediction model of chemoradiotherapy efficacy in patients with esophageal squamous cell carcinoma, including:

[0010] Step S1, obtaining the original RNAseq sequencing data of the patient, and normalizing the expression matrix of the transcriptome data to obtain FPKM;

[0011] Step S2: input the processed expression matrix into the constructed LASSO model, and find the best LASSO model by adjusting the regularization parameter to balance the complexity and fit of the model and screen out highly correlated features;

[0012] Step S3: Use highly correlated features to construct an XGBoost prediction model for classification fitting, use accuracy, sensitivity, specificity, F1, positive prediction rate, and negative prediction rate as model effect evaluation indicators, optimize model hyperparameters, and accurately predict patients' pathological complete remission and pathological incomplete remission.

[0013] In some specific implementations, in step S1, normalizing the expression matrix of the transcriptome data to obtain FPKM specifically includes:

[0014] Step S1.1, calculate the number of reads of each gene, sum the count data of each gene, and obtain the total number of reads of the gene;

[0015] Step S1.2, calculating the length of each gene, that is, the sum of the lengths of all exons of the gene;

[0016] Step S1.3, calculate the PFKM value of each gene, divide the total number of reads by the length of the gene, then divide by the total number of reads, and finally multiply by 1000000 to get the FPKM value.

[0017] In some specific implementations, in step S1, the formula for obtaining FPKM is as follows:

[0018]

[0019] in, Indicates gene of The value, i.e., its standardized expression, represents the number of exon fragments per million mapped reads per kilobase length. Indicates alignment to gene The number of reads or fragments (usually row counts); Indicates gene The length of a sequence, usually in base pairs (bp); It indicates the total number of mapped reads or fragments of all genes in the sample, which is usually called sequencing depth.

[0020] In some specific implementations, in step S2, the formula of the LASSO model is as follows:

[0021]

[0022] in, Represents the objective function value (LASSO loss function), the value that needs to be minimized; Indicates the number of samples; Represents the true value of the i-th sample (target variable); Indicates The sample eigenvalues; Indicates The regression coefficient of each feature; represents the regularization strength (hyperparameter); Indicates The absolute value of the regression coefficient; Indicates the total number of features; Used to standardize the residual sum of squares to make it consistent with the unit of the regularization term; This sum operation represents a sum operation that traverses all samples; Represents a sum operation that traverses all features.

[0023] In some specific implementations, in step S3, the XGBoost prediction model is The final predicted value of samples is:

[0024]

[0025] in, Indicates The predicted value of samples (final output value); Indicates The input feature vector of samples; Indicates The prediction function of the sample, which is the final prediction value of the XGBoost model; Represents the total number of decision trees in the XGBoost model; Indicates The prediction function of the tree; is the function space, representing the set of all possible trees; Indicates that all Sum the predictions of the trees.

[0026] In some specific implementations, in step S3, the XGBoost prediction model minimizes the objective function as shown below:

[0027]

[0028] in, Represents the objective function, which represents the optimization goal to be minimized, including prediction error and model complexity; Indicates The true value (label) of each sample; represents the squared error, which is the square of the difference between the predicted value and the true value; Represents the total number of decision trees in the XGBoost model; Indicates Tree complexity; It is the formula symbol for computational complexity.

[0029] In some specific implementations, in step S3, accuracy (Accuracy, Acc), sensitivity (Sensitivity, Sens), specificity (Spec), F1, positive predictive value (PPV) and negative predictive value (NPV) are expressed as follows:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] Among them, TP is true positive, indicating that the positive class is predicted as the negative class; FN is false negative, indicating that the positive class is predicted as the negative class; FP is false positive, indicating that the negative class is predicted as the positive class; TN is true negative, indicating that the negative class is predicted as the negative class; F1 value represents the harmonic mean of precision and recall; PPV represents the median score of the positive class; NPV represents the median score of the negative class.

[0037] In some specific implementations, in step S3, the optimization model hyperparameters include using an improved particle swarm algorithm to automatically combine and optimize the hyperparameters of the XGBoost prediction model, randomly generating a particle swarm with a combination of hyperparameters, calculating the convergence speed of the particles and adaptively adjusting the inertia weights based on them, calculating the local and global search capabilities of the particles and the local and global learning factors, comprehensively considering the adaptive inertia weights, adaptive local and global learning factors to update the particle speed, and update the particle position; using the performance of the XGBoost prediction model on the validation set as a fitness evaluation; repeating the above steps until the stop condition is met, such as reaching the maximum number of iterations or meeting the accuracy requirements, and obtaining the optimal parameter configuration. The improved particle swarm algorithm comprehensively considers the adaptive inertia weights, adaptive local and global learning factors to update the particle speed, and updates the particle position, which improves the optimization accuracy and convergence speed of the particle swarm algorithm, and can quickly and accurately prepare the optimal parameter combination.

[0038] In some specific implementations, in step S3, RF, SVM or LightGBM is used instead of XGBoost to build a prediction model.

[0039] The second part is a prediction model for the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma constructed by the method described in the first part.

[0040] The third part is the method for predicting the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment, including: obtaining the original RNAseq sequencing data of the patient to be predicted, normalizing the expression matrix of the transcriptome data to obtain FPKM, inputting it into the prediction model for the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma, and outputting the prediction results of the efficacy.

[0041] In some specific embodiments, the method for predicting the efficacy of chemoradiotherapy for patients with esophageal squamous cell carcinoma also includes: predicting the pathological complete remission (pCR) of the patient to be predicted, taking pCR as the outcome, and combining the patient's OS or PFS survival time to draw a survival curve for ESCC patients.

[0042] In some specific embodiments, the ESCC patient survival curve is a Kaplan-Meier survival curve. time points, each with a subscript To express, From 1 to integer, survival probability It can be expressed as:

[0043]

[0044] in, Indicates time point; Indicated in The previous effective number of people; Indicated in Number of deaths; Indicates at the last time point The probability of survival.

[0045] The present invention provides a construction method, model and prediction method of a radiotherapy and chemotherapy efficacy prediction model for patients with esophageal squamous cell carcinoma. The radiotherapy and chemotherapy efficacy of patients are predicted and the patient survival curve is drawn through transcriptome data preprocessing, transcriptome feature screening based on lasso regression, and establishment of an efficacy prediction model based on a traditional machine learning XGBoost feature classifier. A machine learning model is constructed based on transcriptome data of patients with esophageal squamous cell carcinoma undergoing radiotherapy and chemotherapy in actual clinical trials to achieve clinical efficacy prediction and stratify patients into high-risk and low-risk groups. Features highly related to efficacy prediction are effectively screened based on lasso regression, which alleviates model overfitting and reduces the difficulty of model learning. Machine learning modeling of the screened features is performed in combination with XGBoost, and patient risk stratification and classification fitting can effectively improve the accuracy of efficacy prediction and other performances. Leave-one-out cross-validation is used to alleviate overfitting and provide accurate performance estimation.

[0046] The fourth part, a device for predicting the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma, includes:

[0047] at least one processor;

[0048] at least one memory for storing at least one program;

[0049] When the at least one program is executed by the at least one processor, the at least one processor implements the method for constructing a model for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma as described in the first part or the method for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma for diagnosis and treatment of non-diseases as described in the third part.

[0050] Part 5, a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to execute the method for constructing a model for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma described in the first part or the method for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment described in the third part.

[0051] Based on the common knowledge in this field, the above-mentioned preferred conditions can be combined with each other to obtain a specific implementation method.

[0052] Beneficial Effects

[0053] Compared with the prior art, the present invention proposes a prediction model of chemoradiotherapy for esophageal squamous cell carcinoma based on XGBoost, a method for constructing the model and prediction. Based on the transcriptome data of ESCC patients, the lasso regression is used to effectively screen the features highly related to the prediction of the efficacy, alleviate the overfitting of the model, reduce the difficulty of model learning, and effectively extract the transcriptome features highly related to the prediction of the efficacy. The use of L1 regularization will make the weight of some learned features 0, thereby achieving the purpose of sparsification and feature selection; the extracted typical features are classified and fitted using XGBoost, effectively improving the accuracy of the prediction of the efficacy, and combining the efficacy and survival data of the patient, effectively realizing the high-risk and low-risk stratification of the clinical treatment of the patient, and using the leave-one-out cross-validation method to alleviate the overfitting phenomenon and provide accurate performance estimation during prediction, and adopting a variety of evaluation indicators, including accuracy, sensitivity, specificity, F1, positive prediction rate and negative prediction rate, so that the evaluation of the model effect is more accurate and comprehensive. Compared with the efficacy prediction modeling of the impact and pathological images by feature engineering and deep learning methods, the machine learning model constructed based on transcriptome data has significantly improved the performance of prediction accuracy.

[0054] The XGBoost-based machine learning model constructed by the present invention can effectively use the transcriptomic data of patients with esophageal squamous cell carcinoma to predict the efficacy of radiotherapy and chemotherapy. Specifically, the accuracy is 0.956, the AUC is 0.918, the specificity is 1.00, the sensitivity is 0.958, the F1 value is 0.955, the PPV is 1.00, and the NPV is 0.935. At the same time, the prediction results of the model can accurately distinguish between pCR and non-pCR patients. In drawing the OS survival curve, the log-rank p value is 0.033; in drawing the PFS survival curve, the log-rank p is 0.039.

[0055] The present invention adopts the above technical solution to achieve the above purpose, which makes up for the shortcomings of the prior art and has reasonable design and convenient operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to make the above and / or other purposes, features, advantages and examples of the present invention more obvious and easy to understand, the drawings required for use in the specific implementation of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 This is a flowchart showing the efficacy prediction of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma;

[0058] Figure 2 A visual diagram showing a tree in the XGBoost model;

[0059] Figure 3 Survival curve diagram showing the therapeutic effect prediction of ESCC patients. DETAILED DESCRIPTION

[0060] Those skilled in the art can refer to the content of this article and appropriately replace and / or modify the process parameters to achieve the same. However, it should be particularly noted that all similar replacements and / or modifications are obvious to those skilled in the art and are considered to be included in the present invention. The products and preparation methods described in the present invention have been described through preferred examples. It is obvious that relevant personnel can modify or appropriately change and combine the products and preparation methods described herein without departing from the content, spirit and scope of the present invention to implement and apply the technology of the present invention.

[0061] Unless otherwise defined, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the invention belongs. The present invention uses the methods and materials described herein; however, other suitable methods and materials known in the art may also be used. The materials, methods, and examples described herein are illustrative only and are not intended to be limiting. All publications, patent applications, patents, provisional applications, database entries, and other references mentioned herein are incorporated herein by reference in their entirety. In the event of a conflict, the present specification, including definitions, shall prevail.

[0062] Unless specifically stated, the materials, methods, and examples described herein are illustrative only and not limiting. Although methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, suitable methods and materials are still described herein.

[0063] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. At the same time, the embodiments in this application and the features in the embodiments can be combined with each other without conflict.

[0064] It should be understood that any technical solution claimed for protection in the present invention does not involve the diagnosis and treatment of diseases.

[0065] In order to facilitate understanding of the embodiments of the present invention, the abbreviations and key terms that may be involved in the embodiments of the present invention are first explained or defined. For undefined abbreviations or key terms, they are commonly understood by those skilled in the art.

[0066] EC: esophageal cancer;

[0067] ESCC: esophageal squamous cell carcinoma;

[0068] nCRT: neoadjuvant chemoradiotherapy;

[0069] pCR: pathological complete response;

[0070] non-pCR: pathological incomplete response;

[0071] Fragments: refers to each nucleic acid fragment used for sequencing;

[0072] OS: Overall Survival, overall survival;

[0073] PFS: Progression-Free Survival, progression-free survival;

[0074] Acc:Accuracy, accuracy;

[0075] Sens:Sensitivity, sensitivity;

[0076] Spec:Specificity, specificity;

[0077] F1: The harmonic mean of precision and recall.

[0078] PPV: Positive predictive value, positive predictive rate;

[0079] NPV: Negative predictive value, negative predictive rate.

[0080] In addition, the test methods used in the examples are conventional methods unless otherwise specified; the materials, reagents, etc. used are all commercially available unless otherwise specified. The reagents or instruments used without indicating the manufacturer are all conventional products that can be obtained commercially. All the publications and other references mentioned in this article are incorporated herein by reference in their entirety.

[0081] The present invention is described in detail below

[0082] Embodiment 1:

[0083] Provided is a method for constructing a prediction model for the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma, a prediction model constructed according to the method, and a specific process for predicting the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment based on the prediction model. Figure 1 The flowchart is shown, and the specific steps are as follows.

[0084] Data source: The patients included in this study were from a prospective randomized clinical trial in Zhejiang Cancer Hospital, which aimed to evaluate the efficacy of postoperative nCRT in patients with locally advanced ESCC. A total of 45 patients were included, including 16 patients with pathological complete remission (pCR) and 29 patients with pathological incomplete remission (non-pCR). All samples were evaluated using leave-one-out cross-validation.

[0085] RNA was extracted from esophageal cancer pathological tissues, and appropriate methods were selected for sample quality testing based on the characteristics of biological samples. After passing the test, mRNA enrichment, reverse transcription, library construction, and sequencing were performed on the machine. The raw image data (bcl file) from high-throughput sequencing was processed to obtain the raw sequencing data (Raw data) of the sample. The Raw data was checked for sequencing data quality by FastQC quality control, and then high-quality data clean data was obtained after removing the adapters, low-quality sequences, and contamination. The clean data was then compared with the reference genome based on the STAR tool to further obtain the expression matrix of the transcriptome data.

[0086] Step S1: Transcriptome data preprocessing

[0087] The original RNAseq sequencing data of the patient was obtained, and the expression matrix of the transcriptome data was normalized to obtain FPKM (i.e., fragments per kilobase of transcription per million mapped reads). Specifically, the number of reads of each gene was calculated, and the count data of each gene was summed to obtain the total number of reads of the gene; the length of each gene was calculated, that is, the sum of the lengths of all exons of the gene; the PFKM value of each gene was calculated, the total number of reads was divided by the length of the gene, and then divided by the total number of reads, and finally multiplied by 1,000,000 to obtain the FPKM value.

[0088] The formula to get FPKM is as follows:

[0089]

[0090] in, Indicates gene of The value, i.e., its standardized expression, represents the number of exon fragments per million mapped reads per kilobase length. Indicates alignment to gene The number of reads or fragments (usually row counts); Indicates gene The length of a sequence, usually in base pairs (bp); It indicates the total number of mapped reads (or total number of fragments) of all genes in the sample, which is usually called sequencing depth.

[0091] Step S2: Transcriptome feature screening based on LASSO regression

[0092] Using the preprocessed data obtained in step S1, LASSO (Least absolute shrinkage and selection operator) regression is used for feature engineering. LASSO regression is a linear regression method that uses L1 regularization. The use of L1 regularization will make some learned feature weights 0, thereby achieving the purpose of sparsification and feature selection.

[0093] The processed expression matrix is ​​input into the constructed LASSO model, and the best LASSO model is found by adjusting the regularization parameters to balance the complexity and fit of the model and screen out highly correlated features; it is expressed as follows:

[0094]

[0095] in, Represents the objective function value (LASSO loss function), the value that needs to be minimized; Indicates the number of samples; Represents the true value of the i-th sample (target variable); Indicates The sample eigenvalues; Indicates The regression coefficient of each feature; represents the regularization strength (hyperparameter); Indicates The absolute value of the regression coefficient; Indicates the total number of features; Used to standardize the residual sum of squares to make it consistent with the unit of the regularization term; This sum operation represents a sum operation that traverses all samples; Represents a sum operation that traverses all features.

[0096] Step S3: Establishing a therapeutic effect prediction model based on traditional machine learning XGBoost feature classifier

[0097] The XGBoost prediction model was constructed using highly correlated features for classification fitting, and the accuracy, sensitivity, specificity, F1, positive predictive rate, and negative predictive rate were used as model effect evaluation indicators to accurately predict pathological complete remission and pathological incomplete remission. The features obtained by lasso regression screening in step S2 were used to construct XGBoost (eXtreme GradientBoosting) for classification fitting. XGBoost is a tree ensemble model with K trees. The (final) prediction value for the i-th sample is:

[0098]

[0099] in, Indicates The predicted value of samples (final output value); Indicates The input feature vector of samples; Indicates The prediction function of the sample, which is the final prediction value of the XGBoost model; Represents the total number of decision trees in the XGBoost model; Indicates The prediction function of the tree; is the function space, representing the set of all possible trees; Indicates that all Sum the predictions of the trees.

[0100] The minimization objective function is as follows:

[0101]

[0102] in, Represents the objective function, which represents the optimization goal to be minimized, including prediction error and model complexity; Indicates The true value (label) of each sample; represents the squared error, which is the square of the difference between the predicted value and the true value; Represents the total number of decision trees in the XGBoost model; Indicates Tree complexity; It is the formula symbol for computational complexity.

[0103] For the hyperparameters of the XGBoost prediction model, the improved particle swarm algorithm is used to automatically optimize the hyperparameters of the XGBoost prediction model. The specific steps include:

[0104] (1) Randomly generate an initial particle group, where each particle τ represents a set of hyperparameters of the XGBoost prediction model.

[0105] (2) Calculate the convergence speed of particles: , adaptively adjust the inertia weight according to the convergence speed: ,in, It represents the convergence speed of the τth particle and the rate of change of the particle velocity, which is used to adaptively adjust the inertia weight; represents the velocity of the τth particle, which indicates the moving speed of the particle in the search space; represents the velocity of the τth particle in the last iteration (or last time step); Indicates the maximum value of the inertia weight, usually set to 0.9, indicating the global search capability of the particle swarm algorithm at the beginning; Indicates the minimum value of the inertia weight, usually set to 0.4, indicating the local search capability of the particle swarm algorithm in the later stage; Represents the adaptive inertia weight of the τth particle, which is dynamically adjusted according to the convergence speed of the particle.

[0106] (3) Calculate the local search capability of particles: ; Calculate the global search capability of particles: ; Calculate the adaptive local learning factor: ; Calculate the adaptive global learning factor: ;in, represents the local search capability of the τth particle, which represents the distance between the current position of the particle and its individual optimal position; represents the global search capability of the τth particle, and represents the distance between the current position of the particle and the global optimal position; represents the position of the τth particle, and represents the position of the particle in the search space; It represents the maximum value of the learning factor, which is usually set to 2, indicating the strength with which particles in the particle swarm algorithm are attracted by their individual optimal positions; It represents the minimum value of the learning factor, which is usually set to 0. It indicates the intensity with which particles in the particle swarm algorithm are attracted by the global optimal position. and They represent the adaptive local learning factor and the adaptive global learning factor of the τth particle, respectively, and are dynamically adjusted according to the local and global search capabilities of the particle; represents the individual optimal position of the τth particle, that is, the optimal position found by the particle in history; represents the global optimal position, that is, the optimal position found in the history of the entire particle swarm.

[0107] (4) Update particle velocity: , indicating that the new velocity of the particle is an adjustment method after comprehensive consideration of the adaptive inertia weight, adaptive local and global learning factors; update the particle position: , which means that the new position of the particle is its current position plus its velocity vector; where represents a random function, which generates a random number between 0 and 1 to introduce randomness; d represents the dimension of the search space, that is, the dimension of the particle position vector;

[0108] (5) Use the performance of the XGBoost prediction model on the validation set as fitness evaluation; repeat the above steps until the stopping condition is met, such as reaching the maximum number of iterations or meeting the accuracy requirements, to obtain the optimal parameter configuration.

[0109] This embodiment uses the above-mentioned improved particle swarm algorithm to perform automatic combination optimization of the hyperparameters of the XGBoost prediction model. By adaptively adjusting the inertia weight and the learning factor, the search efficiency and optimization accuracy of the particle swarm algorithm are improved. The optimization accuracy of the particle swarm algorithm is improved by combining the global and local search capabilities. Through the adaptive mechanism, the particle swarm algorithm can quickly converge to the optimal solution. Compared with configuring the parameters of the XGBoost prediction model only according to the conventional scheme, the automatic combination optimization configuration of the hyperparameters of the XGBoost prediction model using the above-mentioned improved particle swarm algorithm can reduce at least 30% of the training time, which significantly improves the model efficiency.

[0110] The specific parameter configuration of the XGBoost prediction model is as follows: the model sets the initial learning rate (learning_rate) to 0.01 to control the step size of each iteration and enhance the stability of the model; the maximum depth of the tree (max_depth) is 9, allowing more complex feature interactions; the minimum sample weight and split gain threshold when the node is split are controlled by min_child_weight (set to 1) and gamma (set to 0.2) respectively to avoid overfitting; 80% of the sample subsets (subsample) and 30% of the feature columns (colsample_bytree) are used to build each tree to enhance the generalization ability of the model; the regularization parameter reg_lambda (L2 regularization) is set to 1, which helps to reduce the complexity of the model; the category weight ratio is set by scale_pos_weight to balance the category distribution (currently 1); the number of trees (n_estimators) is set to 40 to ensure that the model has sufficient complexity; the random seed (seed) is set to 3407 to ensure the repeatability of the results. The loss function uses binary, the optimization target is logarithmic loss, and the evaluation indicators (eval_metric) include logloss and auc. Combining these configurations, the model is suitable for tasks with small step size, strong regularization control, and balanced performance and complexity.

[0111] Figure 2It represents the visualization structure diagram of a tree in the XGBoost model, showing the process of how the tree structure model makes decisions on input features step by step. For each sample, starting from the root node, it gradually reaches the leaf node according to the splitting conditions and finally obtains a predicted value. Splitting node: It shows the splitting conditions, such as the feature used for splitting, the splitting threshold, and the path weights of the left and right subtrees. Splitting condition: The condition marked on each splitting node (such as feature_name < threshold) indicates how to divide the sample into the left or right child nodes. The splitting gain of each path represents the contribution of this splitting to the model. The closer the splitting condition is to the root node, the higher the importance of this feature to the overall prediction of the model. Leaf node value: The value of each leaf node represents the weighted result of the prediction, and these values will be summed up to the final model output.

[0112] When making model predictions, six machine learning model evaluation metrics are selected, including accuracy (Accuracy, Acc), sensitivity (Sensitivity, Sens), specificity (Specificity, Spec), F1, positive predictive value (Positive predictive value, PPV), and negative predictive value (Negative predictive value, NPV), which are expressed as follows:

[0113]

[0114]

[0115]

[0116]

[0117]

[0118]

[0119] Among them, TP is true positive, indicating that the positive class is predicted as the negative class; FN is false negative, indicating that the positive class is predicted as the negative class; FP is false positive, predicting the negative class as the positive class; TN is true negative, indicating that the negative class is predicted as the negative class; the F1 value represents the harmonic mean of precision and recall; PPV represents the proportion of those classified as positive that are actually positive; NPV represents the proportion of those classified as negative that are actually negative. The prediction evaluation metrics and results are shown in Table 1.

[0120] Table 1 - Evaluation Metrics and Results of XGBoost for Predicting the Efficacy of Chemoradiotherapy for Esophageal Squamous Cell Carcinoma

[0121]

[0122] Note: AUC, area under the curve; Acc, accuracy; Spec, specificity; Sens, sensitivity; PPV, positive predictive value; NPV, negative predictive value.

[0123] As can be seen from Table 1, the XGBoost-based machine learning model constructed in the present invention can effectively use the transcriptomic data of patients with esophageal squamous cell carcinoma to predict the efficacy of radiotherapy and chemotherapy, and its prediction AUC reaches 0.918, indicating that the model has extremely excellent classification performance, and its accuracy, specificity, and sensitivity are all excellent. Compared with the effect and pathological images, feature engineering, and deep learning method modeling of efficacy prediction, the machine learning model constructed based on transcriptomic data has significantly improved performance in prediction accuracy and other aspects.

[0124] Also includes step S4, patient survival curve drawing

[0125] Based on the prediction of the efficacy prediction model in step S3, the pathological complete remission (pCR) status of each patient was predicted. Taking pCR as the outcome, combined with OS or PFS time, the Kaplan-Meier survival curve of the experimental patients was drawn. time points, each with a subscript To express, From 1 to integer, survival probability It can be expressed as:

[0126]

[0127] in, Indicates time point; Indicated in The previous effective number of people; Indicated in Number of deaths; Indicates at the last time point The probability of survival.

[0128] Figure 3 The Kaplan-Meier survival curves of the predicted samples' overall survival and progression-free survival are shown. In the drawn OS survival curve, the log-rank p value is 0.033; in the drawn PFS survival curve, the log-rank p is 0.039, indicating that the prediction results of the model can accurately distinguish between pCR and non-pCR patients, and effectively achieve high-risk and low-risk stratification of patients for clinical treatment.

[0129] Embodiment 2:

[0130] Based on the above embodiments, a device for predicting the therapeutic effect of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma is also provided, comprising:

[0131] at least one processor;

[0132] at least one memory for storing at least one program;

[0133] When the at least one program is executed by the at least one processor, the at least one processor implements at least one step of the aforementioned method for constructing a radiotherapy and chemotherapy efficacy prediction model for esophageal squamous cell carcinoma patients or the method for predicting the radiotherapy and chemotherapy efficacy of esophageal squamous cell carcinoma patients for non-disease diagnosis and treatment, and can achieve the same technical effect. To avoid repetition, this embodiment will not be repeated.

[0134] Embodiment 3:

[0135] On the basis of the aforementioned embodiment, a computer-readable storage medium is further provided, in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to execute at least one step of the aforementioned method for constructing a radiotherapy and chemotherapy efficacy prediction model for patients with esophageal squamous cell carcinoma or the method for predicting the radiotherapy and chemotherapy efficacy of patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment, and can achieve the same technical effect. To avoid repetition, this embodiment will not be repeated.

[0136] Computer readable media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PR AM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0137] The conventional techniques in the above embodiments are prior arts known to those skilled in the art, and thus will not be described in detail here.

[0138] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0139] Although the present invention has been described in detail and some specific embodiments have been cited, it is obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the present invention.

[0140] Although the above-mentioned specific embodiments have been shown, described and pointed out the novel features applied to various embodiments, it should be understood that various omissions, substitutions and changes can be made to the form and details of the described device or method without departing from the spirit of the present disclosure. In addition, the above-mentioned various features and methods can be used independently of each other, or can be combined in various ways. All possible combinations and sub-combinations are intended to fall within the scope of the present disclosure. Many of the above-mentioned embodiments include similar components, and therefore, these similar components are interchangeable in different embodiments. Although the present invention has been disclosed in the context of certain embodiments and examples, it should be understood by those skilled in the art that the present invention can extend beyond the specifically disclosed embodiments to other alternative embodiments and / or applications and obvious modifications and equivalents thereof. Therefore, the present invention is not intended to be limited by the specific disclosure of the preferred embodiments herein.

[0141] Matters not covered in the present invention are all known technologies.

Claims

1. A method for constructing a prediction model for the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma, characterized in that include: Step S1, obtaining the original RNAseq sequencing data of the patient, and normalizing the expression matrix of the transcriptome data to obtain FPKM; Step S2: input the processed expression matrix into the constructed LASSO model, and find the best LASSO model by adjusting the regularization parameter to balance the complexity and fit of the model and screen out highly correlated features; Step S3, using highly correlated features to construct an XGBoost prediction model for classification fitting, using accuracy, sensitivity, specificity, F1, positive predictive rate and negative predictive rate as model effect evaluation indicators, optimizing model parameters, and accurately achieving the prediction of pathological complete remission and pathological incomplete remission of patients; In step S2, the formula of the LASSO model is as follows: in, Represents the objective function value, the value that needs to be minimized; Indicates the number of samples; Represents the true value of the i-th sample; Indicates The sample eigenvalues; Indicates The regression coefficient of each feature; represents the regularization strength; Indicates The absolute value of the regression coefficient; Indicates the total number of features; Used to standardize the residual sum of squares to make it consistent with the unit of the regularization term; This sum operation represents a sum operation that traverses all samples; Represents a sum operation that traverses all features.

2. The method according to claim 1, characterized in that: In step S1, normalizing the expression matrix of the transcriptome data to obtain FPKM specifically includes: Step S1.1, calculate the number of reads of each gene, sum the count data of each gene, and obtain the total number of reads of the gene; Step S1.2, calculating the length of each gene, that is, the sum of the lengths of all exons of the gene; Step S1.3, calculate the PFKM value of each gene, divide the total number of reads by the length of the gene, then divide by the total number of reads, and finally multiply by 1000000 to get the FPKM value.

3. The method according to claim 1 or 2, characterized in that: In step S1, the formula for obtaining FPKM is as follows: in, Indicates gene of The value, i.e., its normalized expression, is expressed per million mapped reads per kilobase length; Indicates alignment to gene The number of reads or fragments; Indicates gene Length; Indicates the total number of mapped reads or fragments of all genes in the sample.

4. The method according to claim 1, characterized in that: In the step S3, the optimization model hyperparameters includes using an improved particle swarm algorithm to automatically combine and optimize the hyperparameters of the XGBoost prediction model, randomly generating a particle swarm with a combination of hyperparameters, calculating the convergence speed of the particles and adaptively adjusting the inertia weight accordingly, calculating the local and global search capabilities and local and global learning factors of the particles, comprehensively considering the adaptive inertia weight, the adaptive local and global learning factors to update the particle speed, and updating the particle position; using the performance of the XGBoost prediction model on the validation set as a fitness evaluation; repeating the above steps until the stop condition is met to obtain the optimal parameter configuration.

5. A model for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma constructed by the method according to any one of claims 1 to 4.

6. A method for predicting the efficacy of radiotherapy and chemotherapy in patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment, characterized by: include: The original RNAseq sequencing data of the patient to be predicted is obtained, the expression matrix of the transcriptome data is normalized to obtain FPKM, and the radiotherapy and chemotherapy efficacy prediction model for esophageal squamous cell carcinoma patients as described in claim 5 is input to output the prediction result of the efficacy.

7. The method for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment according to claim 6, characterized in that: It also includes: predicting the pathological complete remission status pCR of the patient to be predicted, taking pCR as the outcome, and drawing the survival curve of ESCC patients in combination with the patient's OS or PFS survival time.

8. Device for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma, including: at least one processor; at least one memory for storing at least one program; It is characterized in that: when the at least one program is executed by the at least one processor, the at least one processor implements the method for constructing a model for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma as described in any one of claims 1-4 or the method for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment as described in claim 6 or 7.

9. A computer-readable storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to execute the method for constructing a model for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma as described in any one of claims 1 to 4, or the method for predicting the efficacy of radiotherapy and chemotherapy for patients with esophageal squamous cell carcinoma for non-disease diagnosis and treatment as described in claim 6 or 7 when executed by the processor.

Images