A rule-based learning-based system and method for predicting tumor radiotherapy response

Abstract

This invention discloses a rule-based learning-based system and method for predicting tumor radiotherapy responses. The prediction method includes steps such as data and image acquisition, tumor risk region stratification, medical feature extraction, and rule-based learning prediction model construction. First, a random forest is constructed and rules are generated based on feature data extracted from FDG PET / CT images of patients in the training set. Second, the Gray Wolf Optimizer (GWO) algorithm is used to prune the rules and select better rules. Finally, rule fitting is performed to train a linear model to predict radiotherapy responses in patients in the test set. The constructed prediction model can calculate the support and importance of each rule in the model, thereby providing clinical interpretability of the model. The model provided by this invention has good predictive performance and interpretable radiotherapy response prediction, thus better assisting doctors in making clinical decisions regarding radiotherapy for lung cancer patients.

Description

Technical Field

[0001] This invention relates to the field of medical rule learning, and more specifically, to a rule-based system and method for predicting tumor radiotherapy response. Background Technology

[0002] Radiotherapy, a key component of current cancer treatment, is less invasive and has fewer side effects compared to traditional surgery and chemotherapy. When used in combination with other treatments, it can maximize tumor control. However, current standard treatment regimens for cancer patients suffer from poor prognosis and limited tumor control. Early response assessment using fluorodeoxyglucose (FDG) positron emission tomography (PET) / computed tomography (CT) during radiotherapy has proven important in the treatment of non-small cell lung cancer. PET / CT can quantify changes in the tumor before and after treatment, prior to morphological changes, thus allowing for the tailoring of optimal treatment strategies for patients. The Standard Uptake Value (SUV) has significant prognostic value in cancer patient treatment, providing a reference for clinical treatment selection and prognostic assessment.

[0003] Scientific research has shown that feature data can be extracted from FDG PET / CT scans to more accurately assess and predict outcomes for patients with different cancers at different stages of treatment. Features, including tumor volume and the SUV (ultra-small volume) of the entire tumor, have been shown to be significantly correlated with treatment efficacy. Related studies have also indicated that FDG PET / CT parameters of the tumor, combined with tumor volume and metabolic activity, possess good predictive performance. Radiological features extracted using radiological image analysis techniques have been used to provide accurate and personalized treatment strategies. Radiological features, including FDG PET / CT image features of tissues surrounding the tumor, such as shape and texture, have been used to predict the effectiveness of radiotherapy in locally advanced non-small cell lung cancer. Current research focuses on extracting and evaluating radiomics features to predict patient clinical prognosis.

[0004] Machine learning models, such as Support Vector Machines (SVM) and Decision Trees (DT), have been widely used to predict patient clinical outcomes. Ensemble models, such as Random Forest (RF) and Gradient Boosting Decision Tree (GBDT), can handle complex feature datasets and improve prediction accuracy. However, the complexity of these models makes them "black boxes" lacking clinical interpretability, posing a significant challenge to translating scientific research into clinical applications. Random Forest is a Bagging-based ensemble learning model that enhances the model's decision-making ability by integrating multiple decision trees. A Random Forest consists of decision trees, where each decision tree model stores a set of information for each node in the form of an array. Based on this information, the RuleFit algorithm extracts rules from each decision tree, constructing an interpretable machine learning model for prediction through rule fitting. The quality of the selected rules affects the prediction accuracy of the constructed model, and currently, no efficient method for selecting high-quality rules has been developed. On the other hand, existing models rarely predict the SUV (Radioactive Advantage) value of tumor radiotherapy response, and a rule-based model with high accuracy in predicting SUV value changes has not yet been built.

[0005] Therefore, developing a rule-based method for predicting tumor radiotherapy responses that combines predictive accuracy and clinical interpretability is a pressing technical problem that needs to be solved. Summary of the Invention

[0006] Due to the aforementioned deficiencies in existing technologies, this invention provides a rule-based learning method for predicting tumor radiotherapy responses. First, this invention acquires tumor image data and performs tumor risk stratification. Second, this invention develops a rule-based machine learning model that extracts rules from FDG PET / CT radiomics of the entire tumor / tumor cluster sub-region using random forest, performs rule pruning using a heuristic algorithm, and constructs a model based on this rule set to predict radiotherapy responses. Finally, this invention compares and summarizes the results of each model in predicting SUV changes and treatment effectiveness categories, demonstrating that the adopted method can achieve good radiotherapy response prediction results while maintaining model interpretability.

[0007] To achieve the above objectives, in one aspect, the present invention provides a rule-based method for predicting tumor radiotherapy response, comprising the following steps:

[0008] Step S1, Image Data Acquisition and Tumor Risk Stratification: Acquire FDG PET / CT image data of patients in the pre-treatment (PETpre) and mid-treatment (PETmid) phases of the original dataset; FDG PET / CT image data includes, but is not limited to, the 3D spatial coordinates of tumor voxels and the standard uptake values ​​of tumor voxels; use the K-means clustering algorithm to cluster the tumor voxels of patients and divide them into tumor risk zones.

[0009] Step S2, Radiomics Feature Extraction: Feature data, including multiple radiomics features, are extracted from the FDG PET / CT image data of the above patients' PETpre and PETmid to characterize tumor metabolic activity in response to therapeutic radiation doses. The change in the tumor standard uptake value before and after radiotherapy is used to measure the radiotherapy response, including the FDG PET / CT mid-radiotherapy binary response class used to determine whether radiotherapy is effective and the FDG PET / CT mid-radiotherapy response change measure (ΔSUVmean) used to measure the radiotherapy effect.

[0010] Step S3: Construct a random forest model and generate rules: Construct a random forest based on the training set and extract an initial rule set from the random forest; the random forest consists of decision trees, where each decision tree model stores a set of information for each node in the form of an array. Based on this information, rules are extracted from each decision tree and stored.

[0011] Step S4: Use a heuristic algorithm to prune rules: Since the initial rule set contains a large number of rules, use the Grey Wolf Optimizer (GWO) algorithm to prune the rules and select better rules for prediction.

[0012] Step S5: Rule Fitting and Predicting Radiotherapy Response: The RuleFit algorithm is used to fit the pruned rule set. Each rule is treated as a basic learner, and an ensemble learning model is built based on the rule set. In this model, each input sample is converted into a 0-1 classification result based on whether it matches a rule. The 0-1 matrix composed of the 0-1 classification result of each sample corresponding to each rule and the original feature variables can be used as the new feature variables of the training set for model input.

[0013] To address different prediction needs, the least absolute shrinkage and selection operator (LASSO) is used to linearly combine the base learners derived from single rule transformations. A LASSO logistic regression model is employed for classification tasks, and a LASSO regression model for regression tasks. In classification prediction, the binary response class in the mid-stage of FDG PET / CT radiotherapy is used as the rule output, i.e., the prediction target; in regression prediction, ΔSUVmean is used as the prediction target. This constructs an interpretable machine learning model for radiotherapy response prediction.

[0014] Further, in step S1, the K-means clustering algorithm is used to generate clusters for each patient's pre-treatment and mid-treatment image time points. The n voxels are iteratively divided into k clusters to minimize the loss function corresponding to the clustering results. Based on the FDG PET / CT SUV and the 3D coordinates of the voxels, the tumor voxels of each patient are divided into 2 or 3 cluster sub-regions. The 2 cluster sub-regions are defined as the MTV high-risk region and the MTV low-risk region. The sub-regions with intermediate risk levels among the 3 cluster sub-regions are removed, and the sub-regions with higher and lower risk levels are re-clustered and defined as the MTV high* risk region and the MTV low* risk region.

[0015] Furthermore, in step S2, the radiomics features are extracted using the PET Oncology Radiomics Test Suite (PORTS) in the MATLAB toolbox, including 6 intensity histogram features, 19 texture features of the Gray Tone Spatial Dependent Matrix (GTSDM) class, 5 texture features of the Neighborhood Gray Tone Difference Matrix (NGTDM) class, and 11 texture features of the Gray Level Zone Size Matrix (GLZSM) class.

[0016] Further, in step S2, the formula for calculating the FDG PET / CT mid-stage response change metric (ΔSUVmean), used to measure the radiotherapy effect, is as follows:

[0017]

[0018] Wherein, PreSUVmean represents the average standard uptake value of all pixels before tumor radiotherapy; MidSUVmean represents the average standard uptake value of all pixels during tumor radiotherapy. The criterion for determining whether radiotherapy is effective is whether the average standard uptake value of all pixels before and during radiotherapy satisfies ΔSUVmean >= 0.2; if it satisfies this condition, the tumor radiotherapy targeting that pixel is considered effective; if it does not satisfy this condition, the tumor radiotherapy targeting that pixel is considered ineffective.

[0019] Furthermore, in step S4, in order to select a suitable number of rules from the selected rule set, an evaluation function is used to evaluate the selected rule set:

[0020]

[0021] For each rule in each rule set, cc represents the number of training samples that conform to the rule, and ic represents the number of training samples that do not conform to the rule. For each rule set, rule... cc This represents the sum of the cc values ​​of all rules in the rule set. ic This represents the sum of the ic values ​​of all rules in the rule set; rule count This represents the sum of the number of times each rule is included in the initial rule set (multiple tree classifiers may use the same rule), tree num This represents the total number of tree classifiers in the random forest; m and n are set constants, m is used to keep the denominator non-zero, usually taken as 4, and n is used to control the weights of the two parts, usually taken as 10.

[0022] Furthermore, in step S4, based on the evaluation function, the GWO algorithm is used to iteratively select better rules, thereby completing the pruning of the rule set; the algorithm pseudocode is shown in Table 1, which illustrates the algorithm flow.

[0023] Table 1. Pseudocode of the GWO algorithm

[0024]

[0025]

[0026] Furthermore, in the GWO algorithm, the selected rule set is represented by the position of the gray wolf, and the optimal rule set is represented by the prey. The algorithm simulates the hunting behavior of the gray wolf to filter out high-quality rules. For an initial rule set containing k rules, the GWO algorithm randomly generates p uniformly distributed weight vectors representing the initial wolf pack of p gray wolves, i.e., the p selected rule sets. Each vector (x1, x2, x3, ..., x...) qThe vector contains q elements, representing the positions of individual gray wolves, and represents a rule set containing q rules. Each element x in each vector... i It is a random number between 0 and 1, used to represent the |kx-th rule selected from the initial rule set. i | Rules.

[0027] Furthermore, the iterative part of the GWO algorithm mainly includes the following steps. First, the gray wolf (for each selected rule set) gradually approaches and surrounds the prey. Here, t is the current iteration number, and A and C are the coordination coefficient vectors. A and C are calculated according to the formula described above:

[0028] A = 2a × r1 - a

[0029] C = 2r²

[0030] A changes linearly from 2 to 0 during the iteration process, and r1 and r2 are random vectors between 0 and 1. X(t) represents the current position vector of the gray wolf, and X(t+1) represents the updated position vector of the gray wolf. The gray wolf uses the guidance of α wolf, β wolf, and δ wolf to find prey and approach it, updating its position vector according to the formula.

[0031] D α =C1×X α -X(t),X1=X α -A1×D α

[0032] D β =C2×X β -X(t),X2=X β -A2×D β

[0033] D δ =C3×X δ -X(t),X3=X δ -A3×D δ

[0034] X(t+1)=(X1+X2+X3) / 3

[0035] Where X α X β X δ This represents the positions of the α wolf, β wolf, and δ wolf. (D) α D β D δLet A and B represent the distances between the current wolf and the optimal three wolves. The gray wolf updates its position and moves towards its prey to obtain a better rule set. Subsequently, when |A| < 1, the gray wolf attacks the prey and updates its position; when |A| > 1, the gray wolf performs a random search. After 100 iterations as described, the optimal gray wolf individual is obtained, thus completing rule pruning and obtaining the optimal rule set.

[0036] Furthermore, the number of rules obtained from the pruning should ideally be between 15 and 20. Too few rules will make it difficult to achieve high prediction accuracy, while too many rules will make the model too complex and difficult to construct.

[0037] Furthermore, in step S5, the constructed model is interpretable; the constructed model can evaluate the rules that make up the model, and explain how these rules and features work in the integrated model by calculating the support and importance of each rule and the importance of the features contained therein.

[0038] A second aspect of the present invention provides a rule-learning-based tumor radiotherapy response prediction system, comprising:

[0039] The image data preprocessing module processes the acquired FDG PET / CT image data from the pre-treatment (PETpre) and mid-treatment (PETmid) phases of the patient's treatment. It uses a K-means clustering algorithm to cluster the patient's tumor voxels, delineating tumor risk zones. Feature data, including 41 radiomics features, is extracted from the aforementioned FDG PET / CT image data from the PETpre and PETmid phases. The change in the tumor's standardized uptake value before and after radiotherapy is calculated to measure radiotherapy response, including the FDG PET / CT mid-treatment binary response class used to determine radiotherapy effectiveness and the FDG PET / CT mid-treatment response change metric (ΔSUVmean) used to measure radiotherapy efficacy. Finally, the resulting dataset is divided into training and testing sets.

[0040] The model building module consists of three parts: rule generation, rule trimming, and rule fitting. The first part generates an initial rule set based on the feature dataset obtained from image data preprocessing. The second part uses the GWO algorithm to trim the rules and obtain a better rule set. The third part uses the RuleFit algorithm to fit the rules and build an interpretable machine learning model.

[0041] The training module trains the machine learning model using the training set data.

[0042] The prediction module includes a trained machine learning model and rules for building the model. It is used to input the data to be predicted, i.e. the test set data, into the trained machine learning model to generate prediction results.

[0043] The model analysis module evaluates the rules that make up the model based on the interpretability of the model; it explains how these rules and features work in the integrated model by calculating the support and importance of each rule and the importance of the features it contains.

[0044] Compared with the prior art, the above invention has the following advantages or beneficial effects:

[0045] (1) This invention extracts features from image data and combines the feature data with random forest generation rules to provide interpretable radiotherapy response prediction;

[0046] (2) This invention cleverly uses the GWO algorithm for rule pruning, efficiently selects high-quality rule sets for prediction, and improves the accuracy of prediction;

[0047] (3) This invention utilizes the improved RuleFit algorithm to construct a rule-based radiotherapy response prediction model, which has both interpretability and high prediction accuracy. It can intelligently assist doctors in diagnosing tumor radiotherapy and promote clinical decision support in terms of patient risk stratification, treatment selection and biologically driven adaptation. Attached Figure Description

[0048] The invention, its features and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart of the steps in a tumor radiotherapy response prediction method according to an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of a tumor radiotherapy response prediction system in one embodiment of the present invention;

[0051] Figure 3 This is a flowchart illustrating the steps of constructing a prediction model in a tumor radiotherapy response prediction method according to an embodiment of the present invention.

[0052] Figure 4 This is a flowchart of the GWO algorithm in one embodiment of the present invention. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0054] The terms “comprising,” “including,” and any variations thereof are intended to cover non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units that is not limited to the steps or units listed, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0055] In the following detailed description, numerous specific details are set forth to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that well-known control algorithms are not shown in detail to avoid obscuring the gist of the invention; and that the K-means algorithm, feature extraction, rule generation, GWO algorithm, RuleFit algorithm, and other techniques involved in the following effect examples are readily available prior art.

[0056] Example 1

[0057] like Figure 1 and 3 As shown, this embodiment provides a rule-based learning-based method for predicting tumor radiotherapy response, used to predict the FDG PET / CT mid-stage binary response class that measures the effectiveness of radiotherapy, including the following steps:

[0058] Step S1, Image Data Acquisition and Tumor Risk Stratification: Acquire FDG PET / CT image data of patients in the pre-treatment (PETpre) and mid-treatment (PETmid) phases of the raw dataset; FDG PET / CT image data includes the 3D spatial coordinates of tumor voxels, the standard uptake values ​​of tumor voxels, etc.

[0059] The K-means clustering algorithm was used to generate clusters for each patient's pre-treatment and mid-treatment image time points. The n voxels were divided into k clusters through iteration to minimize the loss function corresponding to the clustering results. Based on FDG PRTSUV and the 3D coordinates of the voxels, the tumor voxels of each patient were divided into 2 or 3 cluster sub-regions. The 2 cluster sub-regions were defined as MTV high-risk region and MTV low-risk region. The sub-regions with intermediate risk levels among the 3 cluster sub-regions were removed, and the sub-regions with high risk levels and low risk levels were re-clustered and defined as MTV high* risk region and MTV low* risk region.

[0060] Step S2, Radiomics Feature Extraction: Feature data was extracted from the FDG PET / CT image data of the above patients' PETpre and PETmid, including 41 radiomics features. Among them, the 41 radiomics features were extracted using the PET Oncology Radiomics Test Suite (PORTS) in the MATLAB toolbox, including 6 intensity histogram features, 19 texture features of the Gray Tone Spatial Dependent Matrix (GTSDM) class, 5 texture features of the Neighborhood Gray Tone Difference Matrix (NGTDM) class, and 11 texture features of the Gray Level Zone Size Matrix (GLZSM) class.

[0061] The formula for calculating the FDG PET / CT mid-term response change measure (ΔSUVmean), used to measure the effectiveness of radiotherapy, is as follows:

[0062]

[0063] Wherein, PreSUVmean represents the average standard uptake value of all pixels before tumor radiotherapy; MidSUVmean represents the average standard uptake value of all pixels during tumor radiotherapy. The criterion for determining whether radiotherapy is effective is whether the average standard uptake value of all pixels before and during radiotherapy satisfies ΔSUVmean >= 0.2; if it satisfies this condition, the tumor radiotherapy targeting that pixel is considered effective; if it does not satisfy this condition, the tumor radiotherapy targeting that pixel is considered ineffective.

[0064] Step S3: Construct a random forest model and generate rules: Construct a random forest based on the training set and extract an initial rule set from the random forest; Each decision tree model in the random forest stores a set of information for each node in the form of an array, and based on this information, extract and store rules from each decision tree.

[0065] Step S4: Perform rule pruning using a heuristic algorithm: Since the initial rule set contains a large number of rules, the GWO algorithm is used to prune the rules and select better rules for prediction (e.g., ...). Figure 4 (As shown); The selected rule set is evaluated using the evaluation function:

[0066]

[0067] For each rule in each rule set, cc represents the number of training samples that conform to the rule, and ic represents the number of training samples that do not conform to the rule. For each rule set, rule... ccThis represents the sum of the cc values ​​of all rules in the rule set. ic This represents the sum of the ic values ​​of all rules in the rule set; rule count This represents the sum of the number of times each rule is included in the initial rule set (multiple tree classifiers may use the same rule), tree num This represents the total number of tree classifiers in the random forest; m and n are set constants, m is used to keep the denominator non-zero, usually taken as 4, and n is used to control the weights of the two parts, usually taken as 10.

[0068] Step S5: Rule Fitting and Radiotherapy Response Prediction: The RuleFit algorithm is used to fit the pruned rule set, with each rule treated as a basic learner. An ensemble learning model is built based on the rule set. Each input sample is converted into a 0-1 classification result based on whether it matches a rule. The 0-1 matrix, composed of the 0-1 classification result for each sample corresponding to each rule and the original feature variables, serves as the new feature variables for the training set input to the model. The basic learners transformed from individual rules are linearly combined, and a LASSO logistic regression model is used to predict whether radiotherapy is effective. The effectiveness of radiotherapy is used as the output of the rule, i.e., the prediction target, for classification prediction.

[0069] like Figure 2 As shown, this embodiment employs a rule-based learning-based tumor radiotherapy response prediction system to implement the above prediction method. The prediction system includes:

[0070] The image preprocessing module is used to process the acquired FDG PET / CT image data of patients in the pre-treatment (PETpre) and mid-treatment (PETmid) stages, and uses the K-means clustering algorithm to cluster the tumor voxels of patients and divide the tumor risk areas.

[0071] Feature data, including 41 radiomics features, were extracted from the FDG PET / CT image data of the aforementioned patients' PETpre and PETmid images. Changes in tumor standardized uptake values ​​before and after radiotherapy were calculated to measure radiotherapy response, specifically the FDG PET / CT mid-radiotherapy binary response class used to determine the effectiveness of radiotherapy. Finally, the resulting dataset was divided into training and testing sets.

[0072] The model building module consists of three parts: rule generation, rule trimming, and rule fitting. The first part generates an initial rule set based on the feature dataset obtained from image data preprocessing. The second part uses the GWO algorithm to trim the rules and obtain a better rule set. The third part uses the RuleFit algorithm to fit the rules and build an interpretable machine learning model.

[0073] The training module trains the machine learning model using the training set data;

[0074] The prediction module includes a trained machine learning model and rules for building the model. It is used to input the data to be predicted, i.e. the test set data, into the trained machine learning model to generate prediction results.

[0075] The model analysis module, based on the interpretability of the model, is used to evaluate the rules that make up the model. It interprets how these rules and features function within the ensemble model by calculating the support and importance of each rule, as well as the importance of the features it contains.

[0076] The prediction system processes the acquired tumor image data using the prediction method to predict the mid-stage binary response class of FDG PET / CT radiotherapy, thus determining whether the radiotherapy is effective. The area under the curve (AUC) is selected as the evaluation metric for classification prediction. AUC, the area under the receiver operating characteristic (ROC) curve, is commonly used as a model evaluation standard; a larger AUC indicates better model classification performance.

[0077] Leave-one-out cross-validation was used for model validation. The validation process involved reserving data from one patient for testing each time, using the remaining patients' data for training, and then using the trained model to predict test samples. Common machine learning methods were used as control methods, including Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT). Based on the image data acquisition and tumor risk stratification methods described above, patient tumors were divided into risk regions, resulting in datasets for five risk regions: MTV, MTV high, MTV low, MTV high*, and MTV low*, used for prediction. Table 2 shows the prediction results of the rule-based learning method and common machine learning methods for the binary reaction class in mid-stage FDG PET / CT radiotherapy. As can be seen from the table, the rule-based learning method achieves stable and relatively good AUC, achieving an AUC of 0.86 in the MTV low* risk region.

[0078] Table 2. Predictive Results of Mid-Term Binary Reactions in FDG PET / CT Radiotherapy

[0079]

[0080] Based on the interpretability of rule-based learning methods, this invention can evaluate the rules constituting the model. Taking a model trained on the entire tumor as an example, this model consists of 15 rules, where X[i] represents the value of feature i. As shown in Table 3, the support and importance of each rule explain how these rules work in the ensemble model. Importance reflects the impact of each rule in the model on the decision outcome, while support is the proportion of data in the dataset that conforms to that rule. Rule-based learning methods can further analyze the rule set and calculate the importance of the features it contains. Table 4 shows the importance of the features, where features 4 (KURT) and 22 (CS) have high importance in the model (1.419 and 4.132, respectively). These evaluations and analyses explain the role of each rule and feature in the model, contributing to more accurate medical diagnosis.

[0081] Table 3 shows the rule set included in the rule learning model.

[0082]

[0083]

[0084] Table 4 shows the features included in the rule set.

[0085]

[0086] Example 2

[0087] This embodiment uses a prediction method similar to that of Embodiment 1 to predict radiotherapy response. The difference from Embodiment 1 is that this embodiment predicts the FDG PET / CT mid-radiotherapy response change metric (ΔSUVmean), which measures the effectiveness of radiotherapy. Therefore, a rule-based regression prediction method is used, with the following steps:

[0088] Step S1, Image Data Acquisition and Tumor Risk Stratification: Acquire FDG PET / CT image data of patients in the pre-treatment (PETpre) and mid-treatment (PETmid) phases of the original dataset; FDG PET / CT image data includes the 3D spatial coordinates of tumor voxels, the standard uptake values ​​of tumor voxels, etc.; use the K-means clustering algorithm to cluster the tumor voxels of patients and divide the tumor risk zone.

[0089] Step S2, Radiomics Feature Extraction: Feature data, including 41 radiomics features, are extracted from the FDG PET / CT image data of the patients' PETpre and PETmid images. The change in tumor standard uptake value before and after radiotherapy is used to measure the radiotherapy response, i.e., the FDG PET / CT mid-radiotherapy response change measure (ΔSUVmean) used to measure the radiotherapy effect. The formula for calculating ΔSUVmean is:

[0090]

[0091] Wherein, PreSUVmean represents the average standard uptake value of all pixels before tumor radiotherapy; MidSUVmean represents the average standard uptake value of all pixels during tumor radiotherapy.

[0092] Step S3: Construct a random forest model and generate rules: Construct a random forest based on the training set and extract an initial rule set from the random forest; Each decision tree model in the random forest stores a set of information for each node in the form of an array, and based on this information, extract and store rules from each decision tree.

[0093] Step S4: Use a heuristic algorithm to prune rules: Since the initial rule set contains a large number of rules, the GWO algorithm is used to prune rules, just like in Example 1, to select better rules for prediction.

[0094] Step S5: Rule Fitting and Radiotherapy Response Prediction: The RuleFit algorithm is used to fit the pruned rule set. Each rule is treated as a basic learner, and an ensemble learning model is built based on the rule set. Each input sample is converted into a 0-1 classification result based on whether it matches a rule. The 0-1 matrix consisting of the 0-1 classification result for each sample corresponding to each rule and the original feature variables serves as the new feature variables for the training set input to the model. The basic learners transformed from individual rules are linearly combined, and a LASSO regression model is used to predict whether radiotherapy is effective. The mid-term response change metric (ΔSUVmean) of FDG PET / CT radiotherapy is used as the rule output, i.e., the prediction target, for regression prediction.

[0095] This embodiment employs a rule-based learning-based tumor radiotherapy response prediction system to implement the above prediction method. The prediction system includes:

[0096] The image preprocessing module is used to process the acquired FDG PET / CT image data of patients in the pre-treatment (PETpre) and mid-treatment (PETmid) stages, and uses the K-means clustering algorithm to cluster the tumor voxels of patients and divide the tumor risk areas.

[0097] Feature data, including 41 radiomics features, were extracted from the FDG PET / CT image data of the above patients' PETpre and PETmid. The change in the tumor standard uptake value before and after radiotherapy was calculated to measure the radiotherapy response, namely the FDG PET / CT radiotherapy mid-term response change measure (ΔSUVmean). Finally, the obtained dataset was divided into training set and test set.

[0098] The model building module consists of three parts: rule generation, rule trimming, and rule fitting. The first part generates an initial rule set based on the feature dataset obtained from image data preprocessing. The second part uses the GWO algorithm to trim the rules and obtain a better rule set. The third part uses the RuleFit algorithm to fit the rules and build an interpretable machine learning model.

[0099] The training module trains the machine learning model using the training set data;

[0100] The prediction module includes a trained machine learning model and rules for building the model. It is used to input the data to be predicted, i.e. the test set data, into the trained machine learning model to generate prediction results.

[0101] The model analysis module, based on the interpretability of the model, is used to evaluate the rules that make up the model. It interprets how these rules and features function within the ensemble model by calculating the support and importance of each rule, as well as the importance of the features it contains.

[0102] The prediction system, after processing the acquired tumor image data using the prediction method, predicted the mid-term response change measure (ΔSUVmean) of FDG PET / CT radiotherapy. The root mean square error (RMSE) was selected as the evaluation index for regression prediction; it is calculated by dividing the sum of the squares of the deviations between the predicted and actual values ​​by the square root of the number of observations, and is commonly used to measure the deviation between the predicted and actual values.

[0103] Leave-one-out cross-validation was used for model validation. The validation process involved reserving data from one patient for testing each time, using the remaining patients' data for training, and then using the trained model to predict test samples. Common machine learning methods were used as control methods, including Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT). Based on the image data acquisition and tumor risk stratification methods described above, patient tumors were divided into risk regions, resulting in datasets for five risk regions: MTV, MTV high, MTV low, MTV high*, and MTV low*, used for prediction. Table 5 shows the prediction results of the FDG PET / CT radiotherapy mid-term response change metric (ΔSUVmean) using the rule-based learning method of this invention and common machine learning methods. As can be seen from the table, the rule-based learning method achieves stable and low RMSE, achieving an RMSE of 0.189 in the MTV high* risk region.

[0104] Table 5. Predicted results of FDG PET / CT mid-term response change measure (ΔSUVmean) for radiotherapy

[0105]

[0106] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention, and will not be elaborated here either.

[0107] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention's technical solutions still fall within the protection scope of the present invention.

Claims

1. A method for predicting tumor radiotherapy response based on rule learning, characterized in that, Includes the following steps: Step S1, Image Data Acquisition and Tumor Risk Stratification: Acquire FDG PET / CT image data of patients in the pre-treatment PETpre and mid-treatment PETmid phases from the original dataset; FDG PET / CT image data includes, but is not limited to, the 3D spatial coordinates of tumor voxels and the standard uptake values ​​of tumor voxels; use the K-means clustering algorithm to cluster the tumor voxels of patients and divide the tumor risk zone; Step S2, Radiomics Feature Extraction: Feature data, including multiple radiomics features, are extracted from the FDG PET / CT image data of the patient's PETpre and PETmid to characterize tumor metabolic activity in response to therapeutic radiation dose; the change in the tumor standard uptake value SUV before and after radiotherapy is used to measure the radiotherapy response, including the FDG PET / CT mid-radiotherapy binary response class used to determine whether radiotherapy is effective and the FDG PET / CT mid-radiotherapy response change metric ΔSUVmean used to measure the radiotherapy effect; Step S3: Construct a random forest model and generate rules: Construct a random forest based on the training set and extract an initial rule set from the random forest; Step S4: Use a heuristic algorithm to prune rules: Use GWO to prune the initial rule set and select the optimal rules for prediction; Step S5: Rule Fitting and Radiotherapy Response Prediction: The RuleFit algorithm is used to fit the pruned rule set. Each rule is treated as a basic learner, and an ensemble learning model is built based on the rule set. In this model, each input sample is converted into a 0-1 classification result based on whether it matches a rule. A 0-1 matrix consisting of the 0-1 classification result of each sample corresponding to each rule and the original feature variables is used as the new feature variables of the training set for model input. For different prediction needs, the LASSO linear model is used to linearly combine the basic learners transformed from individual rules. The LASSO logistic regression model is used for classification tasks, and the LASSO regression model is used for regression tasks. In classification prediction, the binary response class in the mid-stage of FDG PET / CT radiotherapy is used as the output of the rule, i.e., the prediction target. In regression prediction, ΔSUVmean is used as the prediction target. Thus, an interpretable machine learning model is constructed for radiotherapy response prediction. In step S4, in order to select a suitable number of rules from the selected rule set, an evaluation function is used to evaluate the selected rule set: ， For each rule in each rule set, This indicates the number of training samples that conform to this rule. This represents the number of samples in the training set that do not conform to the rule; for each rule set, This indicates the rules in the rule set. the sum of This indicates the rules in the rule set. sum; This represents the sum of the number of times each rule is included in the initial rule set. Multiple tree classifiers may use the same rule. This represents the total number of tree classifiers in the random forest; and For the set constant, To keep the denominator non-zero, we take 4, and n is used to control the weights of the two parts, so we take 10. Based on the evaluation function, the GWO (Grey Wolf) optimization algorithm is used to iteratively select better rules, thereby completing the pruning of the rule set.

2. The method for predicting tumor radiotherapy response based on rule learning according to claim 1, characterized in that, In step S1, the K-means clustering algorithm is used to generate clusters of images at time points before and during treatment for each patient; based on the 3D coordinates of FDG PET / CT SUV and voxels, voxels are divided into 2 or 3 cluster sub-regions; the 2 cluster sub-regions are defined as MTV high-risk region and MTV low-risk region; the sub-regions with intermediate risk levels among the 3 cluster sub-regions are removed, and the sub-regions with higher risk levels and lower risk levels are re-clustered and defined as MTV high* risk region and MTV low* risk region.

3. The method for predicting tumor radiotherapy response based on rule learning according to claim 1, characterized in that, In step S2, radiomics features are extracted using the PET oncology radiomics test suite PORTS in the MATLAB toolbox, including 6 intensity histogram features, 19 texture features of the GTSDM class, 5 texture features of the NGTDM class, and 11 texture features of the GLZSM class.

4. The method for predicting tumor radiotherapy response based on rule learning according to claim 1, characterized in that, In step S2, the formula for calculating the FDG PET / CT mid-term radiotherapy response change ΔSUVmean, used to measure the radiotherapy effect, is as follows: ， Wherein, PreSUVmean represents the average standard uptake value of all pixels before tumor radiotherapy; MidSUVmean represents the average standard uptake value of all pixels during tumor radiotherapy; the criterion for determining whether radiotherapy is effective is whether the average standard uptake value of all pixels before tumor radiotherapy and the average standard uptake value of all pixels during radiotherapy satisfy ΔSUVmean>=0.2; if satisfied, the tumor radiotherapy for the pixel is deemed effective; if not satisfied, the tumor radiotherapy for the pixel is deemed ineffective.

5. The method for predicting tumor radiotherapy response based on rule learning according to claim 1, characterized in that, The number of rules obtained from the pruning is 15-20.

6. The method for predicting tumor radiotherapy response based on rule learning according to claim 1, characterized in that, In step S5, the constructed model is interpretable; the constructed model can evaluate the rules that make up the model, and explain how these rules and features work in the integrated model by calculating the support and importance of each rule and the importance of the features contained therein.

7. A rule-based learning-based tumor radiotherapy response prediction system, used to implement the rule-based learning-based tumor radiotherapy response prediction method of claim 1, characterized in that, include: The image data preprocessing module is used to process the FDG PET / CT image data of the patient's pre-treatment PETpre and mid-treatment PETmid, and to use the K-means clustering algorithm to cluster the patient's tumor voxels and divide the tumor risk area; and to extract feature data, including multiple radiomics features, from the patient's PETpre and PETmid FDG PET / CT image data. The change in tumor standard uptake values ​​before and after radiotherapy was calculated to measure radiotherapy response, including the FDG PET / CT mid-radiotherapy binary response class used to determine whether radiotherapy was effective and the FDG PET / CT mid-radiotherapy response change metric ΔSUVmean used to measure the radiotherapy effect; finally, the obtained dataset was divided into training set and test set. The model building module consists of three parts: rule generation, rule trimming, and rule fitting. The first part generates an initial rule set based on the feature dataset obtained from image data preprocessing. The second part uses the GWO algorithm to trim the rules and obtain a better rule set. The third part uses the RuleFit algorithm to fit the rules and build an interpretable machine learning model. The training module trains the machine learning model using the training set data; The prediction module includes a trained machine learning model and rules for building the model. It is used to input the data to be predicted, i.e. the test set data, into the trained machine learning model to generate prediction results. The model analysis module evaluates the rules that make up the model based on the interpretability of the model; it explains how these rules and features work in the integrated model by calculating the support and importance of each rule and the importance of the features it contains.

Images