Two-stage clustering diagnosis and automatic machine learning prediction method and model

Abstract

The embodiment of the invention provides a two-stage clustering diagnosis and automatic machine learning prediction method and model. The method comprises the following steps: performing feature weighting on 3D space coordinates and an SUV in a tumor PET image to realize first-stage clustering; performing voxel-level clustering analysis on a tumor region in the tumor PET image to realize second-stage clustering, and sketching tumor sub-regions with different metabolic characteristics; performing feature extraction on the whole tumor and different sub-regions based on the tumor sub-regions drawn by clustering in the two stages; constructing an automatic machine learning prediction algorithm based on the extracted features, and predicting labels in the classification task and numerical values in the regression task; and on the basis of a Kernel SHAP method, carrying out interpretability analysis on the feature importance of a prediction result. According to the method and the model provided by the invention, the high-risk and low-risk subregions of the tumor can be accurately identified and predicted and analyzed, so that the accuracy of a personalized radiotherapy plan is improved, and the clinical prognosis of a patient is improved.

Description

Technical Field

[0001] The present application relates to the field of medical machine learning technology, and in particular to a two-stage clustering diagnosis and automatic machine learning prediction method and model. Background Art

[0002] Tumor heterogeneity is one of the characteristics of malignant tumors, which can cause differences in tumor growth rate, invasion and metastasis, drug sensitivity, prognosis, etc. For example, some tumor areas may be sensitive to radiotherapy, while other areas may show drug resistance. Therefore, accurately delineating the subregions of the tumor and dividing the tumor into different biological characteristic areas is the basis for personalized radiotherapy, which can guide more accurate targeted radiotherapy strategies. In addition, the standard uptake value (SUV) is a semi-quantitative indicator commonly used in positron emission computed tomography (PET) in tumor diagnosis. There is a significant correlation between the SUV value of the tumor before radiotherapy and the radiotherapy effect and survival rate of the patient. The change of SUV value during and after radiotherapy can also provide important information about the tumor's response to radiotherapy. Therefore, by analyzing the dynamic changes of SUV values, the patient's radiotherapy effect can be predicted more accurately, thereby guiding personalized radiotherapy strategies. Traditional machine learning methods have been widely used in the prediction of tumor radiotherapy response, but there are still challenges, especially in model selection and hyperparameter adjustment. These steps usually require a lot of manual work and deep expertise, and traditional machine learning models usually lack sufficient interpretability, which makes it more difficult to understand the logic behind the model predictions. Therefore, it is crucial to develop machine learning algorithm models that can provide higher accuracy and better interpretability to predict and analyze changes in tumor and its subregional SUV. Summary of the invention

[0003] The embodiments of the present application provide a two-stage clustering diagnosis and automatic machine learning prediction method and model, which can accurately identify high-risk and low-risk sub-regions of tumors and perform predictive analysis, thereby improving the accuracy of personalized radiotherapy plans and improving the patient's clinical prognosis.

[0004] To solve the above technical problems, in the first aspect, an embodiment of the present application provides a two-stage clustering diagnosis and automatic machine learning prediction method, which includes the following steps: first, based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET image are feature weighted to achieve the first stage clustering; then, based on the New_mid K-means clustering algorithm, the tumor area in the tumor PET image is clustered at the voxel level to achieve the second stage clustering, and tumor sub-regions with different metabolic characteristics are outlined; next, based on the tumor sub-regions outlined by the two-stage clustering, features of the entire tumor and different sub-regions are extracted; then, based on the extracted features, an automatic machine learning prediction algorithm is constructed, and based on the automatic machine learning prediction algorithm, the labels in the classification task and the values ​​in the regression task are predicted to obtain prediction results; finally, based on the Kernel SHAP method, the feature importance of the prediction results is analyzed for interpretability.

[0005] In some exemplary embodiments, based on the K-means clustering algorithm, feature weighting is performed on the 3D spatial coordinates and SUV in the tumor PET image to achieve first-stage clustering, including: using a stepping method to perform multiple groups of weightings on the 3D spatial coordinates and SUV features in the tumor PET image to obtain multiple weight combination clustering results; based on the K-means clustering algorithm, a clustering evaluation index is calculated for each weight combination clustering result; based on the clustering evaluation index, a final feature weighting is performed on the 3D spatial coordinates and SUV in the tumor PET image.

[0006] In some exemplary embodiments, a stepping method is used to perform multiple weightings on the 3D spatial coordinates and SUV features in the tumor PET image to obtain multiple weight combination clustering results, including: using the stepping method to take the weight of the SUV feature in the tumor PET image from 0.1 to 1.0 in steps of 0.1; according to the corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight, multiple weightings are performed on the 3D spatial coordinate features in the tumor PET image to obtain multiple weight combination clustering results; the corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight is:

[0007] w x =w y =w z =(1-w SUV ) / 3

[0008] Among them, w x represents the x-coordinate feature weight in the tumor PET image; w y represents the y-coordinate feature weight in the tumor PET image; w z represents the z-coordinate feature weight in the tumor PET image; wSUV Represents the SUV feature weight in the tumor PET image.

[0009] In some exemplary embodiments, based on the K-means clustering algorithm, the clustering evaluation index of each weight combination clustering result is calculated, including: substituting each weight combination into the K-means clustering algorithm, and calculating the comprehensive score as the clustering evaluation index of the weight combination clustering result; the calculation formula of the comprehensive score is:

[0010]

[0011] CS represents the comprehensive score; SC pre SC represents the clustering silhouette coefficient of tumors in the early stage of radiotherapy; mid DS represents the tumor clustering silhouette coefficient in the middle of radiotherapy; high Dice similarity of high-risk subregions in the early and middle stages of radiotherapy; DS low Represents the Dice similarity of low-risk subregions in the early and middle stages of radiotherapy.

[0012] In some exemplary embodiments, based on the clustering evaluation index, the 3D spatial coordinates and SUV in the tumor PET image are finally feature weighted, including: for a sample, based on the clustering evaluation index of each weight combination clustering result, the SUV feature weight in the tumor PET image corresponding to the optimal clustering evaluation index is selected; for all samples, based on the SUV feature weight in the tumor PET image corresponding to the optimal clustering evaluation index selected for each sample, the average value is calculated to obtain the final SUV feature weight in the tumor PET image for all samples, and according to the correspondence between the 3D spatial coordinates in the tumor PET image and the SUV feature weight, the final 3D spatial coordinate feature weight in the tumor PET image for all samples is calculated.

[0013] In some exemplary embodiments, based on the New_mid K-means clustering algorithm, voxel-level clustering analysis is performed on the tumor area in the tumor PET image to achieve second-stage clustering, and tumor sub-regions with different metabolic characteristics are outlined, including: based on the K-means clustering algorithm, for the clustering analysis of the mid-stage radiotherapy tumor area, the SUV features before radiotherapy and the SUV features during mid-stage radiotherapy are fused and input for voxel-level clustering analysis to construct a New_mid K-means clustering algorithm; based on the New_mid K-means clustering algorithm, the optimal weight of the mid-stage radiotherapy SUV feature is selected by step-by-step debugging; wherein, when selecting, the weight of the mid-stage radiotherapy SUV feature is greater than or equal to the weight of the early stage radiotherapy SUV feature, so as to avoid ignoring the data aggregation degree itself in order to obtain better coincidence stability during the clustering analysis of the mid-stage radiotherapy tumor area.

[0014] In some exemplary embodiments, based on the tumor subregions outlined by two-stage clustering, feature extraction is performed on the entire tumor and different subregions, including: the features extracted from the entire tumor and different subregions include traditional features and radiomics features; the traditional features include the volume of the entire tumor and different subregions, the average SUV value of the entire tumor and different subregions, and the radiotherapy dose of the entire tumor and different subregions; the radiomics features include 6 intensity histogram features, 19 texture features from the grayscale tone spatial dependence matrix, 5 texture features from the neighborhood grayscale tone difference matrix, and 11 texture features from the grayscale region size matrix, a total of 41 radiomics features.

[0015] In some exemplary embodiments, an automatic machine learning prediction algorithm is constructed based on the extracted features, including: building an automatic machine learning core framework based on the extracted features; the automatic machine learning core framework includes a meta-learning module, a Bayesian optimization module and an integrated learning module; wherein the meta-learning module uses metadata of past data sets and model performance to guide model selection and hyperparameter initialization for new tasks; the Bayesian optimization module predicts performance under different configurations by constructing a probabilistic model, thereby guiding the search and tuning process of hyperparameters; the integrated learning module assigns weights to each model based on the model's performance indicators, and then integrates them into a powerful integrated model, which can effectively reduce overfitting and improve the model's generalization ability on unknown data.

[0016] In some exemplary embodiments, based on the Kernel SHAP method, an interpretability analysis is performed on the feature importance of the prediction result, including: based on the SHAP method, fusing kernel regression and weight sampling technology to construct the Kernel SHAP method; based on the Kernel SHAP method, the Shapley value estimation problem is converted into a weighted least squares regression problem, and the contribution value of each feature is estimated using efficient linear model solving technology; based on the contribution value of each feature, an interpretability analysis is performed on the feature importance of the prediction result.

[0017] In the second aspect, the embodiment of the present application also provides a two-stage clustering diagnosis and automatic machine learning prediction model, including a two-stage clustering module and a prediction analysis module connected in sequence; the two-stage clustering module includes a feature weighting module and a New_mid K-means clustering module; the feature weighting module is used to perform feature weighting on the 3D spatial coordinates and SUV value features in the tumor PET image according to the K-means clustering algorithm to achieve the first stage clustering; the New_mid K-means clustering module is used to perform voxel-level clustering analysis on the tumor area in the tumor PET image according to the New_mid K-means clustering algorithm to achieve the second stage clustering, and outline tumor sub-regions with different metabolic characteristics; the prediction analysis module includes a feature extraction module, an automatic machine learning module and a Kernel SHAP interpretability analysis module; the feature extraction module is used to extract features of the entire tumor and different sub-regions according to the tumor sub-regions outlined by the two-stage clustering; the automatic machine learning module is used to construct an automatic machine learning prediction algorithm based on the extracted features, and predict the labels in the classification task and the values ​​in the regression task based on the automatic machine learning prediction algorithm to obtain prediction results; The SHAP interpretability analysis module is used to perform interpretability analysis on the feature importance of prediction results based on the Kernel SHAP method.

[0018] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0019] The embodiment of the present application provides a two-stage clustering diagnosis and automatic machine learning prediction method and model, which includes the following steps: first, based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET image are feature weighted to achieve the first stage clustering; then, based on the New_mid K-means clustering algorithm, the tumor area in the tumor PET image is clustered at the voxel level to achieve the second stage clustering, and tumor sub-regions with different metabolic characteristics are outlined; next, based on the tumor sub-regions outlined by the two-stage clustering, features of the entire tumor and different sub-regions are extracted; then, based on the extracted features, an automatic machine learning prediction algorithm is constructed, and based on the automatic machine learning prediction algorithm, the labels in the classification task and the values ​​in the regression task are predicted to obtain prediction results; finally, based on the KernelSHAP method, the feature importance of the prediction results is analyzed for interpretability.

[0020] This application proposes a two-stage clustering diagnosis and automatic machine learning prediction method and model. First, this application uses a two-stage clustering algorithm to efficiently cluster tumor PET images, thereby distinguishing high- and low-risk sub-regions with different metabolic characteristics inside the tumor, which can provide data support for the clinical formulation of more detailed radiotherapy strategies. Secondly, this application uses traditional features and imaging genomics features to extract and combine the data obtained from voxel clustering and the patient's clinical information, and uses an automatic machine learning prediction algorithm to construct an accurate radiotherapy response prediction model, which can automatically select the optimal feature combination and adopt the most appropriate machine learning algorithm for integration, thereby achieving accurate prediction of the patient's radiotherapy response. Finally, this application verifies the excellent performance of the model through the application results of patients with locally advanced non-small cell lung cancer, showing its significant advantages in accurate segmentation of tumor sub-regions and prediction and analysis of radiotherapy responses. The model provided in this application can accurately identify high- and low-risk sub-regions of tumors and perform predictive analysis, thereby improving the accuracy of personalized radiotherapy plans to improve the clinical prognosis of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplifications do not constitute limitations on the embodiments. Unless otherwise stated, the pictures in the drawings do not constitute proportional limitations.

[0022] Figure 1 A flowchart of a two-stage clustering diagnosis and automatic machine learning prediction method provided in one embodiment of the present application.

[0023] Figure 2 A core framework diagram of automatic machine learning provided in one embodiment of the present application.

[0024] Figure 3 A schematic diagram of the structure of a two-stage clustering diagnosis and automatic machine learning prediction model provided in one embodiment of the present application.

[0025] FIG. 4A to FIG. 4B A two-stage clustering diagnosis and automatic machine learning prediction model provided in one embodiment of the present application is a visualization diagram of the clustering sub-regions in the early stage of radiotherapy and the clustering sub-regions in the middle stage of radiotherapy for patient 001.

[0026] FIG. 5A to FIG. 5C A two-stage clustering diagnosis and automatic machine learning prediction model provided in one embodiment of the present application performs three traditional feature classification tasks and regression tasks on the entire tumor, high-risk tumor sub-regions, and low-risk tumor sub-regions. Kernel SHAP interpretable analysis diagram.

[0027] FIG. 6A to FIG. 6CA two-stage clustering diagnosis and automatic machine learning prediction model provided in one embodiment of the present application performs 41 radiomics feature classification tasks and regression tasks on the entire tumor, high-risk tumor sub-regions, and low-risk tumor sub-regions. Kernel SHAP interpretability analysis diagram. DETAILED DESCRIPTION

[0028] As can be seen from the background technology, tumor heterogeneity is one of the characteristics of malignant tumors, which can cause differences in tumor growth rate, invasion and metastasis, drug sensitivity, prognosis and other aspects. There is a significant correlation between the SUV value of the tumor before radiotherapy and the patient's radiotherapy effect and survival rate. The changes in the SUV value during and after radiotherapy can also provide important information about the tumor's response to radiotherapy. Traditional machine learning methods have been widely used in the prediction of tumor radiotherapy response, but they usually require a lot of manual operations and deep expertise in model selection and hyperparameter adjustment. In addition, traditional machine learning models usually lack sufficient interpretability, which makes it more difficult to understand the logic behind the model prediction. It is crucial to develop machine learning algorithm models that can provide higher accuracy and better interpretability for predicting and analyzing changes in SUV of tumors and their subregions.

[0029] Traditional machine learning methods usually require a lot of manual operations and deep professional knowledge in model selection and hyperparameter adjustment. In addition, traditional machine learning models usually lack sufficient interpretability, which makes it more difficult to understand the logic behind model predictions. In view of the shortcomings of the prior art, the embodiment of the present application provides a two-stage clustering diagnosis and automatic machine learning prediction method and model, which includes the following steps: first, based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET image are feature weighted to achieve the first stage clustering; then, based on the New_mid K-means clustering algorithm, the tumor area in the tumor PET image is clustered at the voxel level to achieve the second stage clustering, and tumor sub-regions with different metabolic characteristics are outlined; next, based on the tumor sub-regions outlined by the two-stage clustering, the entire tumor and different sub-regions are extracted; then, based on the extracted features, an automatic machine learning prediction algorithm is constructed, and the labels in the classification task and the values ​​in the regression task are predicted based on the automatic machine learning prediction algorithm to obtain the prediction results; finally, based on the Kernel SHAP method, the feature importance of the prediction results is analyzed for interpretability. The purpose of this application is to provide a two-stage clustering diagnosis and automatic machine learning prediction method and model, which can accurately identify high- and low-risk sub-regions of tumors and perform predictive analysis, thereby improving the accuracy of personalized treatment plans and improving patients' clinical prognosis.

[0030] The following will describe the various embodiments of the present application in detail with reference to the accompanying drawings. However, it will be appreciated by those skilled in the art that in the various embodiments of the present application, many technical details are provided in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solution claimed in the present application can be implemented.

[0031] See also Figure 1 The present application embodiment provides a two-stage clustering diagnosis and automatic machine learning prediction method and model, the method comprising the following steps:

[0032] Step S101: Based on the K-means clustering algorithm, feature weighting is performed on the 3D spatial coordinates and SUV in the tumor PET image to achieve the first stage clustering.

[0033] Step S102: Based on the New_mid K-means clustering algorithm, voxel-level clustering analysis is performed on the tumor region in the tumor PET image to achieve second-stage clustering, and tumor sub-regions with different metabolic characteristics are delineated.

[0034] Step S103: Based on the tumor sub-regions delineated by the two-stage clustering, feature extraction is performed on the entire tumor and different sub-regions.

[0035] Step S104: construct an automatic machine learning prediction algorithm based on the extracted features, and predict the labels in the classification task and the values ​​in the regression task based on the automatic machine learning prediction algorithm to obtain prediction results.

[0036] Step S105: Based on the Kernel SHAP method, an interpretability analysis is performed on the feature importance of the prediction results.

[0037] The two-stage tumor subregion clustering diagnosis and automatic machine learning prediction model provided in this application is based on the K-means clustering algorithm and the New_mid K-means clustering algorithm, and uses the two-stage clustering algorithm to efficiently cluster PET images, thereby distinguishing high-risk and low-risk subregions with different metabolic characteristics inside the tumor, which can provide data support for the clinical formulation of more detailed treatment strategies. In addition, the two-stage tumor subregion clustering diagnosis and automatic machine learning prediction model is based on the automatic machine learning prediction algorithm and the Kernel SHAP method. After extracting traditional features and imaging genomics features, it combines the data obtained from voxel clustering and the patient's clinical information, and uses the automatic machine learning prediction algorithm to construct an accurate radiotherapy response prediction model. It can automatically select the optimal feature combination and adopt the most appropriate machine learning algorithm for integration, thereby achieving accurate prediction of the patient's radiotherapy response.

[0038] In some embodiments, in step S101, based on the K-means clustering algorithm, feature weighting is performed on the 3D spatial coordinates and SUV in the tumor PET image to achieve first-stage clustering, including:

[0039] Step S1011: Use the stepping method to assign multiple groups of weights to the 3D spatial coordinates and SUV features in the tumor PET image to obtain multiple weight combination clustering results.

[0040] Step S1012: Based on the K-means clustering algorithm, a clustering evaluation index of each weight combination clustering result is calculated.

[0041] Step S1013: Based on the clustering evaluation index, the 3D spatial coordinates and SUV in the tumor PET image are finally weighted.

[0042] In some embodiments, in step S1011, a stepwise method is used to perform multiple weightings on the 3D spatial coordinates and SUV features in the tumor PET image to obtain multiple weight combination clustering results, including:

[0043] Step S1011a: using a stepping method to increase the weight of the SUV feature in the tumor PET image from 0.1 to 1.0 in steps of 0.1.

[0044] Step S1011b: performing multiple weightings on the 3D spatial coordinate features in the tumor PET image according to the correspondence between the 3D spatial coordinates in the tumor PET image and the SUV feature weights, and obtaining multiple weight combination clustering results.

[0045] The corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight is:

[0046] w x =w y =w z =(1-w SUV ) / 3

[0047] Among them, w x represents the x-coordinate feature weight in the tumor PET image; w y represents the y-coordinate feature weight in the tumor PET image; w z represents the z-coordinate feature weight in the tumor PET image; w SUV Represents the SUV feature weight in the tumor PET image.

[0048] In some embodiments, in step S1012, based on the K-means clustering algorithm, a clustering evaluation index of each weight combination clustering result is calculated, including: substituting each weight combination into the K-means clustering algorithm, and calculating a comprehensive score as the clustering evaluation index of the weight combination clustering result; the calculation formula of the comprehensive score is:

[0049]

[0050] CS represents the comprehensive score; SC pre SC represents the clustering silhouette coefficient of tumors in the early stage of radiotherapy; mid DS represents the tumor clustering silhouette coefficient in the middle of radiotherapy; high Dice similarity of high-risk subregions in the early and middle stages of radiotherapy; DS low Represents the Dice similarity of low-risk subregions in the early and middle stages of radiotherapy.

[0051] In some embodiments, in step S1013, the 3D spatial coordinates and SUV in the tumor PET image are finally weighted based on the clustering evaluation index, including:

[0052] Step S1013a: for a sample, based on the clustering evaluation index of each weight combination clustering result, select the SUV feature weight in the tumor PET image corresponding to the optimal clustering evaluation index.

[0053] Step S1013b: For all samples, the SUV feature weights in the tumor PET images corresponding to the optimal clustering evaluation index selected for each sample are averaged to obtain the final SUV feature weights in the tumor PET images for all samples; according to the correspondence between the 3D spatial coordinates in the tumor PET images in step S1011b and the SUV feature weights, the final 3D spatial coordinate feature weights in the tumor PET images for all samples are calculated.

[0054] In some embodiments, in step S102, based on the New_mid K-means clustering algorithm, voxel-level clustering analysis is performed on the tumor region in the tumor PET image to achieve second-stage clustering, outlining tumor subregions with different metabolic characteristics, including:

[0055] Step S1021: Based on the K-means clustering algorithm, for the clustering analysis of the tumor region in the middle stage of radiotherapy, the SUV features in the early stage of radiotherapy and the SUV features in the middle stage of radiotherapy are fused and input for voxel-level clustering analysis to construct a New_mid K-means clustering algorithm.

[0056] Step S1022: Based on the New_mid K-means clustering algorithm, the optimal weight of the mid-stage radiotherapy SUV feature is selected by stepwise debugging.

[0057] It is worth noting that the weight of the SUV feature in the middle stage of radiotherapy should be greater than or equal to the weight of the SUV feature in the early stage of radiotherapy, so as to avoid ignoring the data aggregation degree in the process of clustering analysis of tumor regions in the middle stage of radiotherapy in order to obtain better coincidence stability.

[0058] Specifically, the process of debugging and selecting the optimal weight of the mid-stage radiotherapy SUV feature by the step method is similar to the process of assigning multiple groups of weights to the 3D spatial coordinates and SUV features in the tumor PET image by the step method. However, in the process of debugging and selecting the optimal weight of the mid-stage radiotherapy SUV feature by the step method, it is required that the mid-stage radiotherapy SUV feature weight should be greater than or equal to the early stage radiotherapy SUV feature weight. At the same time, the 3D spatial coordinate feature weight of the early stage radiotherapy should be calculated based on the correspondence between the 3D spatial coordinates and the SUV feature weight in the tumor PET image in step S1011b, while the 3D spatial coordinate feature weight of the mid-stage radiotherapy should be calculated based on the correspondence between the new 3D spatial coordinates and the SUV feature weight. The correspondence between the new 3D spatial coordinates and the SUV feature weight is:

[0059] w xmid =w ymid =w zmid =[1-(w SUVpre +w SUVmid )] / 3

[0060] w SUVpre +w SUVmid =w SUV

[0061] w SUVmid ≥w SUVpre

[0062] Among them, w xmid represents the x-coordinate feature weight; w ymid represents the y-coordinate feature weight; w zmid represents the z-coordinate feature weight; w SUVpre represents the characteristic weight of SUV in the early stage of radiotherapy; w SUVmid represents the characteristic weight of SUV in the middle of radiotherapy; w SUV Represents the total weight of SUV features.

[0063] In some embodiments, in step S103, based on the tumor sub-regions delineated by clustering in two stages, feature extraction is performed on the entire tumor and different sub-regions, including:

[0064] Step S1031: extract features from the entire tumor, where the extracted features include traditional features and radiomics features.

[0065] Step S1032: extract features from high-risk tumor sub-regions, where the extracted features include traditional features and radiomics features.

[0066] Step S1033: extracting features from low-risk tumor sub-regions, where the extracted features include traditional features and radiomics features.

[0067] Specifically, feature extraction was performed for the entire tumor, high-risk tumor subregion, and low-risk tumor subregion, respectively. The extracted features included 3 traditional features and 41 radiomic features. The 3 traditional features included the volume corresponding to the entire tumor or tumor subregion, the average SUV value corresponding to the entire tumor or tumor subregion, and the radiotherapy dose corresponding to the entire tumor or tumor subregion. The 41 radiomic features included 6 intensity histogram features, 19 texture features from the grayscale tone spatial dependency matrix, 5 texture features from the neighborhood grayscale tone difference matrix, and 11 texture features from the grayscale region size matrix.

[0068] In some embodiments, in step S104, an automatic machine learning prediction algorithm is constructed based on the extracted features to predict the labels in the classification task and the values ​​in the regression task, including:

[0069] Step S1041: Based on the extracted features, build an automatic machine learning core framework; the automatic machine learning core framework includes a meta-learning module, a Bayesian optimization module and an integrated learning module.

[0070] Step S1042: Predict the labels in the classification task based on the automatic machine learning prediction algorithm.

[0071] Step S1043: predict the numerical value in the regression task based on the automatic machine learning prediction algorithm.

[0072] In step S1041, the meta-learning module of the core framework of automatic machine learning uses metadata of past data sets and model performance to guide model selection and hyperparameter initialization for new tasks; the Bayesian optimization module predicts performance under different configurations by building a probabilistic model, thereby guiding the hyperparameter search and tuning process; the integrated learning module assigns weights to each model based on the model's performance indicators, and then integrates them into a powerful integrated model, which can effectively reduce overfitting and improve the model's generalization ability on unknown data.

[0073] Specifically, the value to be predicted in the regression task is the mean change rate of the SUV of the entire tumor or tumor subregion. At the same time, samples that meet the corresponding relationship are classified as "responsive" and are labeled "1"; samples that do not meet the corresponding relationship are classified as "unresponsive" and are labeled "0". The label to be predicted in the classification task is "1" or "0"; the corresponding relationship is:

[0074]

[0075] Among them, ΔSUVmean represents the rate of change of the mean SUV of the entire tumor or tumor subregion; SUVpre represents the average SUV value of the entire tumor or tumor subregion in the early stage of radiotherapy; SUVmid represents the average SUV value of the entire tumor or tumor subregion in the middle stage of radiotherapy.

[0076] See also Figure 2 , the embodiment of the present application provides a core framework diagram of automatic machine learning, including a meta-learning module, a Bayesian optimization module and an integrated learning module. In step S1041, the core framework of automatic machine learning is constructed, including:

[0077] Step S1041a: Build a meta-learning module and use metadata of past data sets and model performance to guide model selection and hyperparameter initialization for new tasks.

[0078] Step S1041b: Build a Bayesian optimization module to predict the performance under different configurations by constructing a probability model, thereby guiding the search and tuning process of hyperparameters.

[0079] Step S1041c, build an integrated learning module, assign weights to each model based on the model's performance indicators, and then integrate them into a powerful integrated model, which can effectively reduce overfitting and improve the model's generalization ability on unknown data.

[0080] In some embodiments, in step S105, based on the Kernel SHAP method, an interpretability analysis is performed on the feature importance of the prediction result, including:

[0081] Step S1051: Based on the SHAP method, kernel regression and weight sampling techniques are integrated to construct a Kernel SHAP method.

[0082] Step S1052: Based on the Kernel SHAP method, the Shapley value estimation problem is converted into a weighted least squares regression problem, and the contribution value of each feature is estimated using an efficient linear model solving technique.

[0083] Step S1053: Based on the contribution value of each feature in step S1052, an interpretability analysis is performed on the feature importance of the prediction result in step S104.

[0084] See also Figure 3 The embodiment of the present application also provides a two-stage clustering diagnosis and automatic machine learning prediction model, including a two-stage clustering module 101 and a prediction analysis module 102 connected in sequence; wherein the two-stage clustering module includes a feature weighting module 1011 and a New_mid K-means clustering module 1012; the feature weighting module 1011 is used to perform feature weighting on the 3D spatial coordinates and SUV value features in the tumor PET image according to the K-means clustering algorithm to achieve the first stage clustering; the New_mid K-means clustering module 1012 is used to perform feature weighting on the 3D spatial coordinates and SUV value features in the tumor PET image according to the New_mid The K-means clustering algorithm performs voxel-level clustering analysis on the tumor area in the tumor PET image to achieve the second stage clustering, and outlines the tumor sub-regions with different metabolic characteristics; the prediction and analysis module 102 includes a feature extraction module 1021, an automatic machine learning module 1022 and a KernelSHAP interpretability analysis module 1023; the feature extraction module 1021 is used to extract features of the entire tumor and different sub-regions based on the tumor sub-regions outlined by the two-stage clustering; the automatic machine learning module 1022 is used to construct an automatic machine learning prediction algorithm based on the extracted features, and predict the labels in the classification task and the values ​​in the regression task based on the automatic machine learning prediction algorithm to obtain the prediction results; the Kernel SHAP interpretability analysis module 1023 is used to perform interpretability analysis on the feature importance of the prediction results according to the Kernel SHAP method.

[0085] The two-stage clustering diagnosis and automatic machine learning prediction model provided by the present application is described in detail below with respect to a specific embodiment.

[0086] First of all, the two-stage clustering diagnosis and automatic machine learning prediction model proposed in the embodiment of the present application is based on the Windows 11 system and implemented using Python 3.9.

[0087] In addition, it should be noted that the embodiments of the present application analyze and evaluate the radiotherapy response of locally advanced non-small cell lung cancer.

[0088] First, the feature weighting module uses the stepping method to increase the SUV feature weight in the tumor PET image from 0.1 to 1.0 in steps of 0.1, and multiple groups of weights are assigned to the 3D spatial coordinate features according to the corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight. Then, each weight combination is substituted into the K-means clustering algorithm to obtain the SUV feature weight corresponding to the optimal comprehensive score of each patient, and then the average is calculated to obtain the final SUV feature optimal weight for all patients. At the same time, according to the corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight, the final 3D spatial coordinate feature weight for all patients is calculated. The final optimal SUV feature weight is 0.8, and the corresponding x-coordinate feature weight, y-coordinate feature weight, and z-coordinate feature weight are all 1 / 15.

[0089] Then, a similar step-by-step method is used to debug and select the optimal weight of the mid-radiotherapy SUV feature through the New_mid K-means clustering module, while ensuring that the mid-radiotherapy SUV feature weight should be greater than or equal to the early-radiotherapy SUV feature weight. Finally, when clustering the mid-radiotherapy tumor PET images, the optimal weight of the mid-radiotherapy SUV feature is 0.4, and the optimal weight of the early-radiotherapy SUV feature is 0.4. The corresponding x-coordinate feature weight, y-coordinate feature weight, and z-coordinate feature weight are all 1 / 15. According to the feature weighting results, the New_mid K-means clustering algorithm is used to cluster the early and mid-radiotherapy tumor PET images of each patient to obtain cluster 2 results. The cluster 2 results of the early and mid-radiotherapy tumor PET images of patient 001 can be found in FIG. 4A to FIG. 4B . Figure 4A A visualization diagram of the clustering sub-regions of the pre-radiotherapy period of patient 001 using the two-stage clustering diagnosis and automatic machine learning prediction model provided by the present application is shown; Figure 4B A visualization diagram of the clustering sub-regions of the mid-term radiotherapy of patient 001 using the two-stage clustering diagnosis and automatic machine learning prediction model provided in the present application is shown, wherein the light-colored parts are high-risk sub-regions and the dark-colored parts are low-risk sub-regions.

[0090] The comparative analysis of silhouette coefficient, CH-index, Dice similarity and Euclidean distance between New_mid K-means algorithm and K-means algorithm, spectral clustering algorithm and GMM algorithm when clustering 2 classes is shown in Table 1 and Table 2. It can be seen that New_mid Kmeans clustering has a significant improvement in the stability of sub-region delineation compared with K-means, and the clustering effect is better than that of spectral clustering algorithm and GMM algorithm.

[0091] Table 1. Comparison of silhouette coefficient and CH-index of New_mid K-means algorithm, K-means algorithm, spectral clustering algorithm and GMM algorithm when clustering 2 classes obtained by New_mid K-means clustering module in an embodiment

[0092]

[0093] Table 2 New_mid K-means clustering module in the first embodiment obtains Dice similarity and Euclidean distance comparison between New_mid K-means algorithm and K-means algorithm, spectral clustering algorithm, and GMM algorithm when clustering 2 classes

[0094]

[0095] Next, the feature extraction module extracts features for the whole tumor, high-risk tumor subregions, and low-risk tumor subregions based on the tumor subregions delineated by the two-stage clustering module. The extracted features include 3 traditional features and 41 radiomic features. The 3 traditional features include the volume corresponding to the whole tumor or tumor subregion, the average SUV value corresponding to the whole tumor or tumor subregion, and the radiotherapy dose corresponding to the whole tumor or tumor subregion. The 41 radiomic features include 6 intensity histogram features, 19 texture features from the grayscale tone spatial dependency matrix, 5 texture features from the neighborhood grayscale tone difference matrix, and 11 texture features from the grayscale region size matrix.

[0096] Then, through the automatic machine learning module, the LOOCV (Leave-One-Out Cross-Validation) method was used to predict and compare the classification tasks using the automatic machine learning (AutoML), logistic regression (LR), support vector machine (SVM), random forest (RF), and gradient boosting decision tree (GBDT) methods. The AUC values ​​obtained by the 3 traditional feature classification and 41 radiomics feature classification are shown in Tables 3 and 4. In the regression task, the automatic machine learning, logistic regression, SVM, RF, and GBDT methods were used for prediction and comparison analysis. The RMSE values ​​obtained by the 3 traditional feature regression and 41 radiomics feature regression are shown in Tables 5 and 6. It can be seen that the AUC values ​​obtained by the automatic machine learning algorithm in the classification task are generally higher than those of other classical machine learning algorithms, indicating that its classification predictions in the entire tumor, high-risk tumor subregions, and low-risk tumor subregions are more accurate. At the same time, the RMSE values ​​obtained by the automatic machine learning algorithm in the regression task are generally lower than those of other classic machine learning algorithms, indicating that its regression predictions on the entire tumor, high-risk tumor subregions, and low-risk tumor subregions are more accurate. It can be seen that the automatic machine learning algorithm has excellent prediction performance.

[0097] Table 3 AUC values ​​obtained by automatic machine learning algorithm and classic machine learning algorithm 3 traditional feature classification in the automatic machine learning module of an embodiment

[0098]

[0099] Table 4 AUC values ​​obtained by automatic machine learning algorithm and classical machine learning algorithm in the automatic machine learning module of an embodiment 41 Radiomics feature classification

[0100]

[0101]

[0102] Table 5 RMSE values ​​obtained by automatic machine learning algorithm and traditional feature regression of classic machine learning algorithm 3 in the automatic machine learning module of an embodiment

[0103]

[0104] Table 6 RMSE values ​​obtained by regression of 41 radiomics features using the automatic machine learning algorithm and the classical machine learning algorithm in the automatic machine learning module in an embodiment

[0105]

[0106] Finally, the Kernel SHAP interpretability analysis module is used to analyze the feature importance of the prediction results based on the contribution value of each feature. Kernel SHAP interpretability analysis is performed on the prediction results of the three traditional feature classification tasks and regression tasks for the entire tumor, high-risk tumor subregions, and low-risk tumor subregions. FIG. 5A to FIG. 5C .in, Figure 5A The Kernel SHAP interpretability analysis diagram of the two-stage clustering diagnosis and automatic machine learning prediction model provided by the present application for the entire tumor for 3 traditional feature classification tasks (top) and regression tasks (bottom) is shown; Figure 5B The KernelSHAP interpretability analysis diagram of the model performing 3 traditional feature classification tasks (top) and regression tasks (bottom) on high-risk tumor subregions is shown; Figure 5C The Kernel SHAP interpretability analysis diagram of the model performing 3 traditional feature classification tasks (top) and regression tasks (bottom) on low-risk tumor sub-regions.

[0107] Kernel SHAP interpretability analysis of the prediction results of 41 radiomics feature classification tasks and regression tasks for the entire tumor, high-risk tumor subregions, and low-risk tumor subregions, see FIG. 6A to FIG. 6C .in, Fig. 6A The Kernel SHAP interpretability analysis diagram of the two-stage clustering diagnosis and automatic machine learning prediction model provided in the present application for the classification task (left) and regression task (right) of 41 radiomics features of the entire tumor is shown; Figure 6B The Kernel SHAP interpretability analysis diagram of the model performing 41 radiomics feature classification tasks (left) and regression tasks (right) on high-risk tumor subregions is shown; Figure 6C Kernel SHAP interpretability analysis diagram showing the model’s classification task (left) and regression task (right) of 41 radiomic features for low-risk tumor subregions.

[0108] Based on the above technical scheme, the embodiment of the present application provides a two-stage clustering diagnosis and automatic machine learning prediction method and model, which includes the following steps: first, based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET image are feature weighted to achieve the first stage clustering; then, based on the New_mid K-means clustering algorithm, the tumor area in the tumor PET image is clustered at the voxel level to achieve the second stage clustering, and tumor sub-regions with different metabolic characteristics are outlined; next, based on the tumor sub-regions outlined by the two-stage clustering, features of the entire tumor and different sub-regions are extracted; then, based on the extracted features, an automatic machine learning prediction algorithm is constructed, and based on the automatic machine learning prediction algorithm, the labels in the classification task and the values ​​in the regression task are predicted to obtain prediction results; finally, based on the Kernel SHAP method, the feature importance of the prediction results is analyzed for interpretability.

[0109] This application proposes a two-stage clustering diagnosis and automatic machine learning prediction method and model. First, this application uses a two-stage clustering algorithm to efficiently cluster PET images, thereby distinguishing high- and low-risk sub-regions with different metabolic characteristics inside the tumor, which can provide data support for the clinical formulation of more detailed treatment strategies. Secondly, this application uses traditional features and imaging genomics features to extract and combine the data obtained from voxel clustering and the patient's clinical information, and uses an automatic machine learning prediction algorithm to construct an accurate radiotherapy response prediction model, which can automatically select the optimal feature combination and adopt the most appropriate machine learning algorithm for integration, thereby achieving accurate prediction of the patient's radiotherapy response. Finally, this application verifies the excellent performance of the model through the application results of patients with locally advanced non-small cell lung cancer, showing its significant advantages in accurate segmentation of tumor sub-regions and prediction and analysis of radiotherapy responses. The model provided in this application can accurately identify high- and low-risk sub-regions of tumors and perform predictive analysis, thereby improving the accuracy of personalized treatment plans to improve the clinical prognosis of patients.

[0110] Those skilled in the art can understand that the above-mentioned embodiments are specific examples for implementing the present application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present application. Any person skilled in the art can make their own changes and modifications without departing from the spirit and scope of the present application, so the scope of protection of the present application shall be based on the scope defined in the claims.

Claims

1. A two-stage clustering diagnosis and automatic machine learning prediction method, characterized in that: The following steps are involved: Based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET images were weighted to achieve the first stage clustering; Based on the New_mid K-means clustering algorithm, voxel-level clustering analysis was performed on the tumor regions in the tumor PET images to achieve the second stage of clustering, outlining tumor subregions with different metabolic characteristics; Based on the tumor subregions delineated by two-stage clustering, feature extraction is performed on the entire tumor and different subregions; Building an automatic machine learning prediction algorithm based on the extracted features, and predicting the labels in the classification task and the values ​​in the regression task based on the automatic machine learning prediction algorithm to obtain prediction results; Based on the Kernel SHAP method, an interpretability analysis of the feature importance of the prediction results is performed.

2. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 1, characterized in that: Based on the K-means clustering algorithm, the 3D spatial coordinates and SUV in the tumor PET images are weighted to achieve the first stage of clustering, including: The step-by-step method was used to assign multiple weights to the 3D spatial coordinates and SUV features in the tumor PET images, and multiple weight combination clustering results were obtained; Based on the K-means clustering algorithm, the clustering evaluation index of each weight combination clustering result is calculated; Based on the clustering evaluation index, the 3D spatial coordinates and SUV in the tumor PET image are finally weighted.

3. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 2, characterized in that: The step-by-step method is used to assign multiple weights to the 3D spatial coordinates and SUV features in the tumor PET images, and multiple weight combination clustering results are obtained, including: The weight of the SUV feature in the tumor PET image was adjusted from 0.1 to 1.0 in steps of 0.1 using the step method; According to the correspondence between the 3D spatial coordinates in the tumor PET image and the SUV feature weights, multiple groups of weighted 3D spatial coordinate features in the tumor PET image are assigned to obtain multiple weight combination clustering results; The corresponding relationship between the 3D spatial coordinates in the tumor PET image and the SUV feature weight is: In x =in y =in z =(1-in SUV ) / 3 Among them, w x represents the x-coordinate feature weight in the tumor PET image; w y represents the y-coordinate feature weight in the tumor PET image; w z represents the z-coordinate feature weight in the tumor PET image; w SUV Represents the SUV feature weight in the tumor PET image.

4. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 2, characterized in that: Based on the K-means clustering algorithm, the clustering evaluation index of each weight combination clustering result is calculated, including: Substituting each weight combination into the K-means clustering algorithm, calculating the comprehensive score as the clustering evaluation index of the weight combination clustering result; The calculation formula of the comprehensive score is: CS represents the comprehensive score; SC pre SC represents the clustering silhouette coefficient of tumors in the early stage of radiotherapy; mid DS represents the tumor clustering silhouette coefficient in the middle of radiotherapy; high Dice similarity of high-risk subregions in the early and middle stages of radiotherapy; DS low Represents the Dice similarity of low-risk subregions in the early and middle stages of radiotherapy.

5. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 2, characterized in that: Based on the clustering evaluation index, the 3D spatial coordinates and SUV in the tumor PET image are finally weighted, including: For a sample, based on the clustering evaluation index of each weight combination clustering result, the SUV feature weight in the tumor PET image corresponding to the optimal clustering evaluation index is selected; For all samples, the SUV feature weights in the tumor PET images corresponding to the optimal clustering evaluation index selected for each sample are averaged to obtain the final SUV feature weights in the tumor PET images for all samples; according to the correspondence between the 3D spatial coordinates in the tumor PET images and the SUV feature weights, the final 3D spatial coordinate feature weights in the tumor PET images for all samples are calculated.

6. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 1, characterized in that: Based on the New_mid K-means clustering algorithm, voxel-level clustering analysis was performed on the tumor area in the tumor PET image to achieve the second stage of clustering, outlining tumor sub-regions with different metabolic characteristics, including: Based on the K-means clustering algorithm, for the clustering analysis of tumor regions during the mid-stage of radiotherapy, the SUV features of the early stage of radiotherapy and the SUV features of the mid-stage of radiotherapy were fused and input for voxel-level clustering analysis, and the New_mid K-means clustering algorithm was constructed. Based on the New_mid K-means clustering algorithm, the optimal weight of the mid-stage radiotherapy SUV feature was selected by stepwise debugging. The weight of the mid-stage radiotherapy SUV feature was greater than or equal to the weight of the early-stage radiotherapy SUV feature, so as to avoid ignoring the data aggregation degree in the process of clustering analysis of tumor regions in the mid-stage radiotherapy in order to obtain better coincidence stability.

7. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 1, characterized in that: Based on the tumor subregions delineated by two-stage clustering, feature extraction is performed on the entire tumor and different subregions, including: Features extracted from the whole tumor and different subregions include traditional features and radiomic features; The traditional features include the volume of the whole tumor and different subregions, the average SUV value of the whole tumor and different subregions, and the radiotherapy dose of the whole tumor and different subregions; The radiomics features include 6 intensity histogram features, 19 texture features from the grayscale tone spatial dependency matrix, 5 texture features from the neighborhood grayscale tone difference matrix, and 11 texture features from the grayscale region size matrix, for a total of 41 radiomics features.

8. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 1, characterized in that: Build automatic machine learning prediction algorithms based on the extracted features, including: Build the core framework of automatic machine learning based on the extracted features; The automatic machine learning core framework includes a meta-learning module, a Bayesian optimization module, and an integrated learning module; wherein the meta-learning module uses metadata of past data sets and model performance to guide model selection and hyperparameter initialization for new tasks; The Bayesian optimization module predicts the performance under different configurations by building a probability model, thereby guiding the search and tuning process of hyperparameters; The ensemble learning module assigns weights to each model based on the model's performance indicators, and then integrates them into a powerful ensemble model, which can effectively reduce overfitting and improve the model's generalization ability on unknown data.

9. The two-stage clustering diagnosis and automatic machine learning prediction method according to claim 1, characterized in that: Based on the Kernel SHAP method, an interpretability analysis of the feature importance of the prediction results is performed, including: Based on the SHAP method, the kernel regression and weight sampling techniques are integrated to construct the Kernel SHAP method; Based on the Kernel SHAP method, the estimation problem of Shapley value is transformed into a weighted least squares regression problem, and the contribution value of each feature is estimated using efficient linear model solving technology; Based on the contribution value of each feature, an interpretability analysis is performed on the feature importance of the prediction result.

10. A two-stage clustering diagnosis and automatic machine learning prediction model, characterized in that: It includes a two-stage clustering module and a prediction analysis module connected in sequence; The two-stage clustering module includes a feature weighting module and a New_mid K-means clustering module; The feature weighting module is used to perform feature weighting on the 3D spatial coordinates and SUV value features in the tumor PET image according to the K-means clustering algorithm to achieve the first stage clustering; The New_mid K-means clustering module is used to perform voxel-level clustering analysis on the tumor region in the tumor PET image according to the New_mid K-means clustering algorithm to achieve the second stage clustering and delineate tumor subregions with different metabolic characteristics; The prediction analysis module includes a feature extraction module, an automatic machine learning module and a Kernel SHAP interpretability analysis module; The feature extraction module is used to extract features of the entire tumor and different sub-regions based on the tumor sub-regions outlined by clustering in the two stages; The automatic machine learning module is used to construct an automatic machine learning prediction algorithm based on the extracted features, and predict the labels in the classification task and the values ​​in the regression task based on the automatic machine learning prediction algorithm to obtain prediction results; The Kernel SHAP interpretability analysis module is used to perform interpretability analysis on the feature importance of the prediction results according to the Kernel SHAP method.

Images