Therapeutic effect evaluation method and system for stereotactic radiotherapy of spinal tumor

Abstract

The invention discloses a curative effect evaluation method and system for stereotactic radiotherapy of spinal tumor, and the method comprises the steps: carrying out the automatic segmentation of tumor tissues and surrounding healthy organs in a spinal tumor imaging image after the stereotactic radiotherapy, and enabling a segmented tumor region to serve as a target organ region; extracting target image omics features from the image of the target organ region, wherein the target image omics features comprise shape features, texture features, gray level statistic features and high-order features related to the curative effect evaluation result; and inputting the target image omics characteristics into the first curative effect evaluation model to obtain a curative effect evaluation prediction result. The multi-dimensional features such as the shape feature, the texture feature, the gray statistic feature and the high-order feature are extracted from the image of the target organ region, the curative effect is evaluated based on the multi-dimensional features, and the accuracy of curative effect evaluation is improved.

Description

Technical Field

[0001] The present application relates to the technical field of stereotactic radiotherapy, and more specifically, to a method and system for evaluating the efficacy of stereotactic radiotherapy for spinal tumors. Background Art

[0002] Spinal tumors account for approximately 6% to 10% of all bone tumors in the body. Almost all types of bone tumors can be seen in the spine, such as osteosarcoma, osteoid osteoma, and aneurysmal bone cyst, and metastatic bone tumors account for more than half of spinal tumors.

[0003] At present, the treatment of spinal tumors is mainly preoperative stereotactic radiotherapy or postoperative combined stereotactic radiotherapy. The main feature of stereotactic radiotherapy for spinal tumors is the single large-dose fractionation mode. The single dose is high. Spinal tumors are close to the spinal cord, esophagus / intestinal tract, heart, lungs, stomach and other endangered organs. While ensuring the efficacy of treatment, its toxic side effects must be strictly controlled. Therefore, the efficacy of stereotactic radiotherapy for spinal tumors has become a concern for patients.

[0004] In the prior art, the efficacy of spinal tumor radiotherapy is evaluated based on the size measurement results of the spinal tumor and the organs around the spinal tumor. Evaluating the efficacy only from the size dimension is somewhat one-sided. Therefore, the accuracy of the existing efficacy evaluation method of stereotactic radiotherapy for spinal tumors needs to be improved. Summary of the invention

[0005] The present application provides a method and system for evaluating the efficacy of stereotactic radiotherapy for spinal tumors, which extracts multi-dimensional features such as shape features, texture features, grayscale statistics features and high-order features from images of the target organ area, evaluates the efficacy based on the multi-dimensional features, and improves the accuracy of the efficacy evaluation.

[0006] The present application provides a method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors, comprising:

[0007] Automatically segment the tumor tissue and surrounding healthy organs in spinal tumor imaging images after stereotactic radiotherapy, and use the segmented tumor area as the target organ area;

[0008] Extracting target imaging features from the images of the target organ region, the target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the efficacy evaluation results;

[0009] The target imaging omics features are input into the first efficacy evaluation model to obtain the efficacy evaluation prediction results.

[0010] Preferably, when obtaining the efficacy evaluation prediction result, the image of the target organ region is input into a plurality of efficacy evaluation models constructed by different algorithms, and the efficacy evaluation model with the best model performance is used as the first efficacy evaluation model.

[0011] Preferably, the efficacy evaluation results include local control, toxicity, spinal cord injury and survival time.

[0012] Preferably, a convolutional neural network is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy.

[0013] Preferably, extracting target radiomics features from the image of the target organ region specifically includes:

[0014] The first radiomics features of the region of interest that meet the IBSI criteria are automatically extracted from the images of the target organ region;

[0015] All first radiomics features were z-score normalized;

[0016] Using the univariate selection method, insignificant features were screened out as the second radiomics features;

[0017] Through minimum absolute shrinkage and selection operator regression, indicators related to efficacy evaluation results were screened out from the second radiomics features as target radiomics features.

[0018] The present application also provides a spinal tumor stereotactic radiotherapy efficacy evaluation system, including a segmentation module, a feature extraction module and an evaluation module;

[0019] The segmentation module is used to automatically segment the tumor tissue and surrounding healthy organs in the spinal tumor imaging images after stereotactic radiotherapy, and use the segmented tumor area as the target organ area;

[0020] The feature extraction module is used to extract target imaging features from the images of the target organ area. The target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the efficacy evaluation results;

[0021] The evaluation module is used to input the target imaging genomics features into the first efficacy evaluation model to obtain efficacy evaluation prediction results.

[0022] Preferably, the evaluation module is used to input the image of the target organ region into multiple efficacy evaluation models constructed by different algorithms, and use the efficacy evaluation model with the best model performance as the first efficacy evaluation model.

[0023] Preferably, the efficacy evaluation results include local control, toxicity, spinal cord injury and survival time.

[0024] Preferably, the segmentation module is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy using a convolutional neural network.

[0025] Preferably, the feature extraction module includes a standard extraction module, a normalization module, a first screening module and a second screening module;

[0026] The standard extraction module is used to automatically extract the first radiomics features of the region of interest that meet the IBSI standard from the image of the target organ region;

[0027] The normalization module is used to perform z-score normalization on all first radiomics features;

[0028] The first screening module is used to screen out insignificant features using a univariate selection method as the second radiomics feature;

[0029] The second screening module is used to screen out indicators related to the efficacy evaluation results from the second radiomics features as target radiomics features through minimum absolute shrinkage and selection operator regression.

[0030] Other features and advantages of the present application will become apparent from the following detailed description of exemplary embodiments of the present application with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the application and, together with the description, serve to explain the principles of the application.

[0032] Figure 1 A flow chart of a method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors provided in this application;

[0033] Figure 2 This is a structural diagram of the efficacy evaluation system for stereotactic radiotherapy of spinal tumors provided in this application. DETAILED DESCRIPTION

[0034] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application.

[0035] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0036] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0037] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0038] The present application provides a method and system for evaluating the efficacy of stereotactic radiotherapy for spinal tumors. Multidimensional features such as shape features, texture features, grayscale statistics features, and high-order features are extracted from the image of the target organ area. The efficacy is evaluated based on the multidimensional features, which improves the accuracy of the efficacy evaluation. In addition, the present application can assist clinicians in making high-level diagnoses, allowing grassroots doctors and young doctors to obtain diagnostic capabilities that are homogeneous with senior experts, thereby effectively improving the level of imaging evaluation and prediction of spinal tumors. In addition, the present application also uses clinical characteristics as part of the input data of the efficacy evaluation model, and its combination with imaging genomics features further improves the predictive performance of the efficacy evaluation model.

[0039] like Figure 1 As shown, the method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors provided in this application includes:

[0040] S110: Automatically segment the tumor tissue and surrounding healthy organs in spinal tumor imaging images (such as CT images) after stereotactic radiotherapy, and use the segmented tumor area as the target organ area.

[0041] As an embodiment, a convolutional neural network is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy. The convolutional neural network is used to automatically identify spinal tumors and normal organs from spinal tumor imaging images to improve the recognition ability of the model.

[0042] As an embodiment, the VBNet network model is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy. VBNet is a segmentation model based on the VNet network, which combines the advanced models of VNet and Bottleneck. It replaces the traditional convolutional layers inside the downsampling block and the upsampling block with the Bottleneck structure. The Bottleneck structure consists of three convolutional layers. The first convolutional layer applies a 1×1×1 convolution kernel to reduce the number of channels of the feature map. The second convolutional layer uses a convolution kernel of the same size as the traditional convolutional layer for spatial convolution. The last convolutional layer applies a 1×1×1 convolution kernel to increase the number of channels of the feature map back to the original size. The Bottleneck structure reduces the number of parameters of the convolution kernel while ensuring the segmentation effect of the model, making the model loading and reasoning faster, making it easy to deploy the segmentation network to the cloud or mobile applications. While ensuring the segmentation effect of the model, the number of parameters of the convolution kernel is reduced, making the model loading and reasoning faster.

[0043] It should be noted that during the training phase of the automatic segmentation model, the spinal tumor imaging images need to be preprocessed before being input into the model in order to improve the performance of the model, ensure a more efficient training process, and ultimately obtain more accurate segmentation results.

[0044] As an embodiment, preprocessing of spinal tumor imaging images specifically includes:

[0045] (1) Standardization: Z-score standardization is performed to reduce the impact of the numerical range on model training.

[0046] (2) Contrast adjustment: Enhance the contrast of the image through histogram equalization, gamma correction and other techniques to make the target area more prominent.

[0047] (3) Resampling: Adjust the spatial resolution of the image to ensure that all images have the same pixel spacing. As an example, the median voxel spacing in the dataset, i.e., 1.0×1.0×1.0 mm, is selected as the standard spacing for resampling.

[0048] (4) Cropping and padding: Cropping the image to a fixed size or adding extra pixels (padding) to the edge of the image to meet the model input size requirements. As an example, the block size is unified to 64×64×64 voxels, which can ensure sufficient local feature capture while meeting the model's requirements for global context information.

[0049] (5) Data augmentation: Generate more training samples by rotating, flipping, scaling, and other operations on the original images. This data augmentation strategy can further enhance the robustness of the doctor model.

[0050] Preferably, in the present application, when training the automatic segmentation model, two convolutional neural networks are used for training, and the two VB-Nets are trained at different image resolutions. At low resolution, we train a VB-Net to roughly locate the volume of interest (VOI) of each organ. The VB-Net is trained using resampled images with a voxel size of 6 mm. At high resolution, we train the VB-Net to accurately depict the boundaries of organs within the detected VOI. That is, first use a convolutional neural network with low image resolution to roughly locate the volume of interest (VOI) of each organ, and then use a convolutional neural network with high image resolution to accurately depict the boundaries of organs within the detected VOI. After the training is completed, the convolutional neural network with high image resolution is used for automatic segmentation during the inference process.

[0051] In the training process of the automatic segmentation model, the convolutional neural network is trained using the training set. The back propagation algorithm and the stochastic gradient descent algorithm are used to calculate the loss between the prediction result and the annotation result of the forward propagation, and the weight of each layer in the network is updated iteratively by back propagation until the prediction result of the model is very similar to the annotation result, that is, the loss function is very low and tends to converge. Then the training of the model is stopped and the model is saved. Then the model is verified and the parameters are tuned on the test set. The best model parameters and hyperparameter configuration are selected through the Dice value.

[0052] In this application, the loss function of the convolutional neural network for automatic segmentation is It is the Dice loss, and its formula is as follows:

[0053]

[0054] and Respectively represent The predicted and true values ​​of pixels.

[0055] After obtaining the final automatic segmentation model, the model is used to infer the test set and visualize the target organ area.

[0056] This application uses the Dice coefficient to evaluate the overlap between the tumor segmentation results inferred by the model and the gold standard annotated by senior doctors to evaluate the accuracy of the model segmentation results. The Dice coefficient value is between 0 and 1. The closer it is to 1, the higher the overlap between the model segmentation results and the gold standard, and the better the segmentation performance:

[0057]

[0058] Among them, A represents the binary segmentation result predicted by the model, B represents the binary segmentation result of the gold standard, and |·| represents the number of pixels.

[0059] S120: extracting target imaging features from the image of the target organ region, wherein the target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the efficacy evaluation results.

[0060] As an example, the efficacy evaluation results include local control, toxicity, spinal cord injury, and survival time.

[0061] It should be noted that before extracting the target imaging features, the images of the target organ area need to be preprocessed to improve the quality of the data, reduce noise and inconsistency, and thus improve the accuracy and generalization ability of the model. Preprocessing includes the following steps:

[0062] (1) Image standardization: Make data from different devices or sources consistent and reduce deviations caused by device differences. As an example, image standardization is performed through Z-score standardization.

[0063] (2) Image resampling: Adjust the spatial resolution of the image to ensure that all images have the same pixel spacing. As an example, the median voxel spacing in the dataset, i.e., 1.0×1.0×1.0 mm, is selected as the standard spacing for resampling.

[0064] As an embodiment, the target imaging omics features are features related to the therapeutic effect evaluation results selected from the shape features, texture features, grayscale statistics features and high-order features obtained by automatic extraction.

[0065] As an embodiment, 2264 ROI (Region Of Interest) radiomics features that meet the IBSI standard are automatically extracted (as the first radiomics feature), shape features are extracted from the original image based on the ROI, and texture features, grayscale statistics features and high-order features are extracted from the original image and the filtered image.

[0066] Specifically, shape features include 14 three-dimensional features that reflect the shape and size of the region, such as maximum diameter, volume, surface area, and similarity to a ball. Grayscale statistics features are obtained from 18 features extracted from the original image. These features quantitatively describe the distribution of voxel intensities in the image through common and basic metrics, including average, peak, maximum, and minimum values. Texture features are features that reflect the grayscale distribution of pixels and their surrounding spatial neighborhoods obtained based on 72 features extracted from the original image. They are obtained by calculating features based on the Gray Level Cooccurence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM), Gray Level Size Zone Matrix (GLSZM), Neighbouring Gray Tone Difference Matrix (NGTDM), and Gray Level Dependence Matrix (GLDM). The high-order features are obtained based on 432 grayscale statistical features and 1728 texture features extracted from the image after filtering. The filtering methods include 24 types such as mean filtering, Gaussian filtering, logarithmic filtering, and wavelet transform.

[0067] As can be seen from the above, there are a large number of ROI radiomics features that meet the IBSI standard. In order to select features with good repeatability, full information retention and no redundancy, it is necessary to reduce the dimension and screen the high-throughput radiomics features. First, the data set was randomly divided into a training set and a test set in a ratio of 8:2, and all the first radiomics features were z-score normalized to eliminate the influence of inconsistent feature value dimensions. Subsequently, the univariate selection method was used to screen out insignificant features (p<0.05) as the second radiomics features; the least absolute shrinkage and selection operator (LASSO) regression was used to screen out indicators related to the efficacy evaluation results (local control, toxicity, spinal cord injury and survival time, etc.) from the second radiomics features as the target radiomics features.

[0068] S130: Input the target imaging genomics features into the first efficacy evaluation model to obtain efficacy evaluation prediction results.

[0069] As an embodiment, when obtaining the prediction results of efficacy evaluation, the images of the target organ area are input into multiple efficacy evaluation models (for example, efficacy evaluation models constructed using random forests, decision trees, support vector machines, logistic regression and XGBOOST methods, etc.), and after these efficacy evaluation models are trained, each model is applied to the test data set. The comprehensive performance of the model is quantified by the receiver operating characteristic (ROC) curve. The ROC curve and its area under the curve (AUC) range from [0.5, 1]. The closer the AUC is to 1, the better the classification model performance. In addition, the performance of the classification model is evaluated in detail by indicators such as sensitivity, specificity, accuracy, precision, and F1score. Finally, the model with the best performance is selected as the first efficacy evaluation model.

[0070] Since some clinical characteristics (such as age, gender, prescription dose, fractionation, dose line, site, etc.) will also affect the efficacy of stereotactic radiotherapy, based on such considerations, preferably, the above-mentioned imaging omics features and clinical characteristics corresponding to the imaging image source of the target influencing omics features are combined to construct a third efficacy evaluation model.

[0071] Specifically, for the analysis of radiomics features, the chi-square test was used to select the top k significant features. Then, the least absolute shrinkage and selection operator (LASSO) was used to select the optimal features, which were used to establish a machine learning efficacy prediction model, and the radiomics score (Rad score) of spinal tumors was calculated by the model.

[0072] For the analysis of clinical characteristics, univariate and multivariate logistic regression were used to screen the features with the greatest correlation with the efficacy evaluation results. Specifically, the univariate analysis method was first used to compare the differences in clinical factors between the two groups of patients to obtain factors with statistical differences (P<0.05), and then multivariate logistic regression analysis was used to screen out independent risk factors related to the efficacy evaluation results from the factors with statistical differences as clinical features input into the third efficacy evaluation model.

[0073] Using the comprehensive screening of clinical characteristics and radiomics scores, a machine learning efficacy prediction model was constructed, and then a multi-parameter joint model (i.e., the third efficacy evaluation model) was obtained. The contribution of radiomics scores and clinical characteristics in the model can be visualized through a nomogram.

[0074] On the basis of the above, when obtaining the prediction results of efficacy evaluation, it also includes constructing a second efficacy evaluation model using separate clinical characteristics. The first efficacy evaluation model and the second efficacy evaluation model are used as control models of the third efficacy evaluation model, and the AUC, sensitivity, specificity and other indicators of the first efficacy evaluation model and the second efficacy evaluation model are calculated to verify the effectiveness of the comprehensive model.

[0075] Preferably, in order to improve the stability and generalization ability of the model, the second efficacy evaluation model is trained by an integrated learning method. Specifically, when training the second efficacy evaluation model, the model parameters of the strong learner are optimized by using the model parameters obtained during the training process of multiple weak learners (for example, learning models constructed based on random forests and gradient boosting trees), so as to train a strong learner as the final second efficacy evaluation model.

[0076] On this basis, in order to further improve the performance of the model, preferably, an integrated learning method is used to combine the prediction results of the first efficacy evaluation model, the second efficacy evaluation model and the third efficacy evaluation model to obtain the final efficacy evaluation prediction result.

[0077] As an embodiment, a voting method is used to calculate the efficacy evaluation prediction result based on the prediction results of the three efficacy evaluation models. Specifically, the prediction results of the three different efficacy evaluation models are voted or weighted averaged to obtain the final efficacy evaluation prediction result.

[0078] Ensemble learning methods can leverage the advantages of different models to improve the accuracy and stability of overall predictions, and also enhance the robustness and generalization ability of the model.

[0079] Preferably, the contribution of clinical characteristics and target radiomics features to the model can be visualized through a nomogram, and indicators such as AUC, sensitivity, and specificity of each model can be calculated to evaluate the performance of the model.

[0080] Preferably, during the inference process, the model is also verified by statistical methods.

[0081] As an example, SPSS 25.0 software was used to analyze the data. The Kolmogorov-Smirnov test was used to test the normality of the measurement data. The measurement data that conformed to the normal distribution were analyzed by The independent sample t test was used to compare the two groups. The non-normally distributed quantitative data were expressed as M(Q1, Q3), and the non-parametric Mann-Whitney U test was used to compare the two groups. The count data were described as rate and frequency, and the chi-square test was used to compare the two groups. The multivariate logistic regression method was used to analyze the association between the different characteristics and the efficacy groups, and the indicators with P<0.1 in the comparison between the two groups were included in the multivariate classification model for further analysis. The DeLong test was used to compare the differences between different models. The delong test p value was less than 0.05, indicating that there was a statistical difference between the two models.

[0082] Based on the above, the present application also provides a system for evaluating the efficacy of stereotactic radiotherapy for spinal tumors. Figure 2 As shown, the therapeutic effect evaluation system includes a segmentation module 210 , a feature extraction module 220 and an evaluation module 230 .

[0083] The segmentation module 210 is used to automatically segment the tumor tissue and the surrounding healthy organs in the spinal tumor imaging image after stereotactic radiotherapy, and use the segmented tumor area as the target organ area.

[0084] The feature extraction module 220 is used to extract target imaging features from the image of the target organ region. The target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the efficacy evaluation results.

[0085] The evaluation module 230 is used to input the target radiomics features into the first efficacy evaluation model to obtain the efficacy evaluation prediction result.

[0086] Preferably, the evaluation module 230 is used to input the image of the target organ region into multiple efficacy evaluation models constructed by different algorithms, and use the efficacy evaluation model with the best model performance as the first efficacy evaluation model.

[0087] Preferably, the efficacy evaluation results include local control, toxicity, spinal cord injury and survival time.

[0088] Preferably, the segmentation module 210 is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy using a convolutional neural network.

[0089] Preferably, the feature extraction module 220 includes a standard extraction module 2201, a normalization module 2202, a first screening module 2203 and a second screening module 2204;

[0090] The standard extraction module 2201 is used to automatically extract the first radiomics feature of the region of interest that meets the IBSI standard from the image of the target organ region;

[0091] The normalization module 2202 is used to perform z-score normalization on all first radiomics features;

[0092] The first screening module 2203 is used to screen out insignificant features as the second radiomics features by using a univariate selection method;

[0093] The second screening module 2204 is used to screen out indicators related to the efficacy evaluation results from the second radiomics features as target radiomics features through minimum absolute shrinkage and selection operator regression.

[0094] Although some specific embodiments of the present application have been described in detail by way of example, it should be understood by those skilled in the art that the above examples are only for illustration, not for limiting the scope of the present application. It should be understood by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present application. The scope of the present application is defined by the appended claims.

Claims

1. A method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors, characterized in that: include: Automatically segment the tumor tissue and surrounding healthy organs in spinal tumor imaging images after stereotactic radiotherapy, and use the segmented tumor area as the target organ area; Extracting target imaging features from the image of the target organ region, wherein the target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the efficacy evaluation results; The target imaging omics features are input into the first efficacy evaluation model to obtain the efficacy evaluation prediction results.

2. The method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors according to claim 1, characterized in that: When obtaining the efficacy evaluation prediction results, the images of the target organ area are input into the efficacy evaluation models constructed by multiple different algorithms, and the efficacy evaluation model with the best model performance is used as the first efficacy evaluation model.

3. The method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors according to claim 1, characterized in that: The efficacy evaluation results include local control, toxicity, spinal cord injury and survival time.

4. The method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors according to claim 1, characterized in that: Automatic segmentation of spinal tumor imaging images after stereotactic radiotherapy using convolutional neural network.

5. The method for evaluating the efficacy of stereotactic radiotherapy for spinal tumors according to claim 2, characterized in that: Extracting target radiomics features from the image of the target organ region specifically includes: A first radiomics feature of a region of interest that meets the IBSI standard is automatically extracted from the image of the target organ region; All first radiomics features were z-score normalized; Using the univariate selection method, insignificant features were screened out as the second radiomics features; Through minimum absolute shrinkage and selection operator regression, indicators related to efficacy evaluation results were screened out from the second radiomics features as target radiomics features.

6. A system for evaluating the efficacy of stereotactic radiotherapy for spinal tumors, characterized in that: It includes segmentation module, feature extraction module and evaluation module; The segmentation module is used to automatically segment the tumor tissue and surrounding healthy organs in the spinal tumor imaging image after stereotactic radiotherapy, and use the segmented tumor area as the target organ area; The feature extraction module is used to extract target imaging features from the image of the target organ region, wherein the target imaging features include shape features, texture features, grayscale statistics features and high-order features related to the therapeutic effect evaluation results; The evaluation module is used to input the target imaging genomics features into the first efficacy evaluation model to obtain efficacy evaluation prediction results.

7. The therapeutic effect evaluation system for stereotactic radiotherapy of spinal tumors according to claim 6, characterized in that: The evaluation module is used to input the image of the target organ area into multiple efficacy evaluation models constructed by different algorithms, and use the efficacy evaluation model with the best model performance as the first efficacy evaluation model.

8. The therapeutic effect evaluation system for stereotactic radiotherapy of spinal tumors according to claim 6, characterized in that: The efficacy evaluation results include local control, toxicity, spinal cord injury and survival time.

9. The therapeutic effect evaluation system for stereotactic radiotherapy of spinal tumors according to claim 6, characterized in that: The segmentation module is used to automatically segment spinal tumor imaging images after stereotactic radiotherapy using a convolutional neural network.

10. The therapeutic effect evaluation system for stereotactic radiotherapy of spinal tumors according to claim 6, characterized in that: The feature extraction module includes a standard extraction module, a normalization module, a first screening module and a second screening module; The standard extraction module is used to automatically extract the first radiomics feature of the region of interest that meets the IBSI standard from the image of the target organ region; The normalization module is used to perform z-score normalization on all first radiomics features; The first screening module is used to screen out insignificant features as the second radiomics features by using a univariate selection method; The second screening module is used to screen out indicators related to the efficacy evaluation results from the second imaging omics features as target imaging omics features through minimum absolute shrinkage and selection operator regression.

Images