Helical tomography radiotherapy plan evaluation method and system and electronic equipment

Abstract

The invention relates to the technical field of artificial intelligence, and provides a spiral tomography radiotherapy plan evaluation method, which comprises the following steps: acquiring an RTPlan file, a CT image file and an RTStuct file corresponding to a spiral tomography radiotherapy plan; model input characteristic parameters are extracted from the RTPlan file, the CT image file and the RTStuct file, and the model input characteristic parameters are extracted from the RTPlan file, the CT image file and the RTStuct file; inputting the model input characteristic parameters into a gamma passing rate prediction model for calculation to obtain a gamma passing rate prediction result; and evaluating the spiral tomography radiotherapy plan based on the gamma passing rate prediction result, thereby improving the evaluation efficiency of the radiotherapy plan.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, system, electronic device and computer-readable storage medium for evaluating a helical tomographic radiotherapy plan. Background Art

[0002] Helical Tomotherapy (HT) is an advanced radiotherapy technology that provides highly conformal dose distribution by using intensity-modulated radiotherapy. Helical Tomotherapy plan is a cancer radiotherapy planning program based on the Helical Tomotherapy System (TOMO) device.

[0003] Gamma Passing Rate (GPR) is an indicator used to evaluate the accuracy of dose distribution in radiotherapy plans. It is used to compare the difference between the dose received by the patient in actual treatment and the planned dose distribution. In radiotherapy, ensuring accurate dose delivery is crucial to the treatment effect. Therefore, accurate prediction of gamma passing rate helps to improve the accuracy and efficiency of treatment plans.

[0004] In traditional technology, the method of obtaining gamma pass rate usually involves the following steps: 1. Planning system dose calculation: The planning system dose calculation is performed using the dose calculation algorithm used in clinical patient treatment.

[0005] 2. Dose verification: Dose verification includes point dose verification and dose distribution verification. Point dose verification includes selecting point dose measurement detectors and corresponding phantoms suitable for IMRT dose verification. The principle of selecting the measurement point position is to make the difference between the maximum dose and the minimum dose in the detector volume as small as possible; dose distribution verification includes using film, detector array, EPID, and calculation based on machine log files to verify dose distribution.

[0006] 3. Gamma analysis: Determine the gamma pass rate by comparing the difference between the measured dose distribution and the calculated dose distribution.

[0007] The study found that factors such as the accuracy of phantom positioning, the correctness of equipment parameters, and the accuracy of detector array and EPID measurement have a great impact on the results of dose distribution verification. In addition, the use of film, detector array, EPID and other means has a very high occupancy rate for spiral tomography linear accelerator treatment equipment. Therefore, the traditional verification and analysis process not only needs to ensure the accuracy of each verification operation, but also has a high equipment occupancy rate, resulting in more clinical working time of physicists and low efficiency. Summary of the invention

[0008] In order to improve the efficiency of radiotherapy plan evaluation, the embodiments of the present application provide a helical tomographic radiotherapy plan evaluation method, system, electronic device and computer storage medium.

[0009] The spiral tomographic radiotherapy plan evaluation method provided in the present application includes obtaining the RTPlan file, CT image file and RTStruct file corresponding to the spiral tomographic radiotherapy plan; extracting model input feature parameters from the RTPlan file and the CT image file and the RTStruct file; inputting the model input feature parameters into a gamma pass rate prediction model for calculation to obtain a gamma pass rate prediction result; and evaluating the spiral tomographic radiotherapy plan based on the gamma pass rate prediction result.

[0010] Based on the above technical solution, the gamma pass rate corresponding to the radiotherapy plan can be quickly obtained based on the gamma pass rate prediction model, thereby improving the evaluation efficiency of the radiotherapy plan. At the same time, when making predictions, the patient's CT image files are introduced to obtain more accurate prediction results, thereby improving the accuracy of radiotherapy plan evaluation.

[0011] In one implementation, the gamma pass rate prediction model is constructed based on the XGBoost algorithm.

[0012] In one implementation, the method for obtaining training samples of the gamma pass rate prediction model includes: constructing a case data set, the case data set including case data of completed radiotherapy containing a target number, the case data including historical RTPlan files, historical CT image files, and historical RTStruct files; according to a feature extraction strategy, extracting initial feature combination data from the historical RTPlan files, the historical CT image files, and the historical RTStruct files of the case data; preprocessing the initial feature combination data corresponding to each of the case data; performing feature screening analysis based on the preprocessed initial feature combination data to determine the optimal feature combination; and constructing the training sample based on the optimal feature combination.

[0013] Based on the above technical solution, the features in the case data set are comprehensively extracted and screened, and then the best feature combination is screened out to improve the accuracy of the prediction model.

[0014] In one implementation, the feature extraction strategy includes: extracting treatment planning system related features, absolute leaf opening time statistics related features, relative leaf opening time statistics related features, geometric sinusoidal features and modulated sinusoidal features from the historical RTPlan file; extracting imaging omics features from the historical CT image file and the historical RTStruct file.

[0015] In one embodiment, the treatment planning system related features include modulation factor, number of projections, number of projections per rotation, number of rotations, projection time, Gantry cycle, treatment time, field width, tilt, treatment bed movement distance, treatment bed speed, target length, and treatment time exceeding fractionated dose; the absolute leaf opening time statistics related features include LOT mean, LOT standard deviation, LOT median, LOT mode, LOT sharpness, LOT skewness, LOT maximum, LOT minimum, proportion of LOT less than 100, proportion of LOT less than 50, proportion of LOT less than 30, proportion of LOT less than 20, and proportion of LOT with a projection time interval less than 20 milliseconds; the relative leaf opening time statistics related features include Fractional LOT mean, Fractional LOT standard deviation, Fractional LOT median, Fractional LOT mode, Fractional LOT maximum, Fractional LOT minimum, proportion of Fractional LOT less than 5, proportion of Fractional LOT less than 10, proportion of Fractional LOT less than 25, Fractional LOT minimum, ... The proportion of LOT less than 50, the proportion of Fractional LOT less than 75, the proportion of Fractional LOT less than 90, and the proportion of Fractional LOT less than 95; the geometric sinusoidal graph features include 0 nearest neighbor open leaves, 1 nearest neighbor open leaves, 2 nearest neighbor open leaves, the number of connected components, the length of connected components, the treatment area, the proportion of discontinuous projections, the closed leaf fraction, the closed leaf fraction within the treatment area, and the center of mass; the modulated sinusoidal graph features include the number of openings and closings during treatment, the time variation coefficient of the planned sinusoidal graph, the time variation coefficient of the extended planned sinusoidal graph, the leaf opening time variation coefficient, the extended leaf opening time variation coefficient, the modulation index, the mean value of the sinusoidal graph, the mode of the sinusoidal graph, the standard deviation of the sinusoidal graph, and the symmetry of the average sinusoidal graph.

[0016] Based on the above technical solution, by selecting these feature parameters to construct initial feature combination parameter data, comprehensive extraction of relevant features can be guaranteed.

[0017] In one implementation, the preprocessing of the initial feature combination data corresponding to each case data includes: downsampling the initial feature combination data under the sample groupings corresponding to the regression task and the classification task, respectively.

[0018] Based on the above technical solution, by constructing training samples corresponding to classification tasks and regression tasks, the model can be trained to simultaneously output the gamma pass rate and the classification results under the preset threshold, so that users can obtain the evaluation results intuitively and quickly. At the same time, through the downsampling operation, the sample data can be balanced to ensure that the data distribution is relatively balanced and prevent overfitting in the prediction.

[0019] In one implementation, the feature screening analysis is performed based on the preprocessed initial feature combination data to determine the optimal feature combination, including: performing a first screening of the features in the initial feature combination based on correlation analysis to obtain a first feature combination; performing a second screening of the first feature combination based on a recursive feature elimination cross-validation algorithm to obtain a second feature combination; and performing SHAP value evaluation and selection on the second feature combination to determine the optimal feature combination.

[0020] Based on the above technical solution, the best feature combination is obtained through multiple screenings, which can ensure that the selected best feature combination can improve the prediction accuracy and performance of the model.

[0021] Based on the same inventive concept, an embodiment of the present application further provides a spiral tomotherapy plan evaluation system, which is used to implement the above-mentioned spiral tomotherapy plan evaluation method.

[0022] In addition, an embodiment of the present application also provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, and the program or instruction implements the above method when executed by the processor.

[0023] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings constituting a part of the present application are used to provide further understanding of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute improper limitations on the present application.

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A flow chart of a helical tomotherapy plan evaluation method provided in an embodiment of the present application is shown.

[0027] Figure 2 A flow chart of a method for constructing a gamma pass rate prediction model in an embodiment of the present application is shown.

[0028] Figure 3 A flow chart of a method for obtaining training samples in an embodiment of the present application is shown.

[0029] Figure 4 A flow chart of the pretreatment method in the embodiment of the present application is shown.

[0030] Figure 5 A flow chart of the feature screening analysis method in the embodiments of the present application is shown.

[0031] Figure 6 A schematic diagram of the structure of a helical tomotherapy plan evaluation system provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0033] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, and "first", "second" and various digital numbers are only distinguished for the convenience of description and are not used to limit the scope of the embodiments of the present application.

[0034] The features, structures or characteristics in this application may be combined in one or more embodiments in any suitable manner. In various embodiments of this application, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0035] Some optional features in the embodiments of the present application may be implemented independently in some scenarios without relying on other features to solve corresponding technical problems and achieve corresponding effects. They may also be combined with other features in some scenarios as needed.

[0036] In this application, unless otherwise specified, the same or similar parts between the various embodiments can refer to each other. In the various embodiments of this application, if there is no special description and logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships. The implementation methods of this application do not constitute a limitation on the scope of protection of this application.

[0037] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0038] The embodiment of the present application provides a spiral tomography radiotherapy plan evaluation method, which realizes automatic prediction of gamma pass rate by combining artificial intelligence technology without occupying any experimental verification equipment, thereby improving the efficiency of gamma pass rate acquisition and saving the clinical work time of physicists.

[0039] Please refer to Figure 1 The method provided in the embodiment of the present application specifically includes the following steps.

[0040] S101, obtaining an RTPlan file, a CT image file and an RTStruct file corresponding to a spiral tomographic radiotherapy plan.

[0041] In implementation, the evaluation system can obtain the RTPlan file, CT image file and RTStruct file corresponding to the spiral tomography radiotherapy plan to be evaluated by connecting with the radiotherapy planning system or receiving user input, wherein the CT image file is the CT image file in the patient treatment data.

[0042] S102, extracting model input feature parameters from the RTPlan file, the CT image file, and the RTStruct file.

[0043] Among them, the model input feature parameters include a combination of multiple feature parameters, which are specifically determined based on the input layer of the gamma pass rate prediction model, which will be explained below.

[0044] In one implementation, the corresponding feature parameters can be extracted from the RTPlan file using the open source matlab library TCoMx based on the matlab programming environment, and the corresponding feature parameters can be extracted from the CT image file and the RTStruct file using the open source python library pyRadomics based on the python programming environment.

[0045] S103, inputting the model input feature parameters into the gamma pass rate prediction model for calculation to obtain the gamma pass rate prediction result.

[0046] In an embodiment of the present application, the predicted results output by the model may include dose distributions of global 3% / 2mm, local 3% / 2mm, global 2% / 2mm, and local 2% / 2mm, the gamma pass rates corresponding to the four evaluation criteria, and the classification results under a preset threshold: pass or fail.

[0047] S104, evaluating the helical tomotherapy plan based on the gamma pass rate prediction results.

[0048] It can be seen that based on the method provided in the embodiment of the present application, the user only needs to provide the RTPlan file, CT image file and RTStruct file corresponding to the spiral tomography plan to directly obtain the gamma pass rate prediction result to evaluate the spiral tomography plan, which is not only efficient but also does not occupy equipment resources; further, the gamma pass rate prediction model provided in the embodiment of the present application introduces multi-dimensional feature parameters into the input layer, and at the same time effectively utilizes the imaging genomics features and other information in the patient treatment data, thereby providing a more comprehensive and multi-dimensional data basis for the prediction of the gamma pass rate, which helps to ensure the accuracy of the model's prediction results.

[0049] In one example, the gamma pass rate prediction model provided in an embodiment of the present application can be constructed based on the XGBoost (eXtreme Gradient Boosting) algorithm, wherein XGBoost is an efficient, flexible and scalable machine learning algorithm based on a gradient boosting framework. The gamma pass rate prediction model constructed based on XGBoost has the advantages of high efficiency, good prediction performance, good interpretability, high flexibility and strong stability.

[0050] Please refer to Figure 2 The method for constructing a gamma pass rate prediction model in the embodiment of the present application specifically includes the following steps: S201, obtaining training samples.

[0051] For details, please refer to Figure 3 , the method for obtaining training samples in the embodiment of the present application includes the following specific steps.

[0052] S301, construct a case data set.

[0053] The case data set includes case data of completed radiotherapy with a target number, and each case data includes historical RTPlan files, historical CT image files, historical RTStruct files, and the gamma pass rate of the measured dose distribution.

[0054] In one example, 300 cases that have completed radiotherapy on the Tomo system, mainly including head and neck, abdominal and pelvic tumors, can be selected and all case data can be collected. The gamma pass rate of the measured dose distribution is the actual gamma pass rate obtained after the treatment is completed, where the gamma pass rate of the measured dose distribution includes the gamma pass rate of the dose distribution under four evaluation criteria: global 3% / 2mm, local 3% / 2mm, global 2% / 2mm, and local 2% / 2mm, which are used as the labeled data in the training sample.

[0055] S302, extracting initial feature combination data from the historical RTPlan files, historical CT image files, and historical RTStruct files of the case data according to the feature extraction strategy.

[0056] Specifically, after studying and analyzing the parameter data contained in each file, the features that may be related to the gamma pass rate can be preliminarily determined, and then the corresponding feature extraction strategy can be generated to extract the initial feature combination data.

[0057] In one implementation, the feature extraction strategy includes: extracting 66-dimensional complexity parameter features from historical RTPlan files, wherein the complexity feature parameters include treatment planning system related features, absolute leaf opening time statistics related features, relative leaf opening time statistics related features, geometric sinusoidal features, and modulation sinusoidal features. And extracting radiomics features from historical CT image files and historical RTStruct files.

[0058] In one example, for each patient in the case data set, according to the RTPlan file in the patient case data, based on the matlab programming environment and using the open source matlab library TCoMx, a complexity parameter feature with a dimension of 63 is extracted, including modulation factor (MF), number of projections, number of projections per rotation, number of rotations, projection time, Gantry cycle, treatment time (TT), field width (FW), pitch, treatment bed movement distance, treatment bed speed, target length, and treatment time exceeding fractionated dose, a total of 13 treatment planning system related characteristic parameters; LOT (Leaf Opening Time, leaf clocking time) mean, LOT standard deviation, LOT median, LOT mode, LOT sharpness, LOT skewness, LOT maximum, LOT minimum, proportion of LOT less than 100, proportion of LOT less than 50, proportion of LOT less than 30, proportion of LOT less than 20, and proportion of LOT with projection time interval less than 20 milliseconds, a total of 13 parameters related to absolute leaf opening time statistics, Fractional The LOT mean, fractional LOT standard deviation, fractional LOT median, fractional LOT mode, fractional LOT maximum, fractional LOT minimum, proportion of fractional LOT less than 5, proportion of fractional LOT less than 10, proportion of fractional LOT less than 25, proportion of fractional LOT less than 50, proportion of fractional LOT less than 75, proportion of fractional LOT less than 90 and proportion of fractional LOT less than 95, a total of 13 parameters related to relative leaf opening time statistics; 0 nearest neighbor open leaf, 1 nearest neighbor open leaf, 2 nearest neighbor open leaves, number of connected components, length of connected components, treatment area, discontinuous projection ratio, closed leaf fraction, closed leaf fraction within treatment area (CLS) in 、CLS in,disc 、CLS in,area 、CLS in,area,disc ) and centroid, a total of 13 geometric sinusoidal parameters; the number of opening and closing during treatment, the time variation coefficient of the planned sinusoidal, the time variation coefficient of the extended planned sinusoidal, the time variation coefficient of the blade opening, the time variation coefficient of the extended blade opening, the modulation index, the mean of the sinusoidal, the mode of the sinusoidal, the standard deviation of the sinusoidal, and the mean sinusoidal symmetry, a total of 11 modulation sinusoidal parameters.

[0059] Based on the Python programming environment and using the open source Python library pyRadomics, multi-dimensional radiomics features are extracted, including shape features, first-order features, gray-level co-occurrence matrix features, gray-level size area matrix features, gray-level run matrix features, adjacent gray-level difference matrix features, and gray-level dependence matrix features.

[0060] It is worth noting that the feature parameters in the above examples are feature dimensions selected through correlation analysis based on the record information in the current files. It is understandable that when different application goals or file standards change, the feature extraction strategy can be set according to the actual situation so that the proposed initial feature data can cover the initial parameters related to the gamma prediction rate as much as possible.

[0061] S303, pre-processing the initial feature combination data corresponding to each case data.

[0062] For details, please refer to Figure 4 , the preprocessing method specifically includes the following steps.

[0063] S401, remove the initial feature combination data containing missing values ​​to ensure the integrity of the data.

[0064] S402, grouping the remaining initial feature combination data.

[0065] Specifically, according to the training objectives, accurate training samples for regression tasks and training samples for classification tasks are required. In one example, all remaining initial feature combination data can be divided into two groups, that is, the initial feature combination data corresponding to the regression task and the classification task are the same. In another example, the remaining initial feature data can also be randomly divided into the regression task group or the classification task group.

[0066] S403, down-sampling the grouped initial feature combination data.

[0067] In one example, the downsampling method for regression task grouping includes: selecting 94% as the threshold when regressing the target of 3% / 2mm, selecting 85% as the threshold when regressing the target of 2% / 2mm, respectively counting the number of initial feature combination data less than the threshold and greater than or equal to the threshold, and randomly selecting part of the data from the larger amount of data, the amount of which is 3 times that of the smaller amount of data. The downsampling method for classification task grouping includes: selecting 10 different thresholds from 90% to 99%, binary classification of data at different thresholds, setting the target value greater than or equal to the threshold to 1, and setting the target value less than the threshold to 0, counting the number of 0 and 1, and determining the smaller part, and randomly selecting a part of the data from the larger part so that its amount is the same as the smaller part.

[0068] Through these grouping and downsampling operations, the data is balanced to ensure that the data distribution is relatively balanced and to prevent overfitting in the prediction.

[0069] S404, optimizing the initial feature combination parameters after downsampling.

[0070] Specifically, if a parameter in the initial feature combination after downsampling has only one unique value, the feature can be deleted. At the same time, the repeated initial feature combination data is deleted.

[0071] After completing the preprocessing based on the above method, the following steps are also included.

[0072] S304, performing feature screening analysis based on the preprocessed initial feature combination data to determine the best feature combination.

[0073] Please refer to Figure 5 In this step, the feature screening analysis method specifically includes the following steps: S501, performing a first screening of features in the initial feature combination based on correlation analysis to obtain a first feature combination.

[0074] In implementation, the first feature combination can be constructed by first analyzing the correlations between different features and between each feature and the target variable, screening out features that are highly correlated with the target variable and have low correlations with other features.

[0075] When performing regression and classification predictions, if there is a high correlation between features, it will lead to instability and inaccuracy of the prediction model. This can reduce redundant features and improve the training efficiency and prediction performance of the model.

[0076] In a specific example, the correlation between each feature and the target variable can be calculated first to obtain a correlation coefficient matrix, and then the features can be sorted based on the correlation coefficient matrix. Then, the correlation matrix of the arranged feature matrix is ​​calculated to obtain the correlation between different features. According to the set threshold of 0.9, feature pairs with high correlation are found, and indexes with correlation greater than 0.9 and less than or equal to 1 are screened, and features with high correlation are deleted.

[0077] S502, performing secondary screening on the first feature combination based on a recursive feature elimination cross-validation algorithm to obtain a second feature combination.

[0078] In implementation, recursive feature elimination cross-validation can be performed based on 5-fold cross-validation with mean absolute error as an evaluation indicator to further refine the feature set and obtain the second feature combination.

[0079] S503, performing SHAP value evaluation and selection on the second feature combination to determine the best feature combination.

[0080] In implementation, we can first extract and calculate the SHAP importance of each feature, then sort the features by importance based on the SHAP value, gradually increase the number of features used according to the feature importance, use the XGBoost model for training, and use different seeds for data segmentation, calculate the prediction error, and find the best feature combination.

[0081] S305, constructing training samples based on the best feature combination.

[0082] After determining the best feature combination, the parameter value corresponding to each feature can be obtained based on the corresponding initial combination feature data, and the corresponding measured gamma pass rate and classification under different thresholds can be used as annotations to generate training samples.

[0083] In one example, the best feature combination of the embodiments of the present application includes: For the 3% / 2mm regression task, the best feature combination is: LOT sharpness, proportion of LOT less than 100, mean of sinogram, mode of sinogram, tilt, bed speed, number of projections, number of rotations, and 40-dimensional radiomics features.

[0084] For the 2% / 2mm regression task, the best feature combination is: LOT sharpness, mean of sinogram, mode of sinogram, tilt, couch speed, number of projections, number of rotations, and 34-dimensional radiomics features.

[0085] For the 3% / 2mm classification task, the best feature combination is: LOT mode, LOT minimum, proportion of LOT less than 100, Fractional LOT mode, Fractional LOT maximum, sinusoidal mean, sinusoidal mode, treatment time, target length, treatment bed movement distance, number of projections, number of rotations, and 48-dimensional radiomics features.

[0086] For the 2% / 2mm classification task, the best feature combination was: Fractional LOT standard deviation, Fractional LOT maximum, number of opening and closing times during treatment, modulation factor, mean of sinusoidal graph, mode of sinusoidal graph, time variation coefficient of planned sinusoidal graph, time variation coefficient of extended leaf opening, proportion of Fractional LOT less than 75, and 33-dimensional radiomics features.

[0087] It is worth noting that in the process of extracting the model input feature parameters in the above step S102, the corresponding feature parameters are extracted from each file with reference to the optimal feature combination.

[0088] S202, determining optimal model parameters.

[0089] In one example, a parameter search may be performed using the Hyperopt library, and the optimal parameters of the gamma pass rate prediction model may be found using Bayesian optimization.

[0090] S203, model initialization.

[0091] The gamma pass rate prediction model is initialized based on the optimal feature combination and optimal parameters.

[0092] S204, model training and evaluation.

[0093] In implementation, in order to evaluate the stability of the model, the dataset can be split, trained, and evaluated multiple times with multiple random seeds.

[0094] In one example, a test set can be randomly divided from all training sample sets, where the number of test sets accounts for 20% of the total; and the remaining training sample sets are randomly split into training sets and validation sets, where the validation sets account for 10% of the remaining amount and the training sets account for 90% of the remaining amount.

[0095] After determining the training set, validation set, and test set, the initialized prediction model is first trained using the training set, and the trained prediction model is verified based on the validation set. Then, the feature parameters in the test set are input into the trained prediction model to obtain the output prediction results, and then the prediction results are compared with the true values ​​corresponding to the feature parameters to determine the accuracy of the prediction results. In this way, after multiple splits and evaluations, the average values ​​of each evaluation indicator are calculated, and the mean absolute error (MAE), root mean square error (RMSE), and classification accuracy are calculated to verify the performance of the gamma pass rate prediction model.

[0096] S205, validate the model based on a multicenter dataset.

[0097] In implementation, multi-center data can be used to test the model and evaluate the performance of the prediction model on different data sets to ensure the generalization ability, stability and reliability of the model. Among them, a multi-center data set can be constructed by obtaining case data from different regions and institutions.

[0098] Specifically, feature variables and target variables are first extracted from the multi-center dataset, and then the extracted feature variables and target variables are preprocessed to make them consistent with the training sample format. The trained prediction model is then used to predict the feature variables, and the mean absolute error (MAE), root mean square error (RMSE), and classification accuracy are calculated to verify and evaluate the performance of the prediction model.

[0099] The performance evaluation results of the gamma pass rate prediction model constructed based on the above method are as follows: For the 3% / 2mm regression task, the performance evaluation results are: MAE is 1.3342 and RMSE is 2.2096 on the single-center test set; MAE is 1.4608 and RMSE is 2.2096 on the multi-center test set.

[0100] For the 2% / 2mm regression task, the performance evaluation results are: MAE is 2.7612 and RMSE is 4.0668 on the single-center test set; MAE is 3.1458 and RMSE is 3.8531 on the multi-center test set.

[0101] For the 3% / 2mm classification task, the performance evaluation results are: the classification accuracy on the single-center test set is 80%; the classification accuracy on the multi-center test set is 90%.

[0102] For the 2% / 2mm classification task, the performance evaluation results are: the classification accuracy on the single-center test set is 90%; the classification accuracy on the multi-center test set is 85%.

[0103] It can be seen that based on the above technical scheme, comprehensive complexity parameter features and imaging genomics features are extracted according to the patient's CT image files, RTPlan files, RTStruct files and other treatment data. After feature preprocessing, effective features are systematically screened out through correlation analysis, recursive feature elimination cross-validation and SHAP value evaluation methods, and the Bayesian optimization algorithm is used to optimize the model parameters, which significantly improves the predictive performance and stability of the gamma pass rate prediction model. It is also verified through multi-center data, so that the generalization ability on different data sets is verified, which can effectively improve the accuracy and efficiency of radiotherapy plan evaluation.

[0104] Based on the same inventive concept, the present application also provides a spiral tomotherapy plan evaluation system for executing the above method, please refer to Figure 6 The spiral tomography radiotherapy plan evaluation system includes an input module 61 , a data processing module 62 , an analysis and calculation module 63 and an output module 64 .

[0105] The input module 61 is used to obtain the RTPlan file, CT image file and RTStruct file corresponding to the helical tomography radiotherapy plan to be evaluated, and to verify the integrity of each file. After the verification is passed, each file is sent to the data processing module 62. The method of obtaining each file includes receiving a file uploaded by a user, or requesting a corresponding file from the radiotherapy planning system based on a user operation.

[0106] The data processing module 62 is used to extract the model input characteristic parameters from the RTPlan file, the CT image file and the RTStruct file, and send the input characteristic parameters to the analysis and calculation module 63 .

[0107] The analysis and calculation module 63 is used to input the model input feature parameters into the gamma pass rate prediction model for calculation, obtain the gamma pass rate prediction result, and display it to the user through the output module 64.

[0108] In addition, an embodiment of the present application also provides an electronic device, which includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements a method as in any one of the implementations in the embodiments of the present application; wherein the processor may adopt a general-purpose central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), a graphics processing unit (GPU), or one or more integrated circuits, for executing relevant programs to implement the method in any one of the implementations in the embodiments of the present application.

[0109] The processor may also be an integrated circuit electronic device with signal processing capability. In the implementation process, each step of the method in any implementation of the embodiments of the present application may be completed by an integrated logic circuit of hardware in the processor or by instructions in software form.

[0110] The above-mentioned processor can also be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be combined and executed.

[0111] The software module may be located in a random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register or other mature storage media in the art. The storage medium is located in the memory, and the processor reads the information in the memory, and combines its hardware to complete the functions required to be performed by the units included in the data processing device of the embodiment of the present application, or executes the method in any one of the implementation modes in the embodiment of the present application.

[0112] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method when executed by a processor.

[0113] Those skilled in the art can understand that all or part of the steps in the above-mentioned implementation methods can be completed by instructing related hardware through a program, and the program is stored in a storage medium, including several instructions for making a device (which can be a single-chip microcomputer, chip, etc.) or a processor (processor) perform all or part of the steps of each implementation method of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0114] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for evaluating a helical tomotherapy plan, characterized in that: The method comprises the steps of: Obtain the RTPlan file, CT image file and RTStruct file corresponding to the spiral tomography radiotherapy plan; Extracting model input feature parameters from the RTPlan file, the CT image file and the RTStruct file; Inputting the model input feature parameters into the gamma pass rate prediction model for calculation to obtain the gamma pass rate prediction result; The helical tomotherapy plan is evaluated based on the gamma pass rate prediction result.

2. The method according to claim 1, characterized in that The gamma pass rate prediction model is constructed based on the XGBoost algorithm.

3. The method according to claim 1 or 2, characterized in that: The method for obtaining training samples of the gamma pass rate prediction model includes: Constructing a case data set, wherein the case data set includes case data of completed radiotherapy of a target number, wherein the case data includes a historical RTPlan file, a historical CT image file, and a historical RTStruct file; According to the feature extraction strategy, initial feature combination data is extracted from the historical RTPlan file, the historical CT image file and the historical RTStruct file of the case data; Preprocessing the initial feature combination data corresponding to each case data; Performing feature screening analysis based on the preprocessed initial feature combination data to determine the best feature combination; The training samples are constructed based on the optimal feature combination.

4. The method according to claim 3, characterized in that The feature extraction strategy includes: Extracting treatment planning system related features, absolute blade opening time statistics related features, relative blade opening time statistics related features, geometric sinusoidal features, and modulated sinusoidal features from the historical RTPlan file; Radiomics features are extracted from the historical CT image files and the historical RTStruct files.

5. The method according to claim 4, characterized in that The treatment planning system-related features include modulation factor, number of projections, number of projections per rotation, number of rotations, projection time, Gantry cycle, treatment time, field width, tilt, treatment bed movement distance, treatment bed speed, target length, and treatment time exceeding fractionated doses; The features related to the absolute leaf opening time statistics include LOT mean, LOT standard deviation, LOT median, LOT mode, LOT cusp, LOT skewness, LOT maximum, LOT minimum, proportion of LOT less than 100, proportion of LOT less than 50, proportion of LOT less than 30, proportion of LOT less than 20, and proportion of LOT with a projection time interval less than 20 milliseconds; The relative leaf opening time statistics-related features include the Fractional LOT mean, the Fractional LOT standard deviation, the Fractional LOT median, the Fractional LOT mode, the Fractional LOT maximum, the Fractional LOT minimum, the proportion of Fractional LOT less than 5, the proportion of Fractional LOT less than 10, the proportion of Fractional LOT less than 25, the proportion of Fractional LOT less than 50, the proportion of Fractional LOT less than 75, the proportion of Fractional LOT less than 90, and the proportion of Fractional LOT less than 95; the geometric sinusoidal graph features include 0 nearest neighbor open leaves, 1 nearest neighbor open leaves, 2 nearest neighbor open leaves, the number of connected components, the length of connected components, the treatment area, the discontinuous projection ratio, the closed leaf fraction, the closed leaf fraction within the treatment area, and the centroid; The modulated sinusoidal features include the number of opening and closing times during treatment, the time variation coefficient of the planned sinusoidal, the time variation coefficient of the extended planned sinusoidal, the time variation coefficient of the blade opening, the time variation coefficient of the extended blade opening, the modulation index, the mean of the sinusoidal, the mode of the sinusoidal, the standard deviation of the sinusoidal, and the symmetry of the average sinusoidal.

6. The method according to claim 3, characterized in that The preprocessing of the initial feature combination data corresponding to each case data includes: The initial feature combination data under the sample groups corresponding to the regression task and the classification task are downsampled respectively.

7. The method according to claim 3, characterized in that The performing feature screening analysis based on the preprocessed initial feature combination data to determine the best feature combination comprises: Performing a first screening of the features in the initial feature combination based on correlation analysis to obtain a first feature combination; Performing a secondary screening on the first feature combination based on a recursive feature elimination cross-validation algorithm to obtain a second feature combination; Perform SHAP value evaluation and selection on the second feature combination to determine the optimal feature combination.

8. A helical tomotherapy plan evaluation system, characterized in that: The system is used to implement the helical tomotherapy plan evaluation method according to any one of claims 1 to 7.

9. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction implements the method according to any one of claims 1 to 7 when executed by the processor.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images