Equivalent photon dose prediction method and device of BNCT dose, storage medium and terminal

Abstract

The invention provides an equivalent photon dose prediction method and device for a BNCT dose, a storage medium and a terminal, and the method comprises the steps: inputting a to-be-predicted dose into a first equivalent photon dose calculation model, so as to obtain a corresponding calculation result as a preliminary result; obtaining a prediction result corresponding to the preliminary result by using an equivalent photon mapping AI model, and taking the prediction result as an equivalent photon dose prediction result; the equivalent photon mapping AI model is used for predicting a second result under the same preset dose based on the first result; the first result is a calculation result obtained by inputting a preset dose into a first equivalent photon dose calculation model for calculation; the second result is a calculation result obtained by inputting a preset dose into a second equivalent photon dose calculation model for calculation; wherein the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model. According to the invention, equivalent dose prediction can be rapidly and accurately carried out.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology and relates to a method for predicting an equivalent photon dose of a BNCT dose, and in particular to a method and device for predicting an equivalent photon dose of a BNCT dose, a storage medium and a terminal. Background Art

[0002] Particle therapy for tumors has now become one of the most effective tools for killing tumors. Among the many particle therapy for tumors, Boron Neutron Capture Therapy (BNCT) has received excellent treatment evaluations in clinical trials for recurrent head and neck tumors, malignant glioblastomas, and malignant meningiomas due to its unique "drug-device" characteristics. The radiation field of BNCT is complex, with four main components: 10 When element B is irradiated by thermal neutrons 10 B(n,α) 7 Li + Nuclear reactions produce alpha particles. 7 Li + radiation, the boron dose deposited in the tissue; nitrogen element is irradiated by thermal neutrons 14 N(n,p) 14 The C reaction releases 0.54 MeV proton radiation, and the nitrogen dose deposited in the tissue; when hydrogen is irradiated by epithermal neutrons, 1 Recoil proton rays produced by H(n,n)p, hydrogen dose deposited in tissues; produced when irradiated by thermal neutrons 1 H(n,γ) 2 Photon rays generated by H reactions, stray photon rays generated by hardware equipment such as moderators and collimators under neutron irradiation, and photon doses deposited in tissues. Due to the different action characteristics of the above rays themselves, the four dose values ​​contain multiple ray action results, making it difficult to directly establish the post-irradiation biological effects containing the four dose values, making it difficult to accurately give the irradiation dose suitable for a certain tissue in BNCT treatment, which is not conducive to the comparison of treatment effects between different radiotherapy devices. If two different types of radiation produce the same biological effect results at different absorbed doses, then these two different absorbed dose values ​​are called equivalent doses. In the implementation of BNCT treatment, it is often necessary to establish a precise correlation method between the four doses of BNCT and the photon dose. The dose equivalent to photon irradiation obtained by converting and calculating the above four doses is called the "equivalent photon dose."

[0003] In the prior art, there are two ways to obtain the equivalent photon dose in BNCT. The first is to use the simple multiple calculation method of "fixed RBE value", by directly comparing the dose of the four components of BNCT with the irradiated biological results produced by photon irradiation, calculating the multiple relationship of each component converted into an equivalent photon dose, and obtaining the corresponding multiple relationships of the four doses (the multiple values ​​in BNCT include relative biological effect value (RBE) and composite biological effect value (CBE. Since the definitions of CBE and RBE are similar, the present invention refers to this method as "fixed RBE value" algorithm), and then multiplying and summing the corresponding values ​​to obtain the equivalent dose. This method has a fast calculation speed but low accuracy. The second is the dose-effect model. There are many types of microdose dynamics models, such as the improved random microdose kinetics model, the simplified microdose kinetics model, the improved random microdose kinetics model based on Taylor expansion approximation based on mathematical methods, or the improved random microdose kinetics model based on fast Fourier change. These models have accelerated the calculation time of the BNCT dose-effect model algorithm to a certain extent. In addition, there are dose-effect models based on nanodosimetry. This type of model accurately describes the energy deposition characteristic response of BNCT-generated rays on the scale of DNA, but due to the reduction of spatial scale, the calculation process of the equivalent photon dose of this type of model will also have a long time cost.

[0004] In summary, using the "fixed RBE value" calculation, the simplified theoretical model microdose kinetic model (MK) in the microdose effect model and other simple calculation methods have a relatively fast calculation speed, but the accuracy of the equivalent dose will be lost compared with the theoretical value; and the complex calculation method of the improved stochastic microdose kinetic model (SMK) has more accurate calculation results, but the calculation time cost is high. Therefore, the existing technology lacks a method for obtaining equivalent photon dose that has both calculation efficiency and accuracy. Summary of the invention

[0005] The object of the present invention is to provide a method and device for predicting an equivalent photon dose of a BNCT dose, a storage medium and a terminal, so as to solve the technical problem in the prior art that it is difficult to obtain an equivalent photon dose with both efficiency and accuracy.

[0006] In a first aspect, the present invention provides a method for predicting an equivalent photon dose of a BNCT dose, comprising:

[0007] Inputting the dose to be predicted into the first equivalent photon dose calculation model to obtain corresponding preliminary results;

[0008] Using an equivalent photon mapping AI model to obtain a prediction result corresponding to the preliminary result as an equivalent photon dose prediction result;

[0009] The equivalent photon mapping AI model is used to predict a second result at the same preset dose based on the first result;

[0010] The first result is a calculation result of inputting the preset dose into a first equivalent photon dose calculation model for calculation;

[0011] The second result is a calculation result of inputting the preset dose into a second equivalent photon dose calculation model for calculation;

[0012] Among them, the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model.

[0013] In one embodiment of the present invention, obtaining a prediction result corresponding to the preliminary result using an equivalent photon mapping AI model includes:

[0014] Selecting a prediction sub-model corresponding to the preliminary result according to the dose breakpoint data, if the preliminary result is less than the dose breakpoint data, the corresponding prediction sub-model is the first sub-model, otherwise the corresponding prediction sub-model is the second sub-model;

[0015] Using the preliminary results as input data, using the corresponding prediction sub-model to obtain output results as prediction results of the equivalent photon mapping AI model;

[0016] Among them, the equivalent photon mapping AI model includes a first sub-model and a second sub-model, the first sub-model is a machine learning regression model, and the second sub-model is a neural network model.

[0017] In an embodiment of the present invention, the input data further includes a supplementary feature, and the supplementary feature is a feature obtained by performing a polynomial expansion based on the preliminary result.

[0018] In one embodiment of the present invention, the method for obtaining the dose breakpoint data includes:

[0019] Sorting all data in the training data set of the equivalent photon mapping AI model according to dose size to obtain a sorted data set;

[0020] Using an unsupervised clustering algorithm to divide the sorted data set into a first cluster and a second cluster;

[0021] A maximum dose value of the first cluster and a minimum dose value of the second cluster are obtained, and any dose value between the maximum dose value and the minimum dose value is used as dose breakpoint data.

[0022] In one embodiment of the present invention, the evaluation indicators of the equivalent photon mapping AI model include:

[0023] The root mean square error of the data whose dose sizes in the first cluster and the second cluster are within a preset range from the dose breakpoint data, the root mean square error of the first cluster and the second cluster, and the determination coefficient of the sorted data set.

[0024] In one embodiment of the present invention, the method for acquiring the training data set of the equivalent photon mapping AI model includes:

[0025] Setting a neutron source item identical to that of the BNCT device in the simulation software, setting a tissue region of a preset size at the front end of the source item, and setting a boron concentration in the tissue region to a preset concentration;

[0026] Using materials of different preset tissues to simulate particle transport and energy deposition in all grids of the tissue region, respectively, to obtain corresponding deposition dose results;

[0027] The deposition dose result is input into a first equivalent photon dose calculation model to obtain a corresponding calculation result as a prediction factor, the deposition dose result is input into a second equivalent photon dose calculation model to obtain a corresponding calculation result as a prediction result, and the prediction factor and the corresponding prediction result are combined to obtain a training data set.

[0028] In one embodiment of the present invention, the dose to be predicted includes a boron dose, a hydrogen dose, a nitrogen dose and a photon dose of a preset cell, and the preset cell is a tumor cell, a skin cell, a mucosal cell or a nerve tissue cell.

[0029] In a second aspect, the present invention further provides a device for predicting an equivalent photon dose of a BNCT dose, characterized in that it comprises:

[0030] A calculation module, used for inputting the dose to be predicted into a first equivalent photon dose calculation model to obtain a corresponding calculation result as a preliminary result;

[0031] An inference module, used for obtaining a prediction result corresponding to the preliminary result by using an equivalent photon mapping AI model as an equivalent photon dose prediction result;

[0032] The equivalent photon mapping AI model is used to predict a second result at the same preset dose based on the first result;

[0033] The first result is a calculation result of inputting the preset dose into a first equivalent photon dose calculation model for calculation;

[0034] The second result is a calculation result of inputting the preset dose into a second equivalent photon dose calculation model for calculation;

[0035] Among them, for any dose to be predicted, the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model.

[0036] In a third aspect, the present invention further provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting an equivalent photon dose of a BNCT dose as described above.

[0037] In a fourth aspect, the present invention further provides a terminal, comprising a processor and a memory, wherein the memory is communicatively connected to the processor;

[0038] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the equivalent photon dose prediction method of the BNCT dose as described above.

[0039] As described above, the method and device, storage medium and terminal for predicting the equivalent photon dose of BNCT dose described in the present invention have the following beneficial effects:

[0040] By integrating equivalent dose calculation models with different precision and calculation speeds, and using the equivalent photon mapping AI model to realize the mapping relationship construction of two equivalent photon dose calculation models, the trade-off problem between precision and speed has been successfully solved. Its core advantage lies in using the first equivalent photon dose calculation model to quickly perform preliminary calculations, and then mapping the preliminary results to high-precision results through the equivalent photon mapping AI model, thereby improving prediction accuracy while ensuring calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic flow chart of a method for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention is shown.

[0042] Figure 2 A schematic diagram showing the data flow of the method for predicting the equivalent photon dose of the BNCT dose according to an embodiment of the present invention is shown.

[0043] Figure 3 A schematic diagram of the data flow process for obtaining prediction results based on a two-stage AI model in the method for predicting equivalent photon dose of BNCT dose described in an embodiment of the present invention is shown.

[0044] Figure 4 A schematic diagram of a data flow process for obtaining dose breakpoint data in a method for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention is shown.

[0045] Figure 5A schematic diagram of the structure of a device for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention is shown.

[0046] Figure 6 A schematic diagram of the structure of a terminal according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0047] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0048] The principles and implementation methods of the method and device for predicting the equivalent photon dose of the BNCT dose, the storage medium and the terminal of this embodiment will be described in detail below, so that those skilled in the art can understand the method, device, storage medium and terminal for predicting the equivalent photon dose of the BNCT dose of this embodiment without creative work.

[0049] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention provides a method for predicting an equivalent photon dose of a BNCT dose.

[0050] Figure 1 FIG. 1 is a flow chart showing a method for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention, with reference to FIG. Figure 1 As shown, the method for predicting the equivalent photon dose of BNCT dose according to the embodiment of the present invention mainly includes steps S100 to S200.

[0051] Step S100: inputting the dose to be predicted into the first equivalent photon dose calculation model to obtain the corresponding calculation result as a preliminary result.

[0052] Specifically, the dose to be predicted is the dose value in BNCT treatment, and the equivalent photon dose of the dose to be predicted is the photon irradiation dose corresponding to the dose to be predicted. The purpose of an embodiment of the present invention is to use the dose to be predicted as input data, and obtain the photon irradiation dose corresponding to the dose to be predicted as an equivalent photon dose prediction result. In step S100, the first equivalent photon dose calculation model is first used to perform calculations to obtain preliminary results corresponding to the dose to be predicted. The function of the first equivalent photon dose calculation model is to quickly calculate the equivalent photon metering results as preliminary results. Optionally, the first equivalent photon dose calculation model includes a "fixed RBE value" calculation model, a simplified theoretical model microdose kinetic model (MK) in the microdose effect model, and an improved random microdose kinetic model based on Taylor expansion.

[0053] Step S200: Using the equivalent photon mapping AI model to obtain a prediction result corresponding to the preliminary result as an equivalent photon dose prediction result.

[0054] In this embodiment, the equivalent photon mapping AI model is used to predict the second result under the same preset dose based on the first result; the first result is the calculation result of inputting the preset dose into the first equivalent photon dose calculation model for calculation; the second result is the calculation result of inputting the preset dose into the second equivalent photon dose calculation model for calculation, the preset dose is any dose value in BNCT treatment, and the preset dose is used here to describe the input data and output data during the training or reasoning of the equivalent photon mapping AI model. When the equivalent photon mapping AI model is applied to the equivalent photon dose prediction process of the BNCT dose, the preset dose is the dose to be predicted. That is: the equivalent photon mapping AI model uses the calculation result of the first equivalent photon dose calculation model as the prediction factor, and the calculation result of the second equivalent photon dose calculation model as the prediction result. In other words, in the prediction scenario, for any dose to be predicted, the equivalent photon mapping AI model is used to predict the calculation result of the second equivalent photon dose calculation model for the dose to be predicted based on the calculation result of the first equivalent photon dose calculation model for the dose to be predicted.

[0055] Among them, the calculation speed of the first equivalent photon dose calculation model is faster than that of the second equivalent photon dose calculation model, but the error of the calculation result of the first equivalent photon dose calculation model compared with the true equivalent dose is greater than the error of the calculation result of the second equivalent photon dose calculation model compared with the true equivalent dose, that is, the first equivalent photon dose calculation model is an equivalent photon dose calculation model with fast calculation speed but low precision, and the second equivalent photon dose calculation model is an equivalent photon dose calculation model with high precision but longer calculation time. Optionally, the second equivalent photon dose calculation model includes an improved stochastic microdose kinetic model (SMK) and a nanodose effect model.

[0056] Figure 2 FIG. 1 is a schematic diagram showing the data flow of the method for predicting the equivalent photon dose of BNCT dose according to an embodiment of the present invention, referring to FIG. Figure 2 As shown, the calculation result of the first equivalent photon dose calculation model is used as input data, the calculation result of the second equivalent photon dose calculation model is used as output data, and the preset AI model is trained to obtain the equivalent photon mapping AI model. The core idea of ​​the present invention is to use the first equivalent photon dose calculation model to quickly obtain preliminary results (i.e., preliminary results), and then use the equivalent photon mapping AI model to map the relationship between the calculation results of the two calculation models, and predict the preliminary results of the first equivalent photon dose calculation model to the relatively higher precision results of the second equivalent photon dose calculation model to obtain the equivalent photon dose prediction results, thereby ensuring the prediction efficiency and improving the prediction accuracy. It should be noted that the above steps have given the implementation steps of the embodiment of the present invention. Those skilled in the art can select the first equivalent photon dose calculation model and the second equivalent photon dose calculation model that need to be mapped according to their needs. At the same time, based on the description of the prediction factors and prediction results of the equivalent photon mapping AI model of the present invention, they can select the corresponding preset AI model and collect the corresponding training data set for training and reasoning to construct the equivalent photon mapping AI model.

[0057] In a specific implementation of step S200, the present invention is implemented through a two-stage AI model. Figure 3 FIG. 1 is a schematic diagram showing a data flow process for obtaining prediction results based on a two-stage AI model in a method for predicting equivalent photon dose of BNCT dose according to an embodiment of the present invention, referring to FIG. Figure 3 As shown, using the equivalent photon mapping AI model to obtain prediction results corresponding to the preliminary results includes the following steps:

[0058] Step S201: Select a prediction sub-model corresponding to the preliminary results according to the dose breakpoint data.

[0059] The equivalent photon mapping AI model in this embodiment includes a first sub-model and a second sub-model. The first sub-model is a machine learning regression model, and the second sub-model is a neural network model. The two sub-models are not in a relationship of first execution and later execution in actual prediction, but a relationship of using different sub-models for prediction based on different preliminary results.

[0060] In a specific implementation, the first sub-model is an SVR model, and Support Vector Regression (SVR) is an extension of Support Vector Machine (SVM) and is suitable for regression tasks. The goal of SVR is to find a hyperplane so that the vast majority of data points fall within a certain tolerance error range, while allowing the model to have good generalization capabilities. The second sub-model is an MLP model, and a multi-layer perceptron (MLP) includes an input layer, a hidden layer, and an output layer. The hidden layer includes multiple hidden layers, each hidden layer includes multiple neurons, and uses nonlinear activation functions (such as ReLU, tanh, etc.) to capture complex nonlinear relationships. According to actual needs, those skilled in the art can select other suitable machine learning regression models as the first sub-model and select other neural network models as the second sub-model.

[0061] Specifically, after obtaining the preliminary result output by the first equivalent photon dose calculation model in step S100, the dose size relationship between the preliminary result and the dose breakpoint data is determined. If the preliminary result is less than the dose breakpoint data, the first sub-model is selected as the corresponding prediction sub-model. If the preliminary result is greater than or equal to the dose breakpoint data, the second sub-model is selected as the corresponding prediction sub-model. The advantage of the present invention of splitting the model into two sub-models based on the dose breakpoint data is that, for the same dose to be predicted, the difference between the calculation result of the first equivalent photon dose calculation model and the calculation result of the second equivalent photon dose calculation model has a trend of changing in a quasi-exponential distribution with the change of the dose to be predicted. If the difference is greater than zero, the difference first increases and then decreases with the increase of the dose to be predicted; if the difference is less than zero, the difference gradually decreases with the increase of the dose to be predicted. This embodiment distinguishes different sub-models by dose breakpoints, which is conducive to more accurately extracting the relationship between the calculation result of the first equivalent photon dose calculation model and the calculation result of the second equivalent photon dose calculation model, and avoiding the problem that a single model is difficult to adapt to the overall trend.

[0062] In a specific implementation of step S201 , the present invention obtains dose breakpoint data in a clustering manner. Figure 4 A schematic diagram of a data flow process for obtaining dose breakpoint data in a method for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention is shown. Figure 4 Each group (x, y) is a data sample, x is the prediction factor, y is the prediction result, refer to Figure 4 As shown, the method for obtaining dose breakpoint data includes the following steps:

[0063] S2011: Sort all data in the training data set of the equivalent photon mapping AI model by dose size to obtain a sorted data set.

[0064] Based on the description of steps S100-S200, a person skilled in the art can construct a data set of an equivalent photon mapping AI model without creative work. The data set of the equivalent photon mapping AI model includes multiple data, each of which uses the calculation result of the first equivalent photon dose calculation model as a prediction factor and the calculation result of the second equivalent photon dose calculation model as a prediction result. In this embodiment, all the data in the training data set are first sorted by dose size (here the sorting is based on the dose size of the prediction factor in the training set, but since a person skilled in the art can know based on the description of step S201 that the dose breakpoint data is compared with the calculation result of the first equivalent photon dose calculation model, no special limitation is required here), and a sorted data set is generated. The sorted data set facilitates the unsupervised clustering algorithm to identify the distribution characteristics of the data, and can reduce the interference of disordered noise on the clustering process.

[0065] S2012: Use an unsupervised clustering algorithm to divide the sorted data set into the first cluster and the second cluster.

[0066] The advantage of using an unsupervised clustering algorithm for clustering in this embodiment is that there is no need to pre-label the dose categories, and the cutoff point can be quickly determined based on the data distribution characteristics in the sorted data set to divide the sorted data set into two clusters - a first cluster and a second cluster.

[0067] S2013: Obtain the maximum dose value of the first cluster and the minimum dose value of the second cluster, and use any dose value between the maximum dose value and the minimum dose value as dose breakpoint data.

[0068] The dose breakpoint data is used as the dividing point between the first cluster and the second cluster. The first cluster is used for the training of the first sub-model, and the second cluster is used for the training of the second sub-model. The dose value of any data in the first cluster is less than the dose value of any data in the second cluster. The maximum dose value in the first cluster is selected as the left boundary of the breakpoint interval, and the minimum dose value in the second cluster is selected as the right boundary of the breakpoint interval. The breakpoint interval is formed by the left boundary and the right boundary. Any value in the breakpoint interval can be used as the dose breakpoint data. For example, refer to Figure 4 As shown, Figure 4In the example, the first cluster and the second cluster are separated from (x4, y4) and (x5, y5), that is, at this time, the dose breakpoint data is a data greater than x4 and less than x5, and the breakpoint interval is [x4, x5]. In addition to the above methods, those skilled in the art can choose other methods for obtaining dose breakpoint data, for example, manually setting a fixed dose value as dose breakpoint data. However, the advantages of the above-mentioned method for obtaining dose breakpoint data in this embodiment are that, firstly, no manual intervention is required, and the natural dividing point of the data distribution in the training data set is obtained by sorting and clustering as dose breakpoint data, which is fast and efficient; secondly, a clustering algorithm is used to obtain dose breakpoint data. With the training iteration of the entire model and the adjustment of the clustering algorithm, the dose breakpoint data obtained will also be dynamically updated. In other words, the dose breakpoint data participates in the model training process as one of the optimized parameters, and the dose breakpoint data is continuously optimized with the iteration of model training. The effect of finally obtaining the dose breakpoint data is better than that of the dose breakpoint data with a fixed dose value.

[0069] Optionally, the unsupervised clustering algorithm in the embodiment of the present invention is a K-means algorithm. The K-means algorithm divides clusters according to the distances of different sample points in the data set. Those skilled in the art can also select other suitable unsupervised clustering algorithms according to actual needs. Specifically, there are two ways to construct the data set to be clustered for the K-means algorithm in this embodiment. One is to use the sorted data set as the data set to be clustered, that is, the original data sample (x, y) as the sample to be clustered. The second is to make a difference between the prediction factor and the prediction result of each data in the sorted data set to obtain the corresponding difference, and use the prediction factor and the difference as the sample to be clustered, that is, for the original data sample (x, y), the sample to be clustered is (x, xy). Corresponding to this construction method is that after the K-means clustering is completed, based on the prediction factor in the clustered data sample, the data sample (for example (x3, x3-y3)) is restored to the corresponding original data sample (that is, (x3, y3)) as the final clustering result. Compared with the first method of constructing the data set to be clustered, the second method directly uses the difference as the feature of the sample to be predicted. The clustering process directly calculates the distance based on the prediction factor and the difference. The clustering result can better reflect the data characteristics of the first cluster and the second cluster, so the division effect is better.

[0070] Optionally, the evaluation indicators of the equivalent photon mapping AI model include: the root mean square error of the data whose dose size is within a preset range from the dose breakpoint data in the first cluster and the second cluster, the root mean square error of the first cluster and the second cluster, and the determination coefficient of the sorted data set.

[0071] Specifically, the root mean square error of the data within the preset range of the dose size distance dose breakpoint data in the first cluster and the second cluster refers to the root mean square error between the predicted value and the actual value of the data set formed by the data within the preset range of the dose breakpoint data in the first cluster and the data within the preset range of the dose breakpoint data in the first cluster, that is, the breakpoint area RMSE. Please refer to the following formula for details:

[0072]

[0073] Among them, RMSE near The root mean square error of the dose size in the first cluster and the second cluster from the dose breakpoint data within the preset range, N near Indicates the number of data points in the preset range, y i represents the ith actual value, Represents the i-th predicted value. The data near the breakpoint is usually at the junction of two clusters. The prediction accuracy of the model may be greatly challenged here. Evaluating the RMSE near the breakpoint alone can detect the performance of the model in this breakpoint area. Using the RMSE in the breakpoint area as an evaluation indicator to iterate the training of the entire model is conducive to reducing the jump errors of different clusters.

[0074] The root mean square error of the first cluster and the second cluster refers to the root mean square error between the predicted value and the actual value of the first cluster and the second cluster as a whole (that is, the entire data set). For details, refer to the following formula:

[0075]

[0076] Among them, RMSE combined represents the root mean square error between the first cluster and the second cluster, N1 represents the number of data points in the first cluster, N2 represents the number of data points in the second cluster, Cluster1∪Cluster2 represents the overall data of the first cluster and the second cluster, y i represents the ith actual value, Represents the i-th predicted value. The root mean square error of the first cluster and the second cluster reflects the prediction error of the model for the entire data set. Iterating the model training based on this evaluation indicator is beneficial to improving the global performance.

[0077] The determination coefficient of the sorted data set refers to the determination coefficient between the predicted value and the actual value calculated for the entire sorted data set. For details, refer to the following formula:

[0078]

[0079] Among them, R 2 Represents the coefficient of determination of the sorted data set, SS res Represents the residual sum of squares of the sorted data set, SS totRepresents the total sum of squares of the sorted data set. The closer the coefficient of determination of the sorted data set is to 1, the better the fitting effect is. Using this indicator as an evaluation indicator to iterate the model training can ensure that the model performs consistently and stably on global data.

[0080] The above three evaluation indicators can be integrated together using the following formula in specific applications:

[0081]

[0082] Among them, Weighted_RMSE represents the overall evaluation index of the equivalent photon mapping AI model, w near 、w combined , All are weight coefficients. Those skilled in the art can set different weight coefficients according to actual needs, and the embodiments of the present invention do not specifically limit this. Through the setting of the above evaluation indicators, the key evaluation areas and global performance of the equivalent photon mapping AI model can be taken into comprehensive consideration, and the training model can effectively improve the model performance.

[0083] Optionally, the dose to be predicted includes the boron dose, hydrogen dose, nitrogen dose and photon dose of the preset cells, and the preset cells are tumor cells, skin cells, mucosal cells or neural tissue cells. At this time, the present invention sets corresponding equivalent photon mapping AI models for 4 different preset cell types, that is, there are at least 4 equivalent photon mapping AI models. Further, for any preset cell type, different combinations of first equivalent photon dose calculation models and second equivalent photon dose calculation models can be used to establish multiple equivalent photon mapping AI models for different scenarios. For example, for skin cells, two equivalent photon mapping AI models can be established-the first equivalent photon mapping AI model and the second equivalent photon mapping AI model. These two equivalent photon mapping AI models use different first equivalent photon dose calculation models or second equivalent photon dose calculation models. Those skilled in the art can set other dose combinations and cell types according to actual needs.

[0084] Step S202: Using the preliminary result as input data, using the corresponding prediction sub-model to obtain the output result as the prediction result of the equivalent photon mapping AI model.

[0085] Specifically, the preliminary results are used as input data and input into the corresponding prediction sub-model, and the output data of the prediction sub-model is obtained as the output data of the equivalent photon mapping AI model, that is, the prediction result. When constructing the equivalent photon mapping AI model in the embodiment of the present invention, by considering the characteristics of different dose intervals, different algorithm models are selected as prediction sub-models in a targeted manner, and the machine learning regression model is used to capture the relatively simple nonlinear relationship of the low dose interval (the dose interval less than the dose breakpoint data), and the neural network model is used to fit the relatively complex linear relationship of the high dose interval (the dose interval greater than the dose breakpoint data). The difficulty of fitting a single model to the entire data range is reduced through the two-stage division of sub-models. Therefore, the prediction performance of different dose intervals can be more targeted during model training, and the overall accuracy of the equivalent photon mapping AI model is higher. Furthermore, the input data also includes the dose to be predicted. This setting method can reasonably increase the number of features and information of the prediction factors, which is conducive to improving the accuracy of the equivalent photon mapping AI model.

[0086] Optionally, the input data also includes supplementary features, which are features obtained by performing polynomial expansion based on preliminary results. This embodiment improves the predictive ability of the equivalent photon mapping AI model through supplementary features. Specifically, based on the preliminary results, a series of new features are generated by polynomial expansion to provide additional information dimensions for the equivalent photon mapping AI model, helping to capture more complex nonlinear relationships that may exist in the original data. Polynomial expansion is a feature engineering technology that obtains more new features by generating high-order features of the original features and cross-features of features of different dimensions in the original features. The new features can be used as supplementary features after being standardized. Supplementary features are an enhancement of the input data through polynomial expansion, which can effectively improve the performance of the equivalent photon mapping AI model.

[0087] In a specific embodiment, the present invention directly uses a single tissue material to obtain a training data set. Specifically, the method for obtaining a training data set for an equivalent photon mapping AI model includes the following steps:

[0088] Step S1: setting a neutron source item identical to that of the BNCT device in the simulation software, setting a tissue region of a preset size at the front end of the source item, and setting the boron concentration in the tissue region to a preset concentration.

[0089] A neutron source term is set in the simulation software, the source term is the same as the actual neutron source in the BNCT device, the tissue region is located at the front section of the neutron source term, the size of the tissue region is a preset size, and the boron concentration in the tissue region is a preset concentration. The simulation software used in this embodiment is a Monte Carlo simulation software. Optionally, the Monte Carlo simulation software includes Geant4, MCNP, and FLUCK.

[0090] Step S2: Simulating particle transport and energy deposition in all grids of the tissue region using materials of different preset tissues to obtain corresponding deposition dose results.

[0091] The materials of the tissue area are set to different preset tissues (such as tumors, skin, mucosa, nerve tissue), and the materials of other locations outside the tissue area are set to air. Each time, only a single preset tissue material is used to simulate particle transport and energy deposition in all grids of the tissue area to obtain the deposition dose results (such as boron dose, hydrogen dose, nitrogen dose and photon dose) in each grid of the corresponding material. Optionally, the photon dose in each grid deposition dose result needs to be in the range of 0-30Gy, the boron dose, hydrogen dose and nitrogen dose all belong to the range of 0-40Gy, and the total dose of each grid deposition dose result belongs to the range of 0-70Gy.

[0092] Step S3: Input the deposition dose result into the first equivalent photon dose calculation model to obtain the corresponding calculation result as a prediction factor, input the deposition dose result into the second equivalent photon dose calculation model to obtain the corresponding calculation result as a prediction result, and combine the prediction factor and the corresponding prediction result to obtain a training data set.

[0093] Specifically, for any deposition dose result data, the deposition dose result is input into the first equivalent photon dose calculation model to obtain the first result, and the deposition dose result is input into the second equivalent photon dose calculation model to obtain the second result. The first result is used as a prediction factor and the second result is used as a prediction result to construct a training data corresponding to the deposition dose result. The training data corresponding to all deposition dose results are collected as a training data set.

[0094] Steps S1-S3 directly use single tissue materials to efficiently generate a training data set for the equivalent photon mapping AI model, providing a high-quality data foundation for the training of the equivalent photon mapping AI model.

[0095] In the prior art, the training data set is generally divided into a training set, a test set, and a validation set to train the AI ​​model. In a specific implementation of steps S1 to S3, the data in the test set is obtained by replacing the preset tissue area with the Snyder ellipsoid model, and then obtaining the corresponding deposition dose results based on multiple changes in the boron concentration or the angle of the emitted particles and the irradiation direction, and obtaining the corresponding data set as the test set based on the deposition dose result data using step S3. The advantage of the test set obtained using the above method is that it introduces more complex and diverse scenarios, which can improve the representativeness of the test set data and help improve the robustness of the model.

[0096] The protection scope of the method for predicting the equivalent photon dose of the BNCT dose in the embodiment of the present invention is not limited to the execution order of the steps listed in this embodiment. All solutions implemented by adding, reducing or replacing steps in the prior art based on the principles of the present invention are included in the protection scope of the present invention.

[0097] The equivalent photon dose prediction method of the BNCT dose in the embodiment of the present invention uses the first equivalent photon dose calculation model to quickly obtain preliminary results, and then uses the equivalent photon mapping AI model to map the relationship between the calculation results of the two calculation models to obtain the equivalent photon dose prediction results, thereby ensuring the prediction efficiency and improving the prediction accuracy. In addition, the difficulty of fitting a single model to the entire data range is reduced through the two-stage sub-model division of labor, and the prediction performance of different dose intervals can be more targeted during model training, thereby improving the overall accuracy.

[0098] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides an equivalent photon dose prediction device for BNCT dose.

[0099] Figure 5 FIG. 1 is a schematic diagram showing a device for predicting an equivalent photon dose of a BNCT dose according to an embodiment of the present invention, referring to FIG. Figure 5 As shown, the equivalent photon dose prediction device of the BNCT dose according to the embodiment of the present invention includes a calculation module and a reasoning module.

[0100] A calculation module, used for inputting the dose to be predicted into a first equivalent photon dose calculation model to obtain a corresponding calculation result as a preliminary result;

[0101] An inference module, used for obtaining a prediction result corresponding to the preliminary result by using an equivalent photon mapping AI model as an equivalent photon dose prediction result;

[0102] The equivalent photon mapping AI model is used to predict the second result at the same preset dose based on the first result;

[0103] The first result is a calculation result of inputting the preset dose into the first equivalent photon dose calculation model for calculation;

[0104] The second result is a calculation result of inputting the preset dose into the second equivalent photon dose calculation model for calculation;

[0105] Among them, for any dose to be predicted, the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model.

[0106] The equivalent photon dose prediction device of the BNCT dose of the embodiment of the present invention uses the first equivalent photon dose calculation model to quickly obtain preliminary results (i.e., preliminary results), and then uses the equivalent photon mapping AI model to map the relationship between the calculation results of the two calculation models to obtain the equivalent photon dose prediction results, thereby ensuring the prediction efficiency and improving the prediction accuracy. In addition, the difficulty of fitting a single model to the entire data range is reduced through the two-stage sub-model division of labor, and the prediction performance of different dose intervals can be more specifically optimized during model training, thereby improving the overall accuracy.

[0107] In order to solve the above-mentioned technical problems existing in the prior art, an embodiment of the present invention further provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, all steps of the equivalent photon dose prediction method of the BNCT dose of the embodiment are implemented.

[0108] The specific steps of the method for predicting the equivalent photon dose of BNCT dose and the beneficial effects obtained by applying the readable storage medium provided in the embodiment of the present invention are the same as those in the above embodiment and will not be described in detail here.

[0109] A person of ordinary skill in the art can understand that all or part of the steps in the method for implementing the above-mentioned embodiment can be completed by instructing a processor through a program, and the program can be stored in a computer-readable storage medium, and the storage medium is a non-transitory medium, such as a random access memory, a read-only memory, a flash memory, a hard disk, a solid-state hard disk, a magnetic tape, a floppy disk, an optical disc, and any combination thereof. The above-mentioned storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a digital video disc (DVD)), or a semiconductor medium (for example, a solid-state disk (SSD)), etc.

[0110] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides a terminal. Figure 6 A schematic diagram of the structure of the terminal according to an embodiment of the present invention is shown, referring to Figure 6 As shown, the terminal of the embodiment of the present invention includes a processor and a memory, and the memory is communicatively connected to the processor; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal executes all steps of the equivalent photon dose prediction method of the BNCT dose in the above embodiment.

[0111] The specific steps of the method for predicting the equivalent photon dose of BNCT dose and the beneficial effects obtained by applying the terminal provided by the embodiment of the present invention are the same as those in the above embodiment and will not be described in detail here.

[0112] It should be noted that the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. Similarly, the processor may also be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), 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.

[0113] Although the embodiments disclosed in the present invention are as above, the contents described are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for predicting equivalent photon dose of BNCT dose, comprising: Inputting the dose to be predicted into the first equivalent photon dose calculation model to obtain a corresponding calculation result as a preliminary result; Using an equivalent photon mapping AI model to obtain a prediction result corresponding to the preliminary result as an equivalent photon dose prediction result; The equivalent photon mapping AI model is used to predict a second result at the same preset dose based on the first result; The first result is a calculation result of inputting the preset dose into a first equivalent photon dose calculation model for calculation; The second result is a calculation result of inputting the preset dose into a second equivalent photon dose calculation model for calculation; Among them, the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model.

2. The method for predicting equivalent photon dose of BNCT dose according to claim 1, characterized in that: The prediction results corresponding to the preliminary results obtained by using the equivalent photon mapping AI model include: Selecting a prediction sub-model corresponding to the preliminary result according to the dose breakpoint data, if the preliminary result is less than the dose breakpoint data, the corresponding prediction sub-model is the first sub-model, otherwise the corresponding prediction sub-model is the second sub-model; Using the preliminary results as input data, using the corresponding prediction sub-model to obtain output results as prediction results of the equivalent photon mapping AI model; Among them, the equivalent photon mapping AI model includes a first sub-model and a second sub-model, the first sub-model is a machine learning regression model, and the second sub-model is a neural network model.

3. The method for predicting the equivalent photon dose of BNCT dose according to claim 2, characterized in that: The input data further includes a supplementary feature, which is a feature obtained by performing a polynomial expansion based on the preliminary result.

4. The method for predicting equivalent photon dose of BNCT dose according to claim 2, characterized in that: The method for obtaining the dosage breakpoint data includes: Sorting all data in the training data set of the equivalent photon mapping AI model according to dose size to obtain a sorted data set; Using an unsupervised clustering algorithm to divide the sorted data set into a first cluster and a second cluster; A maximum dose value of the first cluster and a minimum dose value of the second cluster are obtained, and any dose value between the maximum dose value and the minimum dose value is used as dose breakpoint data.

5. The method for predicting equivalent photon dose of BNCT dose according to claim 4, characterized in that: The evaluation indicators of the equivalent photon mapping AI model include: The root mean square error of the data whose dose sizes in the first cluster and the second cluster are within a preset range from the dose breakpoint data, the root mean square error of the first cluster and the second cluster, and the determination coefficient of the sorted data set.

6. The method for predicting equivalent photon dose of BNCT dose according to claim 1, characterized in that: The method for obtaining the training data set of the equivalent photon mapping AI model includes: Setting a neutron source item identical to that of the BNCT device in the simulation software, setting a tissue region of a preset size at the front end of the source item, and setting a boron concentration in the tissue region to a preset concentration; Using materials of different preset tissues to simulate particle transport and energy deposition in all grids of the tissue region, respectively, to obtain corresponding deposition dose results; The deposition dose result is input into a first equivalent photon dose calculation model to obtain a corresponding calculation result as a prediction factor, the deposition dose result is input into a second equivalent photon dose calculation model to obtain a corresponding calculation result as a prediction result, and the prediction factor and the corresponding prediction result are combined to obtain a training data set.

7. The method for predicting equivalent photon dose of BNCT dose according to claim 1, characterized in that: The dose to be predicted includes boron dose, hydrogen dose, nitrogen dose and photon dose of preset cells, and the preset cells are tumor cells, skin cells, mucosal cells or nerve tissue cells.

8. A device for predicting equivalent photon dose of BNCT dose, characterized in that: include: A calculation module, used for inputting the dose to be predicted into a first equivalent photon dose calculation model to obtain a corresponding calculation result as a preliminary result; An inference module, used for obtaining a prediction result corresponding to the preliminary result by using an equivalent photon mapping AI model as an equivalent photon dose prediction result; The equivalent photon mapping AI model is used to predict a second result at the same preset dose based on the first result; The first result is a calculation result of inputting the preset dose into a first equivalent photon dose calculation model for calculation; The second result is a calculation result of inputting the preset dose into a second equivalent photon dose calculation model for calculation; Among them, for any dose to be predicted, the calculation speed of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model, and the error of the first equivalent photon dose calculation model is greater than that of the second equivalent photon dose calculation model.

9. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for predicting the equivalent photon dose of the BNCT dose according to any one of claims 1 to 7 is implemented.

10. A terminal, characterized in that: It comprises a processor and a memory, wherein the memory is communicatively connected to the processor; the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal executes the equivalent photon dose prediction method of the BNCT dose according to any one of claims 1 to 7.

Images