BNCT important clinical parameter prediction method based on BP neural network

Abstract

The invention provides a BNCT important clinical parameter prediction method based on a BP neural network. The method comprises the steps of obtaining neutron energy spectrum and dose distribution data; constructing a multi-layer BP neural network model and training the multi-layer BP neural network model; selecting a treatment position of an outlet of a beam shaper of the BNCT as a point to be measured, and measuring a neutron energy spectrum and neutron distribution at the measurement point; using the trained multilayer BP neural network model to predict distribution of boron dose, hydrogen dose, nitrogen dose and gamma dose in tumors and normal tissues in a human body corresponding to neutron energy spectrum and neutron distribution, and calculating related clinical parameters. According to the method, the neutron source for the BNCT can be rapidly and accurately evaluated, uncertainty and risks in treatment can be reduced through accurate parameter prediction, and the safety of the BNCT is improved; reliable parameter prediction is provided for clinicians, the accuracy and efficiency of clinical decision making are improved, the treatment process is optimized, and therefore the overall medical cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of BNCT parameter prediction, and in particular to a method for predicting important clinical parameters of BNCT based on a BP neural network. Background Art

[0002] Boron seed capture therapy (BNCT) has been proposed as a potential cancer treatment. It uses targeted drugs containing 10B to accumulate in cancer cells through injection or oral administration, and then generates 7Li and α particles through neutron beam irradiation and reaction. They have high linear energy transfer and high relative bioavailability, which can cause lethal damage to cells; and their penetration ability is limited to about 10μm, which is approximately the diameter of a cell, thus limiting the destructive force to the single cell scale, thereby avoiding damage to nearby normal tissue cells.

[0003] Since BNCT involves multiple nuclear reactions between neutrons and matter, the physical process is relatively complex and the secondary particle dose cannot be ignored. The analytical algorithm used in TPS for traditional radiotherapy such as photon radiotherapy can meet the speed requirements, but cannot meet the clinical requirements for calculation accuracy. Although the method based on artificial intelligence for dose prediction is fast, the results are not interpretable enough and are not currently suitable for clinical use. Therefore, the existing TPS for BNCT at home and abroad generally uses the Monte Carlo method, which is known as the "gold standard" for dose calculation, to calculate the dose of BNCT. However, the disadvantage of general Monte Carlo software is that the calculation speed is too slow. It may take an entire night to complete a dose calculation, which is far from meeting the clinical requirements for dose calculation speed.

[0004] Most of the existing BNCT TPSs are connected to general Monte Carlo programs such as PHITS and MCNP for dose calculation. The disadvantage is that the calculation time is too long, and it takes several hours to complete a dose calculation. This cannot meet clinical needs and limits the further promotion of BNCT. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a method for predicting important clinical parameters of BNCT based on a BP neural network.

[0006] The technical solution of the present invention is: a method for predicting important clinical parameters of BNCT based on BP neural network, comprising the following steps:

[0007] S1), obtaining neutron energy spectrum and dose distribution data;

[0008] S2), construct a multi-layer BP neural network model and train it;

[0009] S3), selecting the treatment point at the exit of the BNCT beam shaper as the point to be measured, and measuring the neutron energy spectrum and neutron distribution at the measurement point;

[0010] S4), using the multi-layer BP neural network model trained in step S2), predicting the distribution of boron dose, hydrogen dose, nitrogen dose and gamma dose in tumors and normal tissues in the human body corresponding to the neutron energy spectrum and neutron distribution measured in step S3), and calculating relevant clinical parameters;

[0011] Preferably, in step S1), Monte Carlo calculation software is used to generate dose distribution data including a neutron energy spectrum and a corresponding Snyder phantom.

[0012] Preferably, in step S2), the input of the BP neural network model is a multi-group neutron energy spectrum, and the output is a relevant dose distribution curve and clinical parameters in the human body model.

[0013] Preferably, in step S2), the prediction structure of the BP neural network model includes boron dose, hydrogen dose, nitrogen dose and gamma dose in tumor and normal tissue.

[0014] Preferably, in step S2), during the training process of the BP neural network model, the mean square error MSE is used as the loss function, that is:

[0015]

[0016] Where D i represents the dose obtained by Monte Carlo simulation at position i; represents the predicted dose at position i.

[0017] Preferably, in step S2), during the training of the BP neural network model, ReLU is used as the activation function, that is:

[0018] ReLU(x)=max(0,x);

[0019] In the formula, x represents the input value;

[0020] Preferably, in step S2), the BP neural network model is trained using stochastic gradient descent, that is:

[0021]

[0022] In the formula, θ new is the updated parameter, θ old is the current parameter; η is the learning rate; is the gradient of the loss function J;

[0023] If the mean square error MSE is less than the specified error ε or the maximum number of iterations R is reached, the training is completed.

[0024] Preferably, in step S4), the clinical parameters include dominant depth AD, double dose depth DDD, triple dose depth TDD, and absorbed dose ratio AR.

[0025] The beneficial effects of the present invention are as follows: the present invention can quickly and accurately evaluate the neutron source for BNCT, and accurate parameter prediction can reduce the uncertainty and risk in treatment and improve the safety of BNCT; it provides reliable parameter prediction for clinicians, helps to improve the accuracy and efficiency of clinical decision-making, optimize the treatment process, and thus reduce overall medical costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of a framework of a method according to an embodiment of the present invention;

[0027] Figure 2 This is a comparison diagram of the predicted dose distribution based on the BP neural network and the actual dose distribution in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings:

[0029] like Figure 1 As shown, this embodiment provides a method for predicting important clinical parameters of BNCT based on BP neural network, comprising the following steps:

[0030] S1), obtaining dose distribution data;

[0031] In this embodiment, Monte Carlo calculation software is used to generate dose distribution data including a neutron energy spectrum and a corresponding Snyder phantom.

[0032] S2), construct a multi-layer BP neural network model and train it;

[0033] As preferred in this embodiment, the input of the BP neural network model is a multi-group neutron energy spectrum, and the output is a relevant dose distribution curve and clinical parameters in the human body model. The prediction structure of the BP neural network model includes boron dose, hydrogen dose, nitrogen dose and gamma dose in tumor and normal tissue.

[0034] In this embodiment, during the training process of the BP neural network model, the mean square error MSE is used as the loss function, that is:

[0035]

[0036] Where D irepresents the dose obtained by Monte Carlo simulation at position i; represents the predicted dose at position i.

[0037] For example, taking the boron dose in the tumor as an example, MSE calculates the average of the squares of the differences between the predicted boron dose in the tumor and the actual value, that is:

[0038]

[0039] Where D i,B,tumor represents the boron dose in the tumor obtained by Monte Carlo simulation at position i; represents the predicted intratumoral boron dose at position i.

[0040] In the BP neural network model training process, ReLU is used as the activation function, that is:

[0041] ReLU(x)=max(0,x);

[0042] In the formula, x represents the input value;

[0043] Preferably, in step S2), the BP neural network model is trained using stochastic gradient descent, that is:

[0044]

[0045] In the formula, θ new is the updated parameter, θ old is the current parameter; η is the learning rate; is the gradient of the loss function J;

[0046] If the mean square error MSE is less than the specified error ε or the maximum number of iterations R is reached, the training is completed.

[0047] S3), selecting the treatment point at the exit of the BNCT beam shaper as the point to be measured, and measuring the neutron energy spectrum and neutron distribution at the measurement point;

[0048] S4), using the multi-layer BP neural network model trained in step S2), predicting the distribution of boron dose, hydrogen dose, nitrogen dose and gamma dose in tumors and normal tissues in the human body corresponding to the neutron energy spectrum and neutron distribution measured in step S3), and calculating related clinical parameters; the clinical parameters include advantage depth AD, double dose depth DDD, triple dose depth TDD, and absorbed dose ratio AR.

[0049] In this embodiment, according to the multi-group energy spectrum shown in Table 1, the energy spectrum is converted into 50 multi-group energy spectra and normalized. The value of each energy interval is expressed as [x1, x2, ..., x 50], which constitutes the parameters of the input layer; the output layer corresponds to the dose distribution. The predicted distribution interval is the part of the brain in the Snyder model, which is about 14.8 cm long and is divided into a total of 37 intervals [y1,y2,…,y 37 As for the number of hidden layers, this embodiment selects a single layer of 100 neurons.

[0050] Table 150 Group Multi-Group Neutron Spectra

[0051]

[0052] like Figure 2 As shown, the predicted values ​​of this embodiment are consistent with the actual values. According to the predicted results, it can be calculated that AD is 8.7 cm, DDD is 6.8 cm, TDD is 5.8 cm, and AR is 4.95.

[0053] The above embodiments and descriptions are only for illustrating the principles and best embodiments of the present invention. Without departing from the spirit and scope of the present invention, the present invention may be subject to various changes and improvements, all of which fall within the scope of the present invention to be protected.

Claims

1. A method for predicting important clinical parameters of BNCT based on BP neural network, characterized in that The following steps are involved: S1), obtaining neutron energy spectrum and dose distribution data; S2), construct a multi-layer BP neural network model and train it; S3), selecting the treatment point at the exit of the BNCT beam shaper as the point to be measured, and measuring the neutron energy spectrum and neutron distribution at the measurement point; S4), using the multi-layer BP neural network model trained in step S2), predicting the distribution of boron dose, hydrogen dose, nitrogen dose and gamma dose in tumor and normal tissue of the human body corresponding to the neutron energy spectrum and neutron distribution measured in step S3), and calculating the relevant clinical parameters.

2. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 1, characterized in that: In step S1), Monte Carlo calculation software is used to generate dose distribution data including neutron energy spectrum and corresponding Snyder phantom.

3. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 1, characterized in that: In step S2), the input of the BP neural network model is a multi-group neutron energy spectrum, and the output is a relevant dose distribution curve and clinical parameters in the human body model.

4. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 3, characterized in that: In step S2), the prediction structure of the BP neural network model includes boron dose, hydrogen dose, nitrogen dose and gamma dose in tumor and normal tissue.

5. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 4, characterized in that: In step S2), during the BP neural network model training process, the mean square error MSE is used as the loss function, that is: Where D i represents the dose obtained by Monte Carlo simulation at position i; represents the predicted dose at position i.

6. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 5, characterized in that: In step S2), during the training of the BP neural network model, ReLU is used as the activation function, that is: ReLU(x)=max(0,x); In the formula, x represents the input value.

7. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 6, characterized in that: In step S2), the BP neural network model is trained using stochastic gradient descent, namely: In the formula, θ new is the updated parameter, θ old is the current parameter; η is the learning rate; is the gradient of the loss function J; If the mean square error MSE is less than the specified error ε or the maximum number of iterations R is reached, the training is completed.

8. The method for predicting important clinical parameters of BNCT based on BP neural network according to claim 1, characterized in that: In step S4), the clinical parameters include dominant depth AD, double dose depth DDD, triple dose depth TDD, and absorbed dose ratio AR.

Images