A method for determining the neutron energy spectrum in the emission window based on energy spectrum superposition and machine learning algorithms.

Abstract

This invention discloses a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms, involving the interdisciplinary field of nuclear technology and artificial intelligence. The method includes: predicting the target neutron energy spectrum data generated after target firing when the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed; multiplying the proton fluence ratio by the target neutron energy spectrum data corresponding to the proton, and summing the multiplication results of all energy points and the protons corresponding to the incident direction point to obtain the target neutron energy spectrum; predicting the energy spectrum data when the energy spectrum data corresponding to the neutron energy point at the direction point is not indexed; and substituting the energy spectrum data corresponding to the neutron energy point at the direction point indexed in the BSA database and the predicted energy spectrum data of the neutron energy point at the direction point into the emission window energy spectrum formula for calculation to obtain the emission window neutron energy spectrum. This method improves the accuracy of determining the emission window neutron energy spectrum.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms. Background Technology

[0002] Boron neutron capture therapy (BNCT) is a binary targeted radiotherapy technique and a core component of next-generation radiotherapy research. Its basic principle is to inject patients with a high concentration of boron neutrons... 10 Boron-containing drugs (type B) accumulate in tumor cells through metabolism, with limited distribution in other tissues. These drugs are non-toxic and harmless to the human body and have no effect on cancer cells. The boron-containing drugs are bombarded within the tumor using a low-energy neutron (hyperthermal or thermal neutron) beam, triggering a nuclear reaction. 10 B(n, α) 7 Li releases energy, and the released α and Li both have a range of about 10 μm in vivo, killing tumors at the cellular scale while protecting normal tissue cells to the greatest extent.

[0003] Traditional Energy Spectrum Superposition Method (ESSM) indexes a database to determine the neutron spectrum of the output window. However, when certain data is missing from the database, the accuracy of the calculated neutron spectrum drops significantly. For example, during radiotherapy, image guidance allows for irradiation from different directions to ensure the neutron beam covers the entire tumor, maximizing tumor cell destruction. However, with missing database data, it's impossible to flexibly and accurately adjust the required neutron spectrum based on image guidance, thus affecting the final treatment outcome and even leading to tumor recurrence. Summary of the Invention

[0004] Therefore, it is necessary to provide a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms to address the aforementioned technical problems. This method improves the accuracy of determining the neutron energy spectrum of the emission window.

[0005] The present invention adopts the following technical solution:

[0006] This invention provides a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms, comprising:

[0007] Based on all proton energy points and incident direction points in the accelerator proton energy spectrum, indexing is performed in the target neutron database;

[0008] When the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed, the target neutron energy spectrum data generated after hitting the target at the incident direction point is predicted by machine learning algorithm.

[0009] For each energy point in the accelerator proton energy spectrum data, the proton fluence ratio is multiplied by the target neutron energy spectrum data corresponding to the proton, and the results of multiplying all energy points and the protons corresponding to the incident direction points are summed to obtain the target neutron energy spectrum.

[0010] For each neutron energy point and direction point in the target neutron energy spectrum, indexing is performed in the BSA database. When the energy spectrum data corresponding to the neutron energy point at the direction point is not indexed, the energy spectrum data corresponding to the unindexed neutron energy point at the direction point is predicted by machine learning algorithm.

[0011] The neutron energy spectrum of the emission window is determined based on the energy spectrum data of the neutron energy points at the direction points indexed in the BSA database and the energy spectrum data of the neutron energy points at the direction points predicted by the machine learning algorithm.

[0012] Preferably, the target neutron energy spectrum data generated after the unindexed proton energy points are hit at the incident direction point using machine learning algorithms specifically includes:

[0013] Predict the normalized neutron spectrum of protons at the incident direction point by using a deep neural network for protons whose energy points are not indexed.

[0014] Predict the total neutron fluence rate at the incident direction point for proton energy points that are not indexed using the random forest algorithm;

[0015] By inversely normalizing the standardized neutron spectrum using the total neutron flux rate, the target neutron spectrum data generated after hitting the target at the incident direction point from the unindexed proton energy point is obtained.

[0016] Preferably, the energy spectrum data corresponding to the unindexed neutron energy points at the directional points is predicted using machine learning algorithms, specifically including:

[0017] Predict normalized energy spectrum data at directional points for neutron energy points that are not indexed to the index using deep neural networks;

[0018] Predict the total neutron flux rate at the directional point for neutron energy points that are not indexed using the random forest algorithm;

[0019] By inversely normalizing the standardized energy spectrum data using the total neutron flux rate, the energy spectrum data of the unindexed neutron energy points at the directional points are obtained.

[0020] Preferably, the deep neural network includes fully connected layers N1, N2, and N3 connected in series. nA residual network, a fully connected layer N2, and an activation function.

[0021] Preferably, the energy spectrum data of the neutron energy point includes the total neutron fluence; the formula for calculating the emission window energy spectrum is:

[0022] ;

[0023] in, It is the neutron energy spectrum of the emission window. The total target neutron fluence rate, For energy Direction is Total neutron flux rate, For direction Energy is Total neutron flux rate, It is energy Direction is neutron loss rate, For correction items, This is the upper limit of the summation of energy. Let the summation in the direction be the upper limit. This is a correction term for energy. For the direction correction term, This is the upper limit of the summation of the energy correction terms. The upper limit of the summation of the direction correction terms.

[0024] Preferably, the process of constructing the target neutron database specifically includes:

[0025] Within a preset proton energy range and a preset proton direction range, discretization is performed according to energy value and direction to obtain different proton energy points and different proton incident direction points;

[0026] Record the target neutron energy spectrum data of protons at different proton energy points and at different proton incident directions after hitting the target;

[0027] A target neutron database is constructed based on the target neutron energy spectrum data generated after protons hit the target at different proton energy points, different proton incident directions, and different proton energy points at different proton incident directions.

[0028] Preferably, the process of building the BSA database specifically includes:

[0029] The neutrons in the target neutron database are discretized with preset neutron angle ranges and preset neutron energy ranges to obtain different neutron energy points and different neutron direction points;

[0030] Record the energy spectrum data of neutrons at different neutron energy points and at different neutron direction points after passing through BSA;

[0031] A BSA database is constructed based on the energy spectrum data corresponding to different neutron energy points, different neutron direction points, and different neutron energy points at different neutron direction points.

[0032] This invention provides a device for determining the neutron energy spectrum of an emission window based on energy spectrum superposition and machine learning algorithms, comprising:

[0033] The indexing module is used to index the target neutron database based on all proton energy points and incident direction points in the accelerator proton energy spectrum.

[0034] The first prediction module is used to predict the target neutron energy spectrum data generated after hitting the target at the incident direction point when the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed, by using a machine learning algorithm.

[0035] The determination module is used to multiply the proton fluence ratio of each energy point in the accelerator proton energy spectrum data with the target neutron energy spectrum data corresponding to the proton at the incident direction point, and sum the multiplication results of all energy points and the protons corresponding to the incident direction points to obtain the target neutron energy spectrum.

[0036] The second prediction module is used to index each neutron energy point and direction point in the target neutron energy spectrum in the BSA database. When no energy spectrum data corresponding to the neutron energy point at the direction point is found, the energy spectrum data corresponding to the unindexed neutron energy point at the direction point is predicted by machine learning algorithm.

[0037] The calculation module is used to determine the neutron energy spectrum of the emission window based on the energy spectrum data of the neutron energy points at the direction points indexed in the BSA database and the energy spectrum data of the neutron energy points at the direction points predicted by the machine learning algorithm.

[0038] The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms.

[0039] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithm.

[0040] The above-mentioned at least one technical solution adopted in this invention can achieve the following beneficial effects:

[0041] Based on all proton energy values ​​and incident directions in the accelerator proton energy spectrum, an index is created in the target neutron database. When the target neutron energy spectrum data corresponding to a proton energy point at the incident direction point is not indexed, a machine learning algorithm is used to predict the target neutron energy spectrum data generated after the target is fired for the unindexed proton energy point and incident direction point, thus solving the error problem caused by data missing in traditional methods. For each proton corresponding to an energy point and incident direction point in the accelerator proton energy spectrum data, the proton fluence ratio is multiplied by the target neutron energy spectrum data corresponding to the proton, and the mass flux ratios of all energy points and incident direction points are multiplied. The results of multiplying the protons are summed to obtain the target neutron energy spectrum. A weighted superposition algorithm based on the proton fluence ratio effectively restores the actual neutron yield distribution characteristics of multi-energy proton beam target bombardment. For each neutron energy point and direction point in the target neutron energy spectrum, when the energy spectrum data corresponding to the neutron energy point at the direction point is not indexed, the energy spectrum data corresponding to the unindexed neutron energy point and direction point is predicted using a machine learning algorithm. The energy spectrum data indexed from the BSA database and the predicted energy spectrum data of the neutron energy points are substituted into the exit window energy spectrum formula for calculation to obtain the exit window neutron energy spectrum. This method improves the accuracy of determining the exit window neutron energy spectrum. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0043] Figure 1 A schematic diagram of the process for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithm provided by the present invention;

[0044] Figure 2 A schematic diagram of the target and BSA structure provided by the present invention;

[0045] Figure 3 This is a schematic diagram of the BNN network provided by the present invention;

[0046] Figure 4 Flowchart of the method for determining the neutron energy spectrum of the emission window provided by the present invention;

[0047] Figure 5 A comparison of neutron energy spectra generated by the conventional method and the ESSM-ML method provided by this invention;

[0048] Figure 6 A schematic diagram of a device for determining the neutron energy spectrum of an emission window based on energy spectrum superposition and machine learning algorithm provided by the present invention;

[0049] Figure 7This is a schematic diagram of a computer device for implementing a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms, as provided by the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0051] Devices such as desktop computers, servers, and laptops are capable of executing the present invention. For ease of explanation, the following description will focus on servers as the executing entity.

[0052] The Energy Spectrum Superposition Method (ESSM) can quickly obtain the neutron energy spectrum of the exit window. The neutron energy spectrum of the exit window obtained from the accelerator proton energy spectrum is shown in Equation (1): (1);

[0053] in, It is the neutron energy spectrum of the emission window. The total target neutron fluence rate, For energy Direction is Total neutron flux rate, For direction Energy is Total neutron flux rate, It is energy Direction is neutron loss rate, For correction items, This is the upper limit of the summation of energy. Let the summation in the direction be the upper limit. This is a correction term for energy. For the direction correction term, This is the upper limit of the summation of the energy correction terms. The upper limit of the summation of the direction correction terms.

[0054] Figure 1 This is a schematic diagram of a method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithms in this invention, which specifically includes the following steps:

[0055] S101: Indexing the target neutron database based on all proton energy points and incident direction points in the accelerator proton energy spectrum.

[0056] In an exemplary embodiment, the process of constructing the target neutron database specifically includes: discretizing the proton energy range and the proton direction range according to the energy value and direction to obtain different proton energy points and different proton incident direction points; recording the target neutron energy spectrum data generated after the protons corresponding to different proton energy points at different proton incident direction points hit the target; and constructing the target neutron database based on the target neutron energy spectrum data generated after the protons corresponding to different proton energy points at different proton incident direction points hit the target.

[0057] Specifically, the preset proton energy range and preset proton direction range are set according to specific engineering practices.

[0058] Specifically, the preset proton energy range is 13MeV~14MeV, with an interval of 0.1MeV, generating 11 energy points. The preset proton orientation range is -5°~+5°, with an interval of 1°, generating a total of 11 orientation points. The distribution of energy points generated by the incident protons is closely related to their energy, target material properties, and nuclear reaction type. Figure 2 This is a schematic diagram of the target and BSA structure provided by the present invention. Figure 2 Figure (a) in the diagram is a schematic diagram of the BSA structure. Figure 2 Figure (b) in the diagram is a layered schematic diagram of the target structure, as shown below. Figure 2 Figure (c) shows the neutron energy spectra in each direction of the three surfaces of the cylindrical target: the side surface, the out surface, and the inp surface. Figure 2 In Figure (c), R represents the radial distance between the incident point of the proton beam and the target center. The neutron energy range for each spectrum is from 10. -8 MeV to 12 MeV: Thermal neutrons are 10 -8 MeV to 5×10 -7 MeV with 40 log intervals, and 5 × 10⁻⁶ hyperthermal neutrons in BNCT. -7 MeV to 10 -2 MeV has 101 log-separated intervals, and fast neutrons have 10. -2The recording covers 72 intervals from MeV to 12 MeV, totaling 214 energy points. Neutron emission directions are recorded from cosθ=-1 to cosθ=1, with intervals of 0.05, providing 41 directional points. Furthermore, since the proton beam radius is smaller than the Be target, for outSurf and inpSurf, a circular region with a radius of 0.1 cm, a ring with a width of 0.4 cm, and 15 rings with a width of 0.5 cm are recorded. The Target Database comprises three sub-databases: TargoutSurf, TarginpSurf, and TargsideSurf. However, side-emission neutrons are relatively few and therefore were not used in subsequent training.

[0059] S102: When the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed, the target neutron energy spectrum data generated after hitting the target at the incident direction point is predicted by machine learning algorithm.

[0060] In an exemplary embodiment, the target neutron spectrum data generated after hitting a target at the incident direction point using a machine learning algorithm includes: predicting the normalized neutron spectrum corresponding to the proton at the incident direction point using a deep neural network; predicting the total neutron fluence rate of the unindexed proton energy point at the incident direction point using a random forest algorithm; and inverse normalizing the normalized neutron spectrum using the total neutron fluence rate to obtain the target neutron spectrum data generated after hitting a target at the incident direction point using an unindexed proton energy point.

[0061] Specifically, the deep neural network (DNN) module is mainly used to predict the neutron energy spectrum, such as... Figure 3 As shown in Figure (a), the DNN network uses residual connection blocks. Figure 3 Figure (b) shows the internal architecture of the residual network. n This indicates that the residual blocks can be connected. n Layer. LeakyReLU is used as the activation function, which effectively solves the gradient vanishing problem. The activation function is shown in formula (2):

[0062] (2);

[0063] in, x For input values, For the slope parameter, This is the activation function.

[0064] Dropout layers are regularization layers in deep neural networks, used to reduce the effects of overfitting. Their basic mathematical principle is that for each neuron in the hidden layer, the probability of its existence is... p The case where it is discarded is explained by the mathematical principle shown in formula (3):

[0065] (3);

[0066] in, p For probability, This is the output after the dropout operation. This is the input value for this layer.

[0067] The specific execution process of the DNN network is as follows: (1) Standardize the neutron energy spectrum to eliminate the influence of the dimensions between indicators and increase the accuracy of subsequent predictions; (2) Since neutrons have different distributions in different energy regions, the neutron energy spectrum is divided into thermal neutrons (0~0.5eV), ultrathermal neutrons (0.5eV~10keV) and fast neutrons (10keV~12MeV) for training respectively; (3) Save the DNN model parameters for the main program to call. The DNN modules in the two machine learning modules are Target_databsebroad and BSA_databasebroad.

[0068] The Random Forest (RF) algorithm is mainly used to predict the total neutron fluence rate corresponding to each set of parameters in the database and to inversely normalize the normalized neutron energy spectrum obtained by the DNN. The specific execution process of RF is as follows: (1) Obtain the total neutron fluence rate corresponding to each energy spectrum in the database; (2) Train the RF model of the total neutron fluence rate. The training data is the total neutron fluence rate in (1) and the corresponding parameters of the fluence rate in the database. The output value is the total neutron fluence rate; (3) Save the parameters of the RF model for the main program to call.

[0069] S103: For the proton corresponding to each energy point in the accelerator proton energy spectrum data at the incident direction point, multiply the proton fluence ratio by the target neutron energy spectrum data corresponding to the proton, and sum the multiplication results of all energy points and the protons corresponding to the incident direction points to obtain the target neutron energy spectrum.

[0070] S104: For each neutron energy point and direction point in the target neutron energy spectrum, index it in the BSA database. When the energy spectrum data corresponding to the neutron energy point at the direction point is not indexed, predict the energy spectrum data corresponding to the unindexed neutron energy point at the direction point through machine learning algorithm.

[0071] In an exemplary embodiment, the process of constructing the BSA database specifically includes: discretizing the neutrons in the target neutron database with a preset neutron angle range and a preset neutron energy range to obtain different neutron energy points and different neutron direction points; recording the energy spectrum data of the neutrons corresponding to different neutron energy points at different neutron direction points after passing through the BSA; and constructing the BSA database based on the different neutron energy points, different neutron direction points, and the energy spectrum data corresponding to different neutron energy points at different neutron direction points.

[0072] In an exemplary embodiment, predicting the energy spectrum data corresponding to the unindexed neutron energy points at the direction point using a machine learning algorithm specifically includes: predicting the normalized energy spectrum data of the unindexed neutron energy points at the direction point using a deep neural network; predicting the total neutron fluence rate of the unindexed neutron energy points at the direction point using a random forest algorithm; and inverse normalizing the normalized energy spectrum data using the total neutron fluence rate to obtain the energy spectrum data of the unindexed neutron energy points at the direction point.

[0073] In one exemplary embodiment, the deep neural network includes fully connected layers N1, N2, and N3 connected in series. n A residual network, a fully connected layer N2, and an activation function.

[0074] Specifically, such as Figure 2 As shown in Figure (c), for each ring of the front outSurf and the rear inpSurf, the neutron orientation varies from cosθ=0 to cosθ=1 at intervals of 0.05, producing 21 orientation points. Since neutrons with energies below 0.01 MeV can hardly penetrate the moderator, a total of 71 logarithmically spaced energy points spanning neutron energies from 0.01 MeV to 12 MeV were calculated. The neutron energy spectrum at the emission window was recorded. A database was not created for the sideSurf because neutrons emitted from this surface account for less than 1%. The BSA database comprises two subsets: the BSAoutSurf database and the BSAinpSurf database.

[0075] S105: Determine the neutron energy spectrum of the emission window based on the energy spectrum data of the neutron energy points at the direction points indexed in the BSA database and the energy spectrum data of the neutron energy points at the direction points predicted by the machine learning algorithm.

[0076] In an exemplary embodiment, the calculation of the neutron energy spectrum of the BSA exit window using the Energy Spectrum Superposition Method-Machine learning (ESSM-ML) method mainly consists of the following three steps: (1) The entire particle transport physics process is divided into two parts: a) Protons with specific energy and angle bombard the target material to generate target neutrons; b) After the target neutrons are slowed down and shaped by BSA, they obtain target neutrons in the exit window. (2) For physical process a, according to the upper and lower limits of the proton energy produced by the accelerator, the proton energy is obtained at certain intervals and the target neutron energy spectrum and directionality after the proton energy hits the target are calculated. A database of target neutron energy spectrum and direction after protons of different energies and directions hit the target is obtained. For physical process b, similar to physical process a, the neutron energy spectrum and direction data of the exit window after neutrons of different energies and directions on the target surface pass through the BSA are calculated. (3) The proportion of protons of each energy and direction in the accelerator proton energy spectrum is extracted and multiplied with the corresponding target neutrons in the database and added together to obtain the target neutron energy spectrum. (4) If some energy or direction parameters in the proton energy spectrum are missing in the database, the target database machine learning module is called to calculate the neutron energy spectrum corresponding to the missing parameters. The obtained energy spectrum is then incorporated into the main program for further calculation. (5) When the target neutrons pass through the BSA, the same method is used to obtain the neutron energy spectrum of the final exit window based on the target neutron energy spectrum.

[0077] This invention provides, for example Figure 4 The flowchart shown is for determining the neutron energy spectrum of the emission window. Figure 4 As shown, Figure 4 The index in the Target Database checks whether the energy points and direction points exist, i.e., it retrieves all proton energy points and incident direction points based on the proton energy spectrum of the accelerator and indexes them in the Target Database. The Target Database is the same as the Target Database. Figure 4 The target database machine learning module in this invention is the machine learning algorithm mentioned in this invention. Figure 4 The random forest in this invention is the same as the random forest algorithm in this paper. Figure 4 The Target_databasebroad neural network in this invention is the same as the deep learning neural network in this invention. Figure 4 The BSA_Database mentioned here refers to the BSA database in this invention; Figure 4 The BSA database machine learning module in this invention is the machine learning algorithm in this invention, and the BSA_databasebroad neural network is the deep learning neural network in this invention. Figure 4 The total neutron flux ratio is the same as the total neutron flux ratio in this invention.

[0078] like Figure 4As shown, proton energy flux and angular energy flux are obtained from the proton energy spectrum, and energy flux weights and angular energy flux weights are calculated. The energy spectrum data of the proton energy point at the incident direction point is predicted using the target database machine learning module. Based on the energy flux weights, angular energy flux weights, and the energy spectrum data predicted by the target database machine learning module, the target neutron energy spectrum is calculated. Neutron energy points are obtained from the target neutron energy spectrum, and angular energy flux and neutron energy flux are obtained from the neutron energy spectrum. Angular energy flux weights and neutron energy flux weights are calculated. The neutron energy spectrum data is predicted using the BSA database machine learning module. Based on the angular energy flux, angular energy flux weights, neutron energy flux, neutron energy flux weights, and the energy spectrum data predicted by the BSA database machine learning module, the exit window neutron energy spectrum is calculated.

[0079] The main structure of the algorithm is the Energy Spectrum Superposition (ESSM) algorithm. Based on ESSM, two databases, Target Database and BSADatabase, are established for the two physical processes of proton target generation and neutron entry into the beam shaping component (BSA). Two machine learning algorithm modules are added to each database. Each machine learning module contains a random forest (RF) algorithm and a deep neural network (DNN). The algorithms are trained on the database and predict the energy spectrum of missing parameters in the database.

[0080] Once the target and BSA models are determined, two physical process databases, namely the Target Database and the BSA Database, are pre-established using Monte Carlo simulation, and the corresponding machine learning module parameters are saved. When the accelerator proton energy spectrum and directionality are known, the corresponding neutron parameter information in the database is retrieved. If the database information is accurate, the machine learning module is invoked to quickly obtain the neutron energy spectrum of the emission window.

[0081] Figure 5 A comparison diagram of neutron energy spectra generated by the conventional method and the ESSM-ML method provided by this invention, as shown below. Figure 5 As shown, the total computation time of ESSM-ML was 69 s, the total number of predicted energy spectra was 8280, the single energy spectrum computation time was 0.0052 s, and the total number of calculated particles was 4.629 × 10⁻⁶. 10 The traditional method took 5673 minutes to calculate, while the single-energy spectroscopy method took 4 minutes, with a total of 5 × 10⁻⁶ particles calculated. 10(CPU: AMD 7950x, GPU: RTX 1066s). Therefore, the efficiency of ESSM-ML for calculating the output window energy spectrum is improved by 12,000 times, and the efficiency for calculating a single energy spectrum is improved by more than 46,000 times. Furthermore, the two databases contain a total of 180,000 energy spectra. Even if the database accuracy is improved by an order of magnitude to 1.8 million energy spectra, the total computation time will not exceed 10,000 seconds, but the uncertainty of the final output window energy spectrum can be reduced. If the accuracy of the database energy spectra is improved and the database size is increased, the frequency of calling the machine learning module can be reduced, lowering the computation time for the final output window neutron energy spectrum to below 30 seconds.

[0082] The advantages of the ESSM-ML method provided by this invention include:

[0083] (1) The present invention greatly improves the time required to obtain the neutron energy spectrum for treatment and reduces the patient's waiting time.

[0084] (2) The neutron spectrum can be adjusted at any time according to the treatment status of the tumor during radiotherapy, laying the foundation for image-guided BNCT radiotherapy in the future.

[0085] (3) The present invention calls the machine learning module to quickly obtain the neutron energy spectrum of the exact parameters based on the changes in accelerator parameters, thereby improving the adaptability of the algorithm to different sudden scenarios and reducing the uncertainty of the calculation.

[0086] (4) This invention can greatly reduce the time required to establish a database and reduce computational overhead.

[0087] When applying the method for determining the neutron energy spectrum of the emission window based on energy spectrum superposition and machine learning algorithm provided by this invention, it is not necessary to rely on... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this invention does not impose any restrictions on it.

[0088] The above describes a method for determining the neutron energy spectrum of an emission window based on energy spectrum superposition and machine learning algorithms, provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding apparatus for determining the neutron energy spectrum of an emission window based on energy spectrum superposition and machine learning algorithms, such as... Figure 2 As shown.

[0089] Figure 6 A schematic diagram of an emission window neutron energy spectrum determination device based on energy spectrum superposition and machine learning algorithm provided by the present invention includes:

[0090] The index module 601 is used to index the target neutron database based on all proton energy points and incident direction points in the accelerator proton energy spectrum.

[0091] The first prediction module 602 is used to predict the target neutron energy spectrum data generated after hitting the target at the incident direction point by using a machine learning algorithm when the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed.

[0092] The determination module 603 is used to multiply the proton fluence ratio of each energy point in the accelerator proton energy spectrum data with the target neutron energy spectrum data corresponding to the proton at the incident direction point, and sum the multiplication results of all energy points and the protons corresponding to the incident direction points to obtain the target neutron energy spectrum.

[0093] The second prediction module 604 is used to index each neutron energy point and direction point in the target neutron energy spectrum in the BSA database. When no energy spectrum data corresponding to the neutron energy point at the direction point is found in the index, the energy spectrum data corresponding to the unindexed neutron energy point at the direction point is predicted by machine learning algorithm.

[0094] The calculation module 605 is used to determine the neutron energy spectrum of the emission window based on the energy spectrum data of the neutron energy points at the direction points indexed in the BSA database and the energy spectrum data of the neutron energy points at the direction points predicted by the machine learning algorithm.

[0095] Specific limitations regarding the emission window neutron spectrum determination device based on energy spectrum superposition and machine learning algorithms can be found in the limitations of the emission window neutron spectrum determination method based on energy spectrum superposition and machine learning algorithms mentioned above, and will not be repeated here. Each module in the aforementioned emission window neutron spectrum determination device based on energy spectrum superposition and machine learning algorithms can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independently of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0096] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The method provided is based on the superposition of energy spectra and machine learning algorithms to determine the neutron energy spectrum of the emission window.

[0097] The present invention also provides Figure 7 The schematic diagram of the computer device shown is as follows: Figure 7 As shown, at the hardware level, this computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above. Figure 1 A method for determining the neutron energy spectrum in the emission window based on energy spectrum superposition and machine learning algorithms is provided.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

Claims

1. A method for determining the neutron energy spectrum of an emission window based on energy spectrum superposition and machine learning algorithms, characterized in that, include: Based on all proton energy points and incident direction points in the accelerator proton energy spectrum, indexing is performed in the target neutron database; When the target neutron energy spectrum data corresponding to the proton energy point at the incident direction point is not indexed, the target neutron energy spectrum data generated after hitting the target at the incident direction point is predicted by machine learning algorithm. For each energy point in the accelerator proton energy spectrum data, the proton fluence ratio is multiplied by the target neutron energy spectrum data corresponding to the proton, and the results of multiplying all energy points and the protons corresponding to the incident direction points are summed to obtain the target neutron energy spectrum. For each neutron energy point and direction point in the target neutron energy spectrum, an index is performed in the BSA database. When the energy spectrum data corresponding to the neutron energy point at the direction point is not found in the index, the energy spectrum data corresponding to the unindexed neutron energy point at the direction point is predicted by a machine learning algorithm. The neutron energy spectrum of the emission window is determined based on the energy spectrum data of the neutron energy points at the direction points indexed in the BSA database and the energy spectrum data of the neutron energy points at the direction points predicted by the machine learning algorithm. The spectral data of the neutron energy point includes the total neutron fluence; the formula for calculating the emission window energy spectrum is: ； in, It is the neutron energy spectrum of the emission window. The total target neutron fluence rate, For energy Direction is Total neutron flux rate, For direction Energy is Total neutron flux rate, It is energy Direction is neutron loss rate, For correction items, This is the upper limit of the summation of energy. Let the summation be the upper bound of the direction. This is a correction term for energy. For the direction correction term, This is the upper limit of the summation of the energy correction terms. The upper limit of the summation of the direction correction terms.

2. The method as described in claim 1, characterized in that, The method of predicting the target neutron energy spectrum data generated after hitting the target at the incident direction point using machine learning algorithms for unindexed proton energy points specifically includes: Predict the normalized neutron spectrum of protons at the incident direction point by using a deep neural network for protons whose energy points are not indexed. Predict the total neutron fluence rate at the incident direction point for proton energy points that are not indexed using the random forest algorithm; By inversely normalizing the standardized neutron spectrum using the total neutron flux rate, the target neutron spectrum data generated after hitting the target at the incident direction point from the unindexed proton energy point is obtained.

3. The method as described in claim 1, characterized in that, The method of predicting the energy spectrum data corresponding to the unindexed neutron energy points at the directional points using machine learning algorithms specifically includes: Predict normalized energy spectrum data at directional points for neutron energy points that are not indexed to the index using deep neural networks; Predict the total neutron flux rate at the directional point for neutron energy points that are not indexed using the random forest algorithm; By inversely normalizing the standardized energy spectrum data using the total neutron flux rate, the energy spectrum data of the unindexed neutron energy points at the directional points are obtained.

4. The method as described in claim 3, characterized in that, Deep neural networks consist of N1 fully connected layers connected in series. n A residual network, a fully connected layer N2, and an activation function.

5. The method as described in claim 1, characterized in that, The construction process of the target neutron database specifically includes: Within a preset proton energy range and a preset proton direction range, discretization is performed according to energy value and direction to obtain different proton energy points and different proton incident direction points; Record the target neutron energy spectrum data of protons at different proton energy points and at different proton incident directions after hitting the target; A target neutron database is constructed based on the target neutron energy spectrum data generated after protons hit the target at different proton energy points, different proton incident directions, and different proton energy points at different proton incident directions.

6. The method as described in claim 1, characterized in that, The construction process of the BSA database specifically includes: The neutrons in the target neutron database are discretized with preset neutron angle ranges and preset neutron energy ranges to obtain different neutron energy points and different neutron direction points; Record the energy spectrum data of neutrons at different neutron energy points and at different neutron direction points after passing through BSA; The BSA database is constructed based on the energy spectrum data corresponding to different neutron energy points, different neutron direction points, and different neutron energy points at different neutron direction points.

Images