BNCT treatment plan optimization method

Abstract

The invention provides a BNCT treatment plan optimization method. The method comprises the steps of obtaining a candidate set of BNCT treatment plans; determining a plurality of quality indexes of the BNCT treatment plan and corresponding index standard values, and determining an index scoring function based on the quality indexes and the corresponding index standard values; obtaining index weights of the plurality of quality indexes, and obtaining a plan scoring function based on the index weights and the index scoring function; and obtaining a score regression model based on the plan scoring function, and iteratively calculating a score maximum value in the candidate set based on the score regression model and the plan scoring function to obtain a corresponding optimal BNCT treatment plan. According to the method and the device, the BNCT treatment plan with the highest score can be selected with fewer computing resources by setting the plan scoring function, so that accurate and efficient BNCT treatment is realized.

Description

Technical Field

[0001] The present application belongs to the field of radiation technology and relates to a BNCT treatment plan optimization method. Background Art

[0002] BNCT (Boron Neutron Capture Therapy) is a precision radiotherapy that combines targeted drugs with neutron irradiation. It can selectively kill tumor cells and preserve normal cells. Compared with photon radiotherapy, BNCT effectively reduces the radiation to normal cells and reduces the side effects of radiotherapy. In addition, photon radiotherapy usually requires the dose to be divided into multiple irradiations, and the treatment cycle is long, while BNCT only requires 1 to 2 irradiations to reach the dose required to kill the tumor. The treatment time is very patient-friendly and reduces the patient's treatment burden. Compared with other radiotherapy technologies, BNCT has obvious advantages in treatment time, accuracy and side effect control, and has broad application prospects.

[0003] Treatment planning (TP) plays a vital role in BNCT. A good BNCT treatment plan can ensure the radiation dose to the tumor target area and reduce the radiation dose to normal tissues, ensuring successful treatment. Therefore, how to efficiently optimize the BNCT treatment plan is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0004] The purpose of this application is to provide a BNCT treatment plan optimization method to solve the technical problem of how to efficiently optimize the BNCT treatment plan.

[0005] In a first aspect, the present application provides a BNCT treatment plan optimization method, the method comprising: Obtain a candidate set of BNCT treatment plans; Determining a plurality of quality indicators and corresponding standard values of the indicator of the BNCT treatment plan, and determining an indicator scoring function based on the quality indicators and the corresponding standard values of the indicator; Obtaining indicator weights of a plurality of the quality indicators to obtain a plan scoring function based on the indicator weights and the indicator scoring function; A scoring regression model is obtained based on the plan scoring function, and a maximum score value in the candidate set is iteratively calculated based on the scoring regression model and the plan scoring function to obtain a corresponding optimal BNCT treatment plan.

[0006] In one embodiment of the present application, determining an indicator scoring function based on the quality indicator and the corresponding indicator standard value includes: Obtaining the indicator gap between the current value of the quality indicator and the standard value of the indicator; Perform reward evaluation or penalty evaluation on the metric gap based on the evaluation slope; Optimize the metric gap based on the reward evaluation and the penalty evaluation to obtain the metric scoring function.

[0007] In an embodiment of the present application, performing reward evaluation or penalty evaluation on the metric gap based on the evaluation slope includes: When the metric gap meets the preset gap target, perform reward evaluation on the metric gap using the reward evaluation slope; When the metric gap does not meet the preset gap target, perform penalty evaluation on the metric gap using the penalty evaluation slope; wherein, the reward evaluation slope is less than the penalty evaluation slope.

[0008] In an embodiment of the present application, determining multiple quality metrics for the BNCT treatment plan includes: Determine the region of interest for the BNCT treatment plan, where the region of interest includes the tumor target area, organs at risk, and the planned area; Determine multiple quality metrics for the region of interest based on the radiation criteria of the region of interest In an embodiment of the present application, obtaining the scoring regression model based on the planned scoring function includes: Select the mean function and the covariance function; Specify a sparse prior distribution for the hyperparameters of the covariance function; Fit the planned scoring function based on the mean function, the covariance function, and the sparse prior distribution of the hyperparameters to obtain the scoring regression model.

[0009] In an embodiment of the present application, iteratively calculating the maximum score in the candidate set based on the scoring regression model and the planned scoring function includes: Randomly select multiple BNCT treatment plans from the candidate set as known sample points; Calculate the scores of the known sample points through the planned scoring function, and use the known sample points and the corresponding scores as the known sample set; Iteratively update the posterior distribution of the scoring regression model using the known sample set to obtain the prediction distribution of the unrated plans in the candidate set based on the posterior distribution; Iteratively update the known sample set based on the prediction distribution of the unrated plans until the maximum number of iterations is reached, and then use the BNCT treatment plan with the highest score in the known sample set as the optimal plan.

[0010] In an embodiment of the present application, iteratively updating the known sample set based on the prediction distribution of the unrated plans includes: Select an optimal plan to be scored based on the prediction distribution of the un-scored plans, and calculate the score of the optimal plan to be scored through the calculation scoring function; Add the optimal plan to be scored and the corresponding score to the known sample set to iteratively update the known sample set.

[0011] In an embodiment of the present application, selecting an optimal plan to be scored based on the prediction distribution of the un-scored plans includes: Evaluate the potential value of the un-evaluated plan based on the prediction distribution of the un-scored plans through an acquisition function; Select the un-scored plan with the highest potential value as the optimal plan to be scored.

[0012] In an embodiment of the present application, obtaining a candidate set of BNCT treatment plans includes: Determine multiple irradiation fields; Randomly determine multiple groups of corresponding irradiation weights, and the sum of each group of irradiation weights is 1; the irradiation weights correspond to the irradiation time of the irradiation fields; Randomly and exhaustively combine the multiple irradiation fields with the multiple groups of irradiation weights to form multiple candidate BNCT treatment plans.

[0013] In an embodiment of the present application, determining multiple irradiation fields includes: Determine the geometric center point of the tumor target area; Calculate the distance between the contour points of the tumor target area and the geometric center point to determine the nearest irradiation angle; Taking the nearest irradiation angle as the center, rotate one week in the horizontal plane of the tumor target area to determine multiple irradiation fields at different irradiation angles.

[0014] As described above, the BNCT treatment plan optimization method described in the present application has the following beneficial effects: In the present application, by setting multiple quality indicators of the BNCT treatment plan and the standard values of the indicators to obtain an indicator scoring function, and further obtaining a plan scoring function according to different weights, so as to score multiple BNCT treatment plans, which helps to select the optimal BNCT treatment plan based on the actual situation of the patient to achieve the best treatment effect. And the plan scoring function helps to quantitatively evaluate the quality of the BNCT treatment plan, avoiding the influence of the differences in the experience of the designers or the time cost on the plan quality. In addition, the present application iteratively calculates the maximum score based on the plan scoring function, and can set the number of iterations according to the actual situation. When the number of iterations is reached, the BNCT treatment plan with the maximum score is selected as the optimal plan. This significantly reduces the computing resources, can more efficiently select the optimal plan according to the actual situation, and reduces the optimization time. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Shows a schematic flow chart of the BNCT treatment plan optimization method described in the embodiments of the present application.

[0016] Figure 2 Shows a schematic flow chart of determining the index scoring function described in the embodiments of the present application.

[0017] Figure 3 Shows a schematic flow chart of obtaining the scoring regression model described in the embodiments of the present application.

[0018] Figure 4 Shows a schematic flow chart of iteratively calculating the maximum score in the candidate set described in the embodiments of the present application.

[0019] Figure 5 Shows a schematic logic diagram of the BNCT treatment plan optimization method described in the embodiments of the present application. Detailed implementation manners

[0020] The following uses specific specific examples to illustrate the implementation manners of the present application. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0021] It should be noted that in the following description, reference is made to the accompanying drawings, which describe several embodiments of the present application. It should be understood that other embodiments can also be used, and mechanical composition, structure, electrical, and operational changes can be made without departing from the spirit and scope of the present application. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present application is only defined by the claims of the published patent. The terms used here are only for describing specific embodiments and are not intended to limit the present application. Spatially related terms, such as "upper", "lower", "left", "right", "below", "beneath", "lower part", "above", "upper part", etc., can be used in the text to facilitate the description of the relationship between one element or feature shown in the figure and another element or feature.

[0022] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. It should be further understood that the terms "comprising", "including" indicate the presence of the stated features, operations, elements, components, items, kinds, and / or groups, but do not preclude the presence, occurrence or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are to be construed as inclusive, or meaning any one or any combination.

[0023] Treatment planning quality assessment (TPQA) technology is crucial in the field of radiotherapy as it is used to ensure the accuracy and safety of treatment plans. Accurate quality assessment can help designers optimize the BNCT treatment plan to select the optimal BNCT treatment plan, thereby achieving the best treatment effect. Generally speaking, the main principle for evaluating the quality of the BNCT treatment plan is to ensure the radiation dose to the tumor target area and reduce the radiation dose to normal tissues. To this end, designers often use the Monte Carlo method to simulate the plan continuously to adjust the relevant parameters of the BNCT treatment plan, such as the irradiation time of different irradiation fields, etc., in order to make the treatment plan achieve the optimal effect. This is obviously very time-consuming and laborious, and requires a large amount of computing resources and computing time. In addition, due to the lack of specific evaluation indicators, the BNCT treatment plan is often affected by the differences in designers' experience and time costs, etc., and it is difficult to design a suitable treatment plan according to the real situation of patients through quantitative evaluation of the plan to achieve the best treatment effect.

[0024] To at least solve the above technical problems, the embodiments of the present application provide a method for optimizing a BNCT treatment plan. Through a plan scoring function, it can quantitatively evaluate the quality of the BNCT treatment plan with less computing resources and computing time, avoid the influence of differences in designers' experience or time costs on the plan quality, and help select the optimal BNCT treatment plan based on the real situation of patients to achieve the best treatment effect.

[0025] Figure 1 The flowchart of the method for optimizing a BNCT treatment plan according to the embodiments of the present application is shown. As Figure 1 shown, the method for optimizing a BNCT treatment plan provided by the embodiments of the present application includes steps S1 to S4.

[0026] S1. Obtain a candidate set of the BNCT treatment plan.

[0027] In some embodiments, obtaining a candidate set of BNCT treatment plans includes: determining a plurality of irradiation fields; randomly determining multiple sets of corresponding numbers of irradiation weights, the sum of each set of the irradiation weights being 1; the irradiation weights corresponding to the irradiation times of the irradiation fields; and randomly and exhaustively combining the multiple irradiation fields with the multiple sets of irradiation weights to form multiple candidate BNCT treatment plans.

[0028] In some embodiments, determining a plurality of irradiation fields includes: determining the geometric center point of the tumor target area; calculating the distances between the contour points of the tumor target area and the geometric center point to determine the nearest irradiation angle; and rotating 360 degrees in the horizontal plane of the tumor target area with the nearest irradiation angle as the center to determine multiple irradiation fields at different irradiation angles.

[0029] Specifically, after determining the tumor target area, the contour point positions of the tumor target area can be outlined through a DICOM-RT Struct file, and the arithmetic mean of the sums of these positions can be taken as the geometric center point A of the tumor target area. Then, the distances between the contour points of the skin outlined in the DICOM-RT Struct file and point A are calculated, and the point with the minimum distance is used as the starting point. The vector pointing to point A is the irradiation angle closest to the geometric center point A on the skin surface. Rotate this irradiation angle 360 degrees around point A in the horizontal plane, and record any angle selected therefrom as the irradiation angle of an irradiation field, so that multiple irradiation fields at different irradiation angles can be selected.

[0030] Further, since the irradiation field usually needs to be selected on the side where the skin is closer to the tumor for irradiation and the treatment couch needs to be avoided, these unreasonable angles can be removed to obtain the final multiple irradiation fields.

[0031] Further, after determining the multiple irradiation fields, multiple sets of corresponding numbers of irradiation weights are randomly determined. Each irradiation weight corresponds to the irradiation time of the irradiation field, and the sum of each set of irradiation weights is 1. In this way, the multiple irradiation fields and the multiple sets of irradiation weights are randomly and exhaustively combined to form multiple candidate BNCT treatment plans. That is, in the embodiments of the present application, each candidate BNCT treatment plan actually only differs in the irradiation time of the irradiation field. That is, the candidate set of the BNCT treatment plan includes all possible combinations of the irradiation times of the multiple irradiation fields.

[0032] For example, taking two irradiation fields with a fixed total irradiation time of 60 minutes as an example for introduction. Multiple groups of two irradiation weights are randomly determined in [0, 1], and the sum of the two irradiation weights is 1. For example, the two irradiation weights of the first group are 0.9 and 0.1 respectively; the two irradiation weights of the second group are 0.8 and 0.2 respectively... A total of 10 groups of irradiation weights are randomly determined. Then, the 10 groups of irradiation weights are randomly and exhaustively combined with the two irradiation fields, forming a total of 20 candidate BNCT treatment plans and constituting a candidate set of BNCT treatment plans.

[0033] It should be noted that when randomly determining multiple groups of irradiation weights corresponding to the quantity, the irradiation weight can be set to 0, which means that the irradiation time of a certain irradiation field can be 0.

[0034] S2. Determine multiple quality indicators of the BNCT treatment plan and the corresponding indicator standard values to determine an indicator scoring function based on the quality indicators and the corresponding indicator standard values.

[0035] In some embodiments, the designer of the BNCT treatment plan can design multiple quality indicators for evaluating the treatment plan and the corresponding indicator standard values according to the needs of the treatment. For example, the designer can first determine the region of interest (ROI) of the BNCT treatment plan in the patient's body, including the tumor target area, the organs at risk, or other planned areas, etc. Then, the designer can determine the radiation standard of the region of interest according to the treatment focus and the safety criteria of the plan, so as to determine multiple quality indicators of the region of interest, such as the minimum radiation index of the tumor target area, the average radiation index of the tumor target area, the maximum radiation index of the organs at risk, etc., and set the standard values of the quality indicators correspondingly. Among them, the radiation standard refers to the standard that the radiation needs to reach in the region of interest. Taking the minimum radiation index of the tumor target area as an example, when performing BNCT treatment, in order to be able to irradiate all cancer cells and ensure the treatment effect, the minimum radiation value of the tumor target area should not be less than the prescribed dose at this time.

[0036] Furthermore, the designer can also set the quality indicators in more detail. For example, the tumor target area can be further divided, and the minimum index of the first sub-region of the tumor target area, the minimum index of the second sub-region, etc. can be set. Similarly, the organs at risk can also be subdivided into each organ at risk itself and the corresponding standard values can be set for it. That is, in fact, when the quality indicators are designed in more detail, the designer can evaluate the BNCT treatment plan itself in more dimensions. In addition, when it is necessary to reduce computing resources or time, the designer can also only design key quality indicators.

[0037] It should be understood that the design of any quality index is related to the BNCT treatment plan. Generally speaking, considering the treatment effectiveness and safety of the BNCT treatment plan, the minimum value index of the tumor target area and the maximum value index of the organs at risk are required. The former is to ensure that the tumor target area can receive sufficient radiation dose, while the latter is to ensure that normal tissues are not affected by radiation.

[0038] Figure 2 The flowchart showing the process of determining the index scoring function according to the embodiments of the present application is as follows Figure 2 As shown, determining the index scoring function based on the quality index and the corresponding index standard value includes steps S21 to S23.

[0039] S21. Obtain the index gap between the current value of the quality index and the index standard value.

[0040] S22. Conduct a reward evaluation or a penalty evaluation on the index gap based on the evaluation slope.

[0041] S23. Optimize the index gap based on the reward evaluation and the penalty evaluation to obtain the index scoring function.

[0042] As mentioned above, the quality index actually quantitatively evaluates the quality of the BNCT treatment plan. When the quality of the BNCT treatment plan is higher, the current value of the quality index should be closer to the standard value, and vice versa, that is, the gap between the current value of the quality index and the standard value reflects the evaluation result of the BNCT treatment plan under this quality index. Therefore, the index gap between the current value of the quality index and the index standard value can be obtained, and a reward or penalty evaluation can be conducted on the index gap based on the evaluation slope, so as to optimize the index gap to obtain the index scoring function.

[0043] Among them, when the index gap meets the preset gap target, a reward evaluation slope is used to conduct a reward evaluation on the index gap; when the index gap does not meet the preset gap target, a penalty evaluation slope is used to conduct a penalty evaluation on the index gap; among them, the reward evaluation slope is less than the penalty evaluation slope. Thus, in the process of optimizing the quality index, the index scoring function can output an optimal solution closer to the quality index standard value based on the reward evaluation or the penalty evaluation.

[0044] In some embodiments, the index scoring function can be represented by Equation (1):

[0045] Among them, is the score of the i-th quality index, is the current value of the i-th quality index, is the index standard value of the i-th quality index, is the evaluation slope.

[0046] S3. Obtain the index weights of multiple said quality indexes, so as to obtain a planned scoring function based on the index weights and the said index scoring function.

[0047] As mentioned above, multiple quality indexes can evaluate the treatment of the BNCT treatment plan from multiple dimensions. However, when evaluating, the focus of the BNCT treatment plan may vary due to individual differences and actual treatment conditions. For example, the importance of different organs at risk may vary, etc. Therefore, in this application, by obtaining the index weights of multiple said quality indexes, when evaluating the entire treatment plan, the scores of important quality indexes occupy a more important part. In some embodiments, the planned scoring function can be represented by Equation (2):

[0048] wherein, is the BNCT treatment plan score, is the independent variable vector, represents the irradiation time of the u-th irradiation field, and its domain is ; is the weight of the i-th index; is the importance of the i-th index.

[0049] In some embodiments, the importance can be classified into levels. The designer can, according to the specific situation of the patient, such as different factors such as tumor type, size, depth and location, etc., evaluate the importance level of different indexes, and calculate the corresponding index weights based on the importance level.

[0050] S4. Obtain a scoring regression model based on the said planned scoring function, and iteratively calculate the maximum score in the said candidate set based on the scoring regression model and the planned scoring function, so as to obtain the corresponding optimal BNCT treatment plan.

[0051] It can be seen from Equation (2) that in this application, different candidate BNCT treatment plans are quantitatively scored through the planned scoring function, so as to select the plan with the highest score as the optimal treatment plan. In some embodiments, the plan with the highest score can be taken as the optimization problem to be solved, and the maximum value of the optimization problem can be solved through the Bayesian optimization algorithm, so as to obtain the optimal BNCT treatment plan. Figure 3 shows the flow schematic diagram of obtaining the scoring regression model described in the embodiments of this application. As Figure 3 shown, obtaining the scoring regression model based on the said planned scoring function includes steps S41 to S43.

[0052] S41. Select a mean function and a covariance function.

[0053] In some embodiments, the mean function is set to a zero mean, and the covariance function is selected as an RBF kernel (Radial Basis Function Kernel) to describe the similarity between input points and which can be represented by Equation (3):

[0054] where, is the kernel variance; is the length scale.

[0055] S42. Specify a sparse prior distribution for the hyperparameters of the covariance function.

[0056] In the process of constructing a scoring regression model for the objective function, the present application quantifies the uncertainty of prediction by specifying a prior distribution for the hyperparameters of the covariance function. For the input dimension (the irradiation field with non-zero irradiation time), a larger length scale indicates the relevance of the input in this dimension to the planned scoring function, while for the dimension without input (the irradiation field with zero irradiation time), the length scale tends to concentrate near zero, indicating that this dimension has nothing to do with the planned scoring function. That is, through the length scale, it is possible to ensure that the scoring regression model effectively captures the structure of the data and significantly reduces the computational complexity based on different dimensions (the irradiation times of different irradiation fields). In some embodiments, specifying a sparse prior distribution for the hyperparameters of the covariance function can be represented by Equation (4):

[0057] where, the sparse prior distribution of the kernel variance is a gamma distribution; the sparse prior distribution of the length scale is a half-Cauchy distribution, determines the smoothness of the covariance function in the u-th dimension. When is larger, the sensitivity of this dimension is lower.

[0058] S43. Fit the planned scoring function based on the mean function, the covariance function, and the sparse prior distribution of the hyperparameters to obtain the scoring regression model.

[0059] In the process of obtaining the scoring regression model, a noise error can be set to further improve the anti-interference ability of the scoring regression model. Then in some embodiments, the scoring regression model can be represented by Equation (5):

[0060] where, is the noise term. In some embodiments, the noise term is set to 10-6 , it should be noted that this application is not limited thereto.

[0061] Furthermore, it can be seen from Equation (5) that in fact, the scoring regression model indicates that given the input , the prior distribution of the planned scoring function is a multivariate normal distribution, and its mean is , and the variance is . It is itself an assumption about the distribution of the planned scoring function. That is, in the absence of any samples, the prior distribution is objectively the same for each dimension and does not hold any attitude towards a certain dimension. Assuming that the scores of all BNCT treatment plans in the candidate set follow the prior distribution of the scoring regression model, then given some samples, that is, the input sample observation data, the prior distribution of the model can be updated to obtain the posterior distribution, and the unknown samples should actually also follow the posterior distribution of the model. Figure 4 shows a schematic flowchart of iteratively calculating the maximum score in the candidate set according to an embodiment of the present application. As Figure 4 shown, iteratively calculating the maximum score in the candidate set based on the scoring regression model and the planned scoring function includes steps S441 to S444.

[0062] S441. Randomly select multiple BNCT treatment plans from the candidate set as known sample points.

[0063] S442. Calculate the scores of the known sample points through the planned scoring function, and use the known sample points and the corresponding scores as a known sample set.

[0064] In some embodiments, assuming that the scores of all BNCT treatment plans in the candidate set follow the prior distribution of the scoring regression model, then randomly select multiple BNCT treatment plans from the candidate set as known sample points to form a known sample point set , and calculate the scores of the known sample points (the known sample points and the corresponding scores form a known sample set). Then the scores of the known sample points follow the prior distribution of the scoring regression model, that is , and its probability density function is shown by Equation (6):

[0065] where is the mean vector, that is , , that is the variance; I is the identity matrix.

[0066] S443. Use the known sample set to iteratively update the posterior distribution of the scoring regression model to obtain the prediction distribution of the un-scored plans in the candidate set based on the posterior distribution.

[0067] In some embodiments, a set of known sample points Y that follows a scoring regression model and the scores corresponding to each known sample point therein are used as observed data and input into a No-U-Turn Sampler (NUTS) together with a sparse prior distribution of the hyperparameters of the covariance function for sampling to update the distribution of the model, i.e., to obtain the posterior distribution of the model.

[0068] Then the scores of the un-scored plans (unknown sample points in the set of unknown sample points in the candidate set) should also follow the posterior distribution of the scoring regression model, and the conditional distribution X|Y also follows the posterior distribution of the scoring regression model, and it can be shown by Equation (7):

[0069] where the conditional mean is and the conditional covariance is Then, by combining Equation (6) and Equation (7), the predicted mean and predicted variance of X can be obtained, which is also the predicted distribution of the un-scored plans in the candidate set.

[0070] S444. Iteratively update the known sample set based on the predicted distribution of the un-scored plans until the maximum number of iterations is reached, and then take the BNCT plan with the maximum score value in the known sample set as the optimal BNCT treatment plan.

[0071] In fact, to obtain the optimal plan among the candidate BNCT treatment plans is to find the candidate plan with the maximum score. After obtaining the scores of the known sample point set, the scores of the unknown sample points, i.e., the un-scored plans in the candidate set, can be further calculated and compared with the scores of the known sample points, so as to update the maximum score value in the known sample set and further update the posterior distribution of the scoring regression model, and iterate repeatedly until the maximum score value is found. In some embodiments, iteratively updating the known sample set based on the predicted distribution of the un-scored plans includes: selecting the optimal plan to be scored based on the predicted distribution of the un-scored plans, and calculating the score of the optimal plan to be scored through the scoring function; adding the optimal plan to be scored and the corresponding score to the known sample set to iteratively update the known sample set.

[0072] To further reduce computing resources, this application does not need to calculate the scores for all unrated plans in the candidate set. Instead, it evaluates the potential value of unrated plans based on their predicted distributions, selects the optimal plan to be scored, and calculates its score. That is, in the embodiments of this application, the acquisition function is used to evaluate the potential value of the unevaluated plans based on the predicted distributions of the unrated plans, and selects the unrated plan with the highest potential value as the optimal plan to be scored.

[0073] In some embodiments, the acquisition function is used to select the improvement degree in the unknown sample point set X and select the unknown sample point with the largest improvement degree to calculate its score, and add it to the known sample point set Y, and compare it with the maximum known score in the known sample set and update the maximum score and the posterior distribution of the score regression model. Then, the posterior distribution of the updated score regression model is used to re-predict the predicted distribution in the unknown sample point set X again, and the acquisition function is used again to find , and iterate repeatedly until the maximum number of iterations is reached and then stop, output the maximum score at this time, and use the corresponding BNCT treatment plan as the optimal plan.

[0074] In some embodiments, the EI acquisition function is selected to select the unknown sample point with the largest improvement degree in the unknown sample point set X (unrated plans) as the optimal plan to be scored. This acquisition function provides a quantitative evaluation of the improvement degree by calculating the integral on the predicted distribution of the unknown sample point set X, balances the exploration and exploitation of unknown sample points, and helps the efficient progress of the optimization process. Among them, the EI acquisition function can be shown by Equation (8):

[0075] Among them, is the predicted mean of in the unknown sample point set X, is the predicted variance of in the unknown sample point set X; represents converting a potentially better observed value into a standard Gaussian distribution, is the probability density function of the standard normal distribution, is the cumulative distribution function of the standard normal distribution.

[0076] It should be noted that in the process of calculating the score of the BNCT treatment plan through the plan scoring function, the Monte Carlo method needs to be used to simulate the dose distributions of multiple irradiation fields to determine whether the treatment plan can meet the quality indicators.

[0077] Figure 5It shows a logical schematic diagram of the BNCT treatment plan optimization method described in the embodiments of the present application. As Figure 5 shown, a candidate set of the BNCT treatment plan is obtained, and multiple plans are selected from it as known sample points, and their scores are calculated to form a known sample point set. After that, the prior distribution of the score regression model is obtained, and the posterior distribution of the model is updated based on the known sample points, so as to predict the prediction distribution of the un-scored plans in the candidate set. After obtaining the prediction distribution, the improvement degree of the un-scored plan is calculated, and the un-scored plan with the largest improvement degree is selected as the plan to be scored, and its score is calculated, so as to update the known sample point set, and iterate repeatedly until the maximum number of iterations is reached, then the maximum score in the known sample point set is output, and the corresponding plan is used as the optimal plan.

[0078] It should be noted that, in addition to the number of iterations, a score threshold can also be designed as the stop condition, that is, when the score threshold is reached, the iteration stops, and the maximum score at this time is output.

[0079] Therefore, the present application obtains an index scoring function by setting multiple quality indexes and index standard values, and further obtains a plan scoring function according to different weights, which helps to select the optimal BNCT treatment plan based on the actual situation of the patient to achieve the best treatment effect. And the plan scoring function helps to quantitatively evaluate the quality of the BNCT treatment plan, avoiding the influence of the differences in the experience of designers or time costs on the plan quality. In addition, the present application iteratively calculates the maximum score based on the plan scoring function, and can set the number of iterations according to the actual situation. When the number of iterations is reached, the BNCT treatment plan with the maximum score is selected as the optimal plan. This significantly reduces the computing resources, can more efficiently select the optimal plan according to the actual situation, and reduces the optimization time. The protection scope of the BNCT treatment plan optimization method in the embodiments of the present application is not limited to the execution order of the steps listed in this embodiment. Any solution realized by adding or subtracting steps of the prior art and replacing steps according to the principle of the present application is included in the protection scope of the present application.

[0080] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0081] The above embodiments are only illustrative of the principles and effects of the present application and are not intended to limit the present application. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed in the present application should still be covered by the claims of the present application.

Claims

1. A BNCT treatment plan optimization method, characterized in that The method includes: Obtaining a candidate set of BNCT treatment plans; Determining multiple quality indicators of the BNCT treatment plan and corresponding indicator standard values, so as to determine an indicator scoring function based on the quality indicators and the corresponding indicator standard values; Obtaining the indicator weights of the multiple quality indicators, so as to obtain a plan scoring function based on the indicator weights and the indicator scoring function; Obtaining a scoring regression model based on the plan scoring function, and iteratively calculating the maximum score in the candidate set based on the scoring regression model and the plan scoring function, so as to obtain the corresponding optimal BNCT treatment plan.

2. The BNCT treatment plan optimization method according to claim 1, characterized in that Determining an indicator scoring function based on the quality indicators and the corresponding indicator standard values includes: Obtaining the indicator gap between the current value of the quality indicator and the indicator standard value; Performing a reward evaluation or a penalty evaluation on the indicator gap based on an evaluation slope; Optimizing the indicator gap based on the reward evaluation and the penalty evaluation to obtain the indicator scoring function.

3. The BNCT treatment plan optimization method according to claim 2, wherein Performing a reward evaluation or a penalty evaluation on the indicator gap based on an evaluation slope includes: When the indicator gap meets a preset gap target, performing a reward evaluation on the indicator gap using a reward evaluation slope; When the indicator gap does not meet the preset gap target, performing a penalty evaluation on the indicator gap using a penalty evaluation slope; wherein, the reward evaluation slope is less than the penalty evaluation slope.

4. The BNCT treatment plan optimization method according to claim 1, wherein, Determining multiple quality indicators of the BNCT treatment plan includes: Determining the region of interest of the BNCT treatment plan, where the region of interest includes a tumor target area, an organ at risk, and a planned area; Determining multiple quality indicators of the region of interest based on the radiation standard of the region of interest.

5. The BNCT treatment plan optimization method according to claim 1, characterized in that, Obtaining a scoring regression model based on the plan scoring function includes: Selecting a mean function and a covariance function; Specifying a sparse prior distribution for the hyperparameters of the covariance function; Fitting the plan scoring function based on the mean function, the covariance function, and the sparse prior distribution of the hyperparameters to obtain the scoring regression model.

6. The BNCT treatment plan optimization method according to claim 1, wherein Iteratively calculating the maximum score in the candidate set based on the scoring regression model and the plan scoring function includes: Randomly extracting multiple BNCT treatment plans from the candidate set as known sample points; Calculating the scores of the known sample points through the plan scoring function, and using the known sample points and the corresponding scores as a known sample set; Iteratively updating the posterior distribution of the scoring regression model using the known sample set, so as to obtain the prediction distribution of the unrated plans in the candidate set based on the posterior distribution; Iteratively updating the known sample set based on the prediction distribution of the unrated plans until the maximum number of iterations is reached, and then using the BNCT treatment plan with the highest score in the known sample set as the optimal plan.

7. The BNCT treatment plan optimization method according to claim 6, wherein Iteratively updating the known sample set based on the prediction distribution of the unrated plans includes: Selecting an optimal plan to be scored based on the prediction distribution of the unrated plans, so as to calculate the score of the optimal plan to be scored through the calculation scoring function; Adding the optimal plan to be scored and the corresponding score to the known sample set to iteratively update the known sample set.

8. The BNCT treatment plan optimization method according to claim 7, characterized in that, Selecting the optimal plan to be scored based on the prediction distribution of the un-scored plans includes: Evaluating the potential value of the un-evaluated plan based on the prediction distribution of the un-scored plan through an acquisition function; Selecting the un-scored plan with the highest potential value as the optimal plan to be scored.

9. The BNCT treatment plan optimization method according to claim 1, characterized in that Obtaining a candidate set of BNCT treatment plans includes: Determining a plurality of irradiation fields; Randomly determining multiple groups of corresponding irradiation weights, the sum of each group of the irradiation weights being 1; the irradiation weights corresponding to the irradiation time of the irradiation fields; Randomly and exhaustively combining the multiple irradiation fields with the multiple groups of irradiation weights to form multiple candidate BNCT treatment plans.

10. The BNCT treatment plan optimization method according to claim 9, wherein Determining a plurality of irradiation fields includes: Determining the geometric center point of the tumor target area; Calculating the distance between the contour points of the tumor target area and the geometric center point to determine the nearest irradiation angle; Rotating one week in the horizontal plane of the tumor target area with the nearest irradiation angle as the center to determine multiple irradiation fields at different irradiation angles.

Images