Generating and applying robust dose prediction models

Abstract

Nominal values and perturbations of the nominal values of parameters associated with a previously defined radiation treatment plan are accessed (202). For each treatment field of the treatment plan, a field-specific planning target volume (fsPTV) is determined (204) based on those perturbations. At least one clinical target volume (CTV) and at least one organ at risk (OAR) volume are also delineated. Each OAR includes at least one sub-volume that is delineated (208) based on a spatial relationship between each OAR and the CTV and the fsPTV for each treatment field. Dose distributions for the sub-volumes are determined (210) based on the nominal values and the perturbations. One or more dose prediction models are generated (f212) for each sub-volume. The dose prediction model(s) are trained using the dose distributions.

Description

BACKGROUND

[0001] It is well known to treat cancer using radiation therapy. Typically, radiation therapy involves directing a beam of high-energy protons, photons, ions, or electrons radiation ("treatment radiation") into a target or target volume (e.g., a volume including a tumor or lesion).

[0002] Prior to treating a patient with radiation, a treatment plan is developed for the patient. The plan uses simulations and optimizations based on past experience to define various aspects of the therapy. Generally, the intent of the treatment plan is to deliver enough radiation to unhealthy tissue while minimizing exposure of surrounding healthy tissue to radiation.

[0003] The goal of the planner is to find a solution that is optimal for multiple clinical objectives, which can be contradictory, i.e., an improvement in one objective can have an adverse effect on achieving another objective. For example, a treatment plan that avoids the liver from receiving a certain dose of radiation can result in the stomach receiving too much radiation. These types of tradeoffs result in an iterative process in which the planner creates different plans to find an optimized (best fit) one for achieving the desired results.

[0004] A critical component of a treatment plan is the prediction of the dose and dose distribution of radiation that will be applied to the patient. In knowledge-based dose prediction, information from previously planned radiation treatments can be used to gain knowledge about what dose distributions can be achieved in a new case without having to perform an actual plan. One approach to knowledge-based dose prediction is to use a set of treatment plans to create a model that can then be used to predict a dose for a new case. The prediction can be converted into an optimization objective that, when used in conjunction with an optimization algorithm, results in a complete treatment plan. SUMMARY

[0005] The present invention provides a computer-implemented method as defined in the claims.

[0006] It is desirable that the dose prediction models are "robust" in that they account for perturbations that can occur during treatment. For example, the position of the patient can change during treatment, and it is preferred that the dose prediction models account for patient movement when using the models to optimize the objectives of a radiation or radiation therapy treatment plan prior to treating the patient.

[0007] Embodiments according to the present invention provide an improved method of radiation treatment planning. More specifically, embodiments according to the present invention relate to the generation of robust dose prediction models, and to the application of those models to develop and optimize radiation treatment plans.

[0008] In one aspect of the application, a computer-implemented method is provided, comprising: accessing a plurality of treatment plans, the plurality of treatment plans having perturbations of nominal values of parameters of the treatment plans associated with the plurality of treatment plans;

[0009] determining, for each treatment field of the treatment plans, a field-specific planned target volume (fsPTV) based on the perturbations;

[0010] accessing information delineating at least one clinical target volume (CTV) and information delineating at least one organ at risk (OAR) volume, wherein each of the OAR volumes includes at least one sub-volume delineated based on a spatial relationship between the each OAR volume and the CTV and the fsPTV for the each treatment field;

[0011] determining a dose distribution for each of the sub-volumes, wherein the dose distribution includes a dose distribution based on the nominal values and a dose distribution based on the perturbations; and

[0012] generating a dose prediction model for each of the sub-volumes, wherein the model includes a model trained using the dose distribution based on the nominal values and a model trained using the dose distribution based on the perturbations.

[0013] Each of the dose prediction models can include information associating the each model with the dose distribution used to train the each model.

[0014] The method can include applying the dose prediction models to a radiation treatment plan, wherein the applying includes: using the dose prediction models to calculate dose-volume histograms for the radiation treatment plan, wherein the dose-volume histograms include dose-volume histograms calculated using the nominal values and dose-volume histograms calculated using the perturbations of the nominal values; and using the dose prediction models and based on the dose-volume histograms to generate an objective for the radiation treatment plan.

[0015] In an embodiment, nominal values of parameters associated with previously defined radiation treatment plans and perturbations of the nominal values are accessed. For each treatment field of those treatment plans, a field-specific planned target volume (fsPTV) is determined based on the perturbations. At least one clinical target volume (CTV) and at least one organ at risk (OAR) volume are also delineated. Each OAR includes at least one sub-volume delineated based on a spatial relationship between the each OAR and the CTV and the fsPTV for the each treatment field. Dose distributions for the sub-volumes are determined based on the nominal values and based on the perturbations. For example, if there are N perturbations, N+1 dose distributions are determined: a dose distribution based on the nominal values, and a dose distribution based on each of the perturbations.

[0016] In embodiments, multiple robust dose prediction models are generated for each sub-volume. For example, if there are N perturbations, N+1 dose prediction models are generated for each sub-volume: a dose prediction model trained using the dose distribution based on the nominal values, and a dose prediction model trained using each dose distribution based on each perturbation.

[0017] In other embodiments, only one robust dose prediction model is generated for each sub-volume. However, this model is trained using all dose distributions. For example, if there are N perturbations, the dose prediction model is trained using the N+1 dose distributions described above.

[0018] Thus, in embodiments according to the present application, robust dose prediction models are generated that are trained for potential perturbations in the planning parameters during treatment and thus take these potential perturbations into account.

[0019] In embodiments where multiple robust dose prediction models are generated for each sub-volume, when applying the models to a radiation treatment plan, the perturbations are inserted as parameters into the models to develop optimization objectives. More specifically, the models are used to predict dose-volume histograms (DVHs) for the plan. If there are N perturbations and thus N+1 models, a set of N+1 DVHs per sub-volume is determined for the plan. The dose prediction models can then be used to generate optimization objectives based on the DVHs.

[0020] In one aspect of the present application, a computer-implemented method is provided, comprising: accessing a plurality of treatment plans, the plurality of treatment plans having perturbations of nominal values of parameters of the treatment plans associated with the plurality of treatment plans;

[0021] determining, for each treatment field of the treatment plan, a field-specific planning target volume (fsPTV) based on the perturbations;

[0022] accessing information delineating at least one clinical target volume (CTV) and information delineating at least one organ at risk (OAR) volume, wherein each of the OAR volumes comprises at least one sub-volume delineated based on a spatial relationship between each of the OAR volumes and the CTV and the fsPTV for the each treatment field;

[0023] determining a plurality of dose distributions, the plurality of dose distributions comprising a dose distribution based on the nominal values and a dose distribution based on the perturbations; and generating, for each of the sub-volumes, a dose prediction model trained using the plurality of dose distributions.

[0024] The dose prediction model can comprise information associating the dose prediction model with the dose distributions used to train the dose prediction model.

[0025] The method can further include applying the dose prediction model to the radiation treatment plan, wherein the applying includes: calculating a dose-volume histogram for the radiation treatment plan using the dose prediction model and using the nominal values; and generating an objective for the radiation treatment plan using the dose prediction model and based on the dose-volume histogram.

[0026] In embodiments where only a single robust dose prediction model is generated for each sub-volume, the model is used to predict the nominal DVH for each sub-volume. The dose prediction model can then be used to generate optimization objectives based on the nominal DVH.

[0027] According to embodiments of the present invention, the radiation treatment plan is improved by increasing the robustness of the dose prediction model, and thus the treatment itself, which in turn is used to develop and optimize the radiation treatment plan for the patient being treated. The robust model can be used to deliver sufficient dose to the CTV in perturbed treatment scenarios, while ensuring that the OAR does not receive too much dose in any perturbed scenario. The robust model is particularly important for proton beam treatment, but is also important for other modalities.

[0028] Those skilled in the art will realize, after reading the following detailed description, that these and other objects and advantages of embodiments according to the present invention are achieved, as illustrated in the various figures.

[0029] This Summary is provided to introduce some concepts in a simplified form that are further described in the detailed description below. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in limiting the scope of the claimed subject matter. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings, which are incorporated in and form a part of the specification, illustrate embodiments of the present disclosure and, together with the detailed description, serve to explain the principles of the present disclosure.

[0031] Figure 1A is a flowchart of a computer-implemented method of generating and applying a dose prediction model in embodiments according to the present invention.

[0032] Figure 1A is a block diagram illustrating elements in a method of generating and applying a dose prediction model in embodiments according to the present invention.

[0033] Figure 2 is a flowchart of a computer-implemented method of generating a new and robust dose prediction model in embodiments according to the present invention.

[0034] Figure 3 shows sub-volumes used when generating a new and robust dose prediction model in embodiments according to the present invention.

[0035] Figure 4 is a flowchart of a computer-implemented method of generating a new and robust dose prediction model in accordance with other embodiments of the application.

[0036] Figure 5 is a flowchart of a computer-implemented method of applying a new and robust dose prediction model in accordance with embodiments of the application.

[0037] Figure 6 is a block diagram of an example of a computer system on which embodiments described herein can be implemented. DETAILED DESCRIPTION

[0038] Reference will now be made in detail to various embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. While the disclosure will be described in conjunction with these embodiments, it is understood that they are not intended to limit the disclosure to these embodiments. On the contrary, the disclosure is intended to cover alternatives, modifications, and equivalents, which can be included within the spirit and scope of the disclosure as defined by the appended claims. Furthermore, in the following detailed description of the present disclosure, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be recognized that the present disclosure can be practiced without these specific details. In other instances, well-known methods, procedures, components, and circuits have not been described in detail as not to unnecessarily obscure aspects of the present disclosure.

[0039] Some portions of the detailed description that follows are presented in terms of procedures, logic blocks, processing, and other symbolic representations of operations on data bits within a computer memory. These descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. In this application, a procedure, logic block, process, or the like, is conceived to be a self-consistent sequence of steps or instructions leading to a desired result. The steps are those utilizing physical manipulations of physical quantities. Usually, though not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared, and otherwise manipulated in a computer system. It has proven convenient at times, principally for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, samples, pixels, or the like.

[0040] It should be borne in mind, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless specifically stated otherwise as apparent from the following discussion, it is appreciated that throughout the present disclosure, discussions utilizing terms such as "determining", "accessing", "generating", "applying", "representing", "indicating", "storing", "using", "adjusting", "comprising", "counting", "calculating", "associating", or the like, refer to the action and processes of a computer system, or similar electronic computing device or processor (e.g., a Figure 6actions and processes of the computer system 600) are described in terms of computer-executable instructions. After realizing the embodiments, those skilled in the art will appreciate that those computer-executable instructions, and the like, can be loaded onto one or more computers or other devices, including a computer system 600, from a computer-readable storage medium or communication medium, so as to implement various aspects of the embodiments as discussed in the above description. Figure 1A , 2 , 4, and 5). The computer system or similar electronic computing device manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system memories, registers or other such information storage, transmission or display devices. Terms such as "dose," "dose rate," or some other parameter or attribute generally refer to a dose value, dose rate value, attribute value, or parameter value, respectively; the use of these terms will become clear from the context of the surrounding discussion.

[0041] The embodiments described herein can be discussed in the general context of computer-executable instructions residing on some form of computer- readable storage medium, such as program modules, executed by one or more computers or other devices. By way of example, and not limitation, computer-readable storage media can comprise non-transitory computer storage media and communication media. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform particular tasks or implement particular abstract data types. The functionality of the program modules can be combined or distributed as desired in various embodiments.

[0042] Computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read only memory (ROM), electrically erasable programmable ROM (EEPROM), flash memory or other memory technology, compact disc ROM (CD-ROM), digital versatile disks (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed to retrieve that information.

[0043] Communication media can embody computer-executable instructions, data structures, and program modules, and includes any information delivery media. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, radio frequency (RF), infrared, and other wireless media. Combinations of the any of the above can also be included within the scope of computer readable media.

[0044] Some portions of the detailed description are presented in terms of methods and procedures, such as the steps and ordering of acts of the flowcharts of FIGS. Figure 1A , 2 , 4, and 5). The steps and ordering of acts disclosed in the flowcharts of the figures describing these methods are merely examples. Embodiments are well suited to performing various other steps or variations of the steps recited in the flowcharts of the figures, and in the sequence presented, other sequences can be utilized within the scope of the embodiments.

[0045] Figure 1A 、 2 , 4 and 5 are flowcharts 100, 200, 400 and 500, respectively, of examples of computer-implemented operations for generating a dose prediction model or applying such a model to a radiation treatment plan in embodiments according to the present application. The flowcharts 100, 200, 400 and 500 can be implemented as computer-executable instructions (e.g., of the models 650 and 651) residing in the memory of a computer system 600 (e.g., of FIG. 1) on some form of computer- readable storage medium. Figure 6 Figure 6

[0046] Figure 1A is a flowchart 100 providing an overview of a computer-implemented method of generating and applying a dose prediction model in embodiments according to the present application. Reference is also made to Figure 1B discussed in Figure 1A , Figure 1B is a block diagram showing elements in the process of Figure 1A .

[0047] In block 102, a set of previously defined treatment plans 110 are accessed from a treatment plan database (e.g., knowledge-based database 112).

[0048] In block 104, information in those treatment plans is selected, such as information for each organ at risk (OAR) and dose volume histogram (DVH), for training a new dose prediction model.

[0049] In block 106, a new dose prediction model 114 is trained. In embodiments, the training process includes identifying acceptable clinical objectives, including acceptable tradeoffs between target dose coverage and dose to organs at risk (OARs). The training process can also include calculating DVHs using the trained model, which can be compared to DVHs in the treatment plan training set.

[0050] In block 108, once the prediction model is trained, it can be added to the database 112 or another database, and can also be used to predict DVHs and doses for treatment plans 116 being developed for radiation treatment of a patient. The dose prediction model can be used to develop treatment plans for (but not limited to) intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT).

[0051] Figure 2 is a flowchart 200 of a computer-implemented method of generating a new and robust dose prediction model in embodiments according to the present application. Reference is also made to Figure 3 discussed in Figure 2 , Figure 3 ​​The subvolumes used when generating the dose prediction model are shown.

[0052] exist Figure 2 In box 202, a set of previously defined treatment plans (which may be referred to herein as the training set) is selected and accessed from, for example, a database. New dose prediction models are generated for specific anatomical regions (e.g., head and neck, or trunk, etc.), and the training set will include treatment plans for the same anatomical regions. In embodiments, 20 or more treatment plans are accessed. Generally, the number of plans accessed is sufficient to properly train and generate a robust dose prediction model. Therefore, depending on the results of the training process, the number of plans selected may be more or less than 20. The number of plans selected may also depend on the anatomical region being modeled.

[0053] The selected treatment plan includes nominal values ​​for parameters that influence the treatment dose to be delivered. These nominal values ​​for the treatment plan may be stored in the plan itself or linked to a plan. Generally, the nominal values ​​for the training set are accessed in box 202. These parameters may include, for example, patient movement (displacement of the isocenter of the treatment field relative to the patient's position), beam calibration (e.g., changes in the computed tomography (CT) calibration curve), field size, beam attenuation, and imaging during planning and treatment.

[0054] Also in box 202, multiple perturbations to the nominal value (e.g., uncertainty, tolerance, range) are defined or accessed. Ideally, although the invention is not limited thereto, the perturbations are the same as those used to train a set of previously defined processing plans. The perturbations of the processing plans may be stored in the plan itself or linked to the plan.

[0055] Also in box 202, and refer to Figure 3 Define at least one OAR volume 302 and define at least one clinical target volume (CTV) 304. CTV 304 includes the target volume (e.g., the volume of the tumor being treated) and also takes into account uncertainties in the boundaries of the target volume (e.g., possible unimaged tumor spread).

[0056] exist Figure 2 In box 204, for each processing field (e.g., for a predefined set of processing plans) Figure 3 For fields 310 and 312), the perturbation-based field-specific planned target volume (fsPTV) is determined. The planned target volume (PTV) includes the CTV and also takes into account the perturbation. fsPTV is a separate PTV for each processing field.

[0057] In block 206, in embodiments, a geometry-based expected dose (GED) is computed using the fsPTV for each treatment field as the target volume. In essence, the GED provides an estimate of the dose distribution. The GED provides a measure that maps the treatment beam and patient geometry to a DVH.

[0058] In embodiments, a nominal GED is computed using the nominal values of the parameters and using the fsPTV for each treatment field as the target volume. The nominal GED is then used to estimate the GED corresponding to perturbations from the nominal values. More specifically, the nominal GED can be shifted laterally with respect to the treatment beam according to isocenter shifts caused by perturbations (e.g., patient movement), or the nominal GED can be shifted towards or away from the beam source due to, for example, perturbations associated with calibration curves affecting beam range or with the CT image.

[0059] In block 208, (from block 202) the information delineating the at least one CTV 304 and delineating the at least one OAR volume 302 is accessed, and each OAR volume is divided into one or more sub-volumes, the sub-volumes delineated based on spatial relationships between each OAR volume and the CTV and the fsPTV for each treatment field.

[0060] More specifically, in embodiments, the sub-volumes include: an in-field fsPTV region that includes at least a portion of the OAR volume overlapping a projection of the fsPTV from the at least one treatment field; an in-field CTV region that includes at least a portion of the OAR volume overlapping a projection of the CTV from the at least one treatment field; an overlapping fsPTV region that includes at least a portion of the OAR volume within the union of all fsPTVs; and an overlapping CTV region that includes at least a portion of the OAR volume inside the interior of any CTV(s). The term “projection” is a term of art and can be defined differently for different treatment modalities (e.g., beam types). For example, for a photon beam, it is the projection of the target treatment field in the beam direction, extending through the target volume. As another example, for a proton beam, it is the projection of the treatment field closest to the target volume (between the beam source and the target volume, but not extending beyond the target volume).

[0061] In general, in embodiments, Figure 3 the in-field region 306 of the OAR volume 302 is divided into an in-field fsPTV region that includes at least a portion of the OAR volume overlapping a projection of the fsPTV from the at least one treatment field and an in-field CTV region that includes at least a portion of the OAR volume overlapping a projection of the CTV from the at least one treatment field; and Figure 3The off-field region 308 of the OAR 308 is divided into an overlapping fsPTV region and an overlapping CTV region, the overlapping fsPTV region including at least a portion of the OAR volume within the union of all fsPTVs, the overlapping CTV region including at least a portion of the OAR volume inside any CTV(s).

[0062] In Figure 2 In block 210, data is extracted from the treatment plans in the training set. More specifically, a dose distribution is determined for each sub-volume identified in block 206. The dose distribution per sub-volume includes a dose distribution based on the nominal value and a dose distribution based on the perturbation. For example, if there are N perturbations, N+1 dose distributions per sub-volume are determined for each treatment plan: one dose distribution based on the nominal value and one dose distribution based on each perturbation.

[0063] In block 212 (also during the data extraction phase), the dose prediction models are applied to the dose distributions in the training set. Figure 2 In an embodiment of the method, a plurality of dose prediction models is generated for each sub-volume. The models include a model trained using the dose distribution based on the nominal value and a model trained using the dose distribution based on the perturbation. Thus, for example, if there are N perturbations, N+1 dose prediction models are generated for each sub-volume, one model for each perturbation.

[0064] In an embodiment, each dose prediction model includes information associating the model with the dose distribution used to train the model. Thus, for example, each dose prediction model has information identifying the perturbation associated therewith.

[0065] Figure 4 is a flowchart 400 of a computer-implemented method of generating new and robust dose prediction models in accordance with other embodiments of the application. The above description of Figure 2 blocks 202, 204, 206, 208, and 212 of the method of Figure 4 is described above. In the flowchart 400, block 412 replaces block 212 of the method of Figure 2 .

[0066] In Figure 4 In block 412 (during the data extraction phase), a single dose prediction model is generated for each sub-volume. However, the model is trained using all dose distributions. For example, if there are N perturbations plus the nominal value, N+1 dose distributions are used to train the dose prediction model. In an embodiment, the dose prediction model includes information associating the model with the dose distributions used to train the model.

[0067] Figure 5 is a flowchart 500 of a computer-implemented method of applying new and robust dose prediction models in accordance with embodiments of the application.

[0068] In block 502, a proposed radiation treatment plan is accessed.

[0069] In block 504, in embodiments where multiple robust dose prediction models are generated for each sub-volume, multiple DVHs are computed for the radiation treatment plan. One DVH is computed using the nominal values, and multiple DVHs are computed using the multiple perturbations.

[0070] More specifically, in embodiments where multiple robust dose prediction models are generated for each sub-volume, when the models are applied to the proposed radiation treatment plan, the perturbations are inserted as parameters into the models to develop optimization objectives. If there are N perturbations and thus N+1 models, a set of N+1 DVHs is determined for each sub-volume for the plan.

[0071] In block 506, the dose prediction models are used to generate optimization objectives based on the DVHs. For example, an optimization objective can be that the maximum dose to an OAR cannot exceed a certain value in any perturbed scenario.

[0072] In block 508, in embodiments where only a single robust dose prediction model is generated for each sub-volume, the model is used to compute a nominal DVH per sub-volume.

[0073] In block 510, the dose prediction model is used to generate optimization objectives based on the nominal DVH.

[0074] Figure 6 A block diagram of an example of a computer system 600 upon which embodiments described herein can be implemented is shown. In its most basic configuration, the system 600 includes at least one processing unit 602 and memory 604. This most basic configuration is illustrated in Figure 6 by dashed line 606. The system 600 can also have additional features and / or functionality. For example, the system 600 can also include additional storage such as removable storage and / or non-removable storage including, but not limited to, magnetic or optical disks or tape. Such additional storage is illustrated in Figure 6 by removable storage 608 and non-removable storage 620. The system 600 can also contain a communication connection 622 that allows devices to communicate over a network, such as a logical connection to one or more remote computers.

[0075] The system 600 also includes input devices 624 such as a keyboard, mouse, pen, voice input device, touch input device, etc. Output devices 626 such as a display device, speakers, a printer, etc. are also included.

[0076] In Figure 6In the example of FIG. 6A, the memory 604 includes computer-readable instructions, data structures, program modules, etc. associated with a model 650 for generating dose prediction models and a model 651 for applying dose prediction models to radiation treatment plans, as described above in connection with 1A, 2, 4, and 5. However, in contrast, the models 650 and 651 can reside in any one of the computer storage media used by the system 600, or can be distributed across some combination of computer storage media, or can be distributed across some combination of networked computers.

[0077] Thus, as described above, embodiments according to the present application provide an improved method of radiation treatment planning. More specifically, embodiments according to the present application relate to the generation of robust dose prediction models, and to the application of those models to develop and optimize radiation treatment plans. The robust dose prediction models are trained and thus take into account potential perturbations in the planning parameters during treatment.

[0078] Embodiments according to the present application improve radiation treatment planning by increasing the robustness of the dose prediction models, which in turn are used to develop and optimize radiation treatment plans for a patient being treated. Robust models can be used to deliver sufficient dose to the CTV under perturbed treatment scenarios, while ensuring that the OARs do not receive too much dose under any perturbed scenario. Robust models are particularly important for proton beam treatment, but are also important for other modalities.

[0079] The dose prediction models are trained using a training set comprising a plurality of treatment plans, where the plurality of parameters include nominal values and perturbations of the nominal values. For each plan, a plurality of fsPTV volumes are determined, and structures in the treatment region are divided into a plurality of sub-volumes. The dose distribution is determined taking into account all of these factors. Thus, developing one or more dose prediction models is a complex task that is beyond the capabilities of a human and relies on the use of a computing system.

[0080] Applying the dose prediction models to a proposed radiation treatment plan can also be a complex task. For example, depending on the modality of treatment, the degrees of freedom available include beam shaping (collimation), beam weighting (spot scanning), beam intensity or energy, beam direction, dose rate, and number and placement of spots. Parameters such as those mentioned earlier herein that affect the dose rate are also taken into account. If the target volume is divided into sub-volumes or voxels, the parameter values can be per sub-volume or per voxel (e.g., a value per sub-volume or voxel). Thus, consistently and efficiently generating and evaluating high-quality treatment plans is beyond the capabilities of a human and relies on the use of a computing system, particularly in light of the time constraints associated with using radiation therapy to treat diseases such as cancer, and in light of the large number of patients undergoing or needing to undergo radiation therapy during any given time period.

[0081] Although the subject matter has been described in language specific to structural features and / or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.

Claims

1. A computer-implemented method comprising: accessing a plurality of treatment plans having associated therewith perturbations of nominal values of parameters of the treatment plans; determining, for each treatment field of the treatment plans, a field-specific plan target volume fsPTV based on the perturbations; accessing information delineating at least one clinical target volume CTV and information delineating at least one organ at risk OAR volume, wherein each of the OAR volumes comprises at least one sub-volume delineated based on a spatial relationship between each of the OAR volumes and the CTV and the fsPTV for the each treatment field; and one of: (a) determining a dose distribution for each of the sub-volumes, wherein the dose distribution comprises a dose distribution based on the nominal values and a dose distribution based on the perturbations; and generating a dose prediction model for each of the sub-volumes, wherein the model comprises a model trained using the dose distribution based on the nominal values and a model trained using the dose distribution based on the perturbations; (b) determining a plurality of dose distributions comprising a dose distribution based on the nominal values and a dose distribution based on the perturbations; and generating a dose prediction model for each of the sub-volumes, the model trained using the plurality of dose distributions.

2. The method of claim 1, comprising step (a), wherein each of the dose prediction models comprises information associating the each model with a dose distribution used to train the each model.

3. The method of claim 1 or 2, comprising step (a), further comprising applying the dose prediction models to a radiation treatment plan.

4. The method of claim 3, wherein the applying comprises: calculating a dose-volume histogram for the radiation treatment plan using the dose prediction models, wherein the dose-volume histogram comprises a dose-volume histogram calculated using the nominal values and a dose-volume histogram calculated using the perturbations of the nominal values; and generating an objective for the radiation treatment plan using the dose prediction models and based on the dose-volume histogram.

5. The method of claim 1, comprising step (b), wherein the dose prediction model comprises information associating the dose prediction model with the dose distributions used to train the dose prediction model.

6. The method of claim 1 or 5, comprising step (b), further comprising applying the dose prediction model to a radiation treatment plan.

7. The method of claim 6, wherein the applying comprises: calculating a dose-volume histogram for the radiation treatment plan using the dose prediction model and using the nominal values; and generating an objective for the radiation treatment plan using the dose prediction model and based on the dose-volume histogram.

8. The method of any of claims 1-2, 4-5, and 7, wherein the at least one sub-volume comprises a sub-volume selected from the group consisting of: an in-field fsPTV region comprising at least a portion of an OAR volume overlapping a projection of a fsPTV from at least one of the treatment fields; an in-field CTV region comprising at least a portion of an OAR volume overlapping a projection of a CTV from at least one of the treatment fields; an overlapping fsPTV region comprising at least a portion of an OAR volume within a union of all fsPTVs; and an overlapping CTV region comprising at least a portion of an OAR volume inside any of the at least one CTV.

9. The method of any of claims 1-2, 4-5, and 7, wherein the determining comprises computing a geometry-based expected dose (GED).

10. The method of claim 9, wherein the GED is computed using the fsPTV for the each treatment field as the target volume.

11. The method of claim 9, wherein the computing comprises: computing a nominal GED using the nominal values of the parameters and using the fsPTV for the each treatment field as the target volume; and estimating a GED corresponding to the perturbation of the nominal values using the nominal GED.

12. A computer-implemented method comprising: accessing a radiation treatment plan; computing dose-volume histograms for the radiation treatment plan using a plurality of dose prediction models, wherein the dose prediction models have associated therewith nominal values of parameters of the radiation treatment plan and perturbations of the nominal values, and wherein the computing comprises: computing a dose-volume histogram using the nominal values; and computing a dose-volume histogram using the perturbations; and generating an objective for the radiation treatment plan using the plurality of dose prediction models and based on the plurality of dose-volume histograms for the radiation treatment plan.

13. The method of claim 12, wherein the computing a dose-volume histogram using the nominal values comprises using a dose prediction model trained with the nominal values.

14. The method of claim 12 or 13, wherein the computing a dose-volume histogram using the perturbations comprises using a dose prediction model trained with the perturbations.

15. The method of claim 12 or 13, wherein each model of the plurality of dose prediction models comprises information associating the each model with a dose distribution used to train the each model. ​

Images