Methods and apparatuses to facilitate administration of therapeutic radiation to a patient

Abstract

The control circuit accesses (201, 203) patient image content and field geometry information about a particular radiation treatment platform. The control circuit then generates (205) a predicted three-dimensional dose map for a radiation treatment plan as a function of both the patient image content and the field geometry information.

Description

Technical Field

[0001] These teachings generally involve irradiating the patient's planned target volume according to the radiotherapy plan, and more specifically involve predictive dose maps corresponding to the radiotherapy plan. Background Technology

[0002] The use of radiation to treat medical conditions encompasses the known areas of current technological efforts. For example, radiation therapy is a crucial component of many treatment programs aimed at reducing or eliminating unwanted tumors. Unfortunately, the radiation applied does not inherently distinguish between unwanted substances and adjacent tissues, organs, etc., which are desired or even essential for the patient's continued survival. As a result, radiation is typically administered in a carefully managed manner to at least attempt to confine it to a given target volume. So-called radiation therapy planning generally operates in this regard.

[0003] Radiation therapy planning typically involves specifying values ​​for each of various treatment platform parameters for each period across multiple consecutive fields. Treatment plans for radiation therapy courses are usually generated through a process known as optimization. As used herein, "optimization" will be understood as improving candidate treatment plans without necessarily ensuring that the result of optimization is actually a single optimal solution. Such optimization typically involves automatically adjusting one or more treatment parameters (usually while adhering to one or more corresponding constraints of these aspects) and mathematically calculating possible corresponding treatment outcomes to identify a given set of treatment parameters that represents a good trade-off between desired therapeutic outcomes and avoiding undesirable indirect effects.

[0004] Recent advances in radiotherapy planning have improved the overall quality of planning and ultimately led to better patient outcomes. Unfortunately, these advances have also increased the complexity of treatment planning and the time required to develop a radiotherapy plan. Obtaining the best plan for a given patient can rely heavily on the planner's expertise and often requires several iterative interactions between the planner and the oncologist. To reduce both the time required for planning and the variation in the quality of treatment plans, some existing technological approaches seek to automate at least a portion of the planning process.

[0005] Such attempts at automation include using artificial intelligence to perform tasks such as organ segmentation, tumor identification, and three-dimensional dose prediction. Three-dimensional dose prediction refers to predicting the likely radiation dose to occur at various locations within the planned target volume and / or at one or more organs at risk of the patient when treating a patient according to a given radiotherapy plan. Unfortunately, training AI models can be very challenging when using heterogeneous patient data that varies in the location, shape, and size of the planned treatment volume, as well as the type of treatment (e.g., lateral / full arc, coplanar / non-coplanar, etc.). Variations in field geometry pose an even greater challenge when performing three-dimensional dose prediction on a single two-dimensional slice at a time. Summary of the Invention

[0006] In one aspect, the invention provides an apparatus as defined in claim 1 that facilitates the provision of radiotherapy planning for the administration of therapeutic radiation to a patient via a specific radiotherapy platform. Optional features are specified in the claims dependent on claim 1.

[0007] Devices that facilitate the provision of radiotherapy planning for the administration of therapeutic radiation to a patient via a specific radiotherapy platform may include a memory containing image content about the patient and field geometry information about the specific radiotherapy platform stored therein.

[0008] According to another aspect of the invention, a method is provided for facilitating the provision of radiotherapy planning for the administration of therapeutic radiation to a patient via a specific radiotherapy platform, as defined in claim 11. Optional features are specified in the claims dependent on claim 11. In one arrangement, the method further includes administering therapeutic radiation to the patient via a specific radiotherapy platform. In another arrangement, the claimed method does not include administering therapeutic radiation to the patient via a specific radiotherapy platform.

[0009] Methods for facilitating the provision of radiotherapy planning to deliver therapeutic radiation to a patient via a specific radiotherapy platform may include providing a memory containing image content about the patient and field geometry information about the specific radiotherapy platform stored therein.

[0010] The method may include operating the device for which protection is sought. Attached Figure Description

[0011] The above needs are at least partially met by providing methods and equipment for administering therapeutic radiation to patients, as described in the following detailed description, particularly when studied in conjunction with the accompanying drawings, in which:

[0012] Figure 1 Including block diagrams configured according to various embodiments of these teachings;

[0013] Figure 2 Including flowcharts of various embodiments configured according to these teachings; and

[0014] Figure 3 This includes neural network processing views configured according to various embodiments of these teachings.

[0015] The elements in the accompanying drawings are shown for simplicity and clarity and are not necessarily drawn to scale. For example, the size and / or relative positioning of some elements in the drawings may be enlarged relative to other elements to aid in understanding the various embodiments of the invention. Furthermore, common but well-known elements that are useful or necessary in commercially viable embodiments are generally not depicted to facilitate less obstructed observation of these various embodiments of the teachings. Certain actions and / or steps may be described or depicted in a particular order of occurrence, and those skilled in the art will understand that such specificity regarding the order is not actually necessary. The terms and expressions used herein have the ordinary technical meaning consistent with those given by those skilled in the art, unless otherwise set forth in this document. Unless otherwise specifically indicated, the word “or” as used herein should be interpreted as having a separate structure rather than a connected structure. Detailed Implementation

[0016] Generally, these various embodiments facilitate the provision of radiotherapy plans for administering therapeutic radiation to a patient via a specific radiotherapy platform by generating a predicted three-dimensional dose map for the radiotherapy plan. The dose map can then be used in various ways to compare and / or review a given radiotherapy plan to aid in evaluation (as one example), such as whether the plan is ready for use when administering therapeutic radiation to a patient. As another example, such dose predictions can be used to check whether an optimized dose is entirely different from the dose expected based on historical processing data of the corresponding application settings. (Since the dose prediction may be based only on a portion of the planning information, this teaching can be useful and applicable even when the entire set of planning information is not yet available.)

[0017] One method involves using control circuitry to access image content about the patient and field geometry information about a specific radiotherapy platform. The control circuitry then generates a predicted three-dimensional dose map for radiotherapy planning based on both the image content about the patient and the field geometry information.

[0018] These teachings will be adapted to a variety of image content concerning patients. Examples include, but are not limited to, at least one organ mask and at least one computed tomography image. As used herein, it will be understood that one or more organ masks may include, for example, a contoured planned target volume mask (to accommodate, for example, a tumor located in a region between organs or a tumor extending across several organs) and / or at least one contoured mask of an organ at risk.

[0019] In one method, the aforementioned field geometry information includes at least one image that graphically represents at least a portion of the field geometry information. In other words, when generating a predicted three-dimensional dose map, at least some of the field geometry information is provided and used as an image. In another method, the field geometry information encoded as an image is encoded as an image with the same resolution as, for example, a computed tomography image including at least a portion of image content about a patient. These teachings can be adapted to other methods in these aspects if needed. As an example, the field geometry information may be encoded as a vector.

[0020] One method involves a control circuit configured to generate a predicted 3D dose map by feeding image content and field geometry information about the patient as input to a convolutional neural network model. The convolutional neural network model processes the field geometry information along with the image content about the patient to generate the predicted 3D dose map. Alternatively, in this case, the image content and field geometry information about the patient can be fed as input to the convolutional neural network model as a stack of two-dimensional images.

[0021] The latter may include providing a stack of two-dimensional images via corresponding channels. As illustrative examples of these aspects, these channels may include at least in part a computed tomography image channel, a contoured planned target volume image channel, a contoured organ at risk image channel, and a field geometry information channel.

[0022] Because this type of 3D dose prediction model receives field geometry information as input in addition to patient-based images, it can be more easily trained on heterogeneous datasets that exhibit variations in the location, size, and shape of the planned target volume, thus overcoming significant technical limitations of various existing efforts to characterize these aspects. This adaptability can, in turn, greatly reduce planning time and / or minimize computational requirements to achieve useful results over a specific time period.

[0023] Furthermore, those skilled in the art will understand that such three-dimensional dose prediction models can be trained using both coplanar and non-coplanar treatment plans. As a result, these teachings can support faster comparisons than usual between three-dimensional dose maps predicted with different field geometries, and thus help planners select the most appropriate field geometry for a given patient's radiotherapy plan.

[0024] These and other benefits will become clearer after a comprehensive review and study of the following detailed explanation. Refer now to the accompanying drawings, especially... Figure 1 Now we will present an illustrative device 100 that is compatible with many of these teachings.

[0025] In this particular example, enabling device 100 includes control circuitry 101. As a “circuit,” control circuitry 101 therefore includes a structure comprising at least one (and typically many) conductive paths (such as paths made of conductive metals such as copper or silver) that transmit electricity in an ordered manner. These paths (one or more) typically also include corresponding electrical components (including, where appropriate, passive (such as resistors and capacitors) and active (such as any of a variety of semiconductor-based devices)) to allow the circuitry to implement the control aspects of these teachings.

[0026] This type of control circuit 101 may include a fixed-purpose hardwired hardware platform (including, but not limited to, application-specific integrated circuits (ASICs) (which are integrated circuits designed for a specific purpose rather than for general use), field-programmable gate arrays (FPGAs), etc.), or may include partially or fully programmable hardware platforms (including, but not limited to, microcontrollers, microprocessors, etc.). These architectural choices of such structures are well known and understood in the art and need not be further described herein. This control circuit 101 is configured (e.g., by using corresponding programming that will be well understood by those skilled in the art) to perform one or more of the steps, actions, and / or functions described herein.

[0027] Control circuitry 101 is operatively coupled to memory 102. Memory 102 may be integrated into control circuitry 101 or physically separated from control circuitry 101 (whole or part) as needed. Memory 102 may also be local to control circuitry 101 (where, for example, both share a common circuit board, chassis, power supply, and / or housing), or may be partially or entirely remote to control circuitry 101 (where, for example, memory 102 is physically located in another facility, a large metropolitan area, or even a country compared to control circuitry 101).

[0028] In addition to the imaging information and field geometry information described above, the memory 102 can, for example, be used to non-transiently store computer instructions that, when executed by the control circuit 101, cause the control circuit 101 to operate as described herein. (As used herein, this reference to "non-transient" will be understood to mean the non-transient state of the stored content (and thus excludes the case where the stored content constitutes only a signal or wave) rather than the volatility of the storage medium itself, and therefore includes both non-volatile memories (such as read-only memory (ROM)) and volatile memories (such as dynamic random access memory (DRAM)).

[0029] In this example, control circuitry 101 may also be operatively coupled to user interface 103. User interface 103 may include any of a variety of user input mechanisms (such as, but not limited to, keyboards and keypads, cursor control devices, touch-sensitive displays, voice recognition interfaces, gesture recognition interfaces, etc.) and / or user output mechanisms (such as, but not limited to, visual displays, audio transducers, printers, etc.) to facilitate receiving information and / or instructions from the user and / or providing information to the user.

[0030] If necessary, the control circuitry 101 can also be operatively coupled to a network interface (not shown). With this configuration, the control circuitry 101 can communicate with other components (both within and outside the device 100) via the network interface. Network interfaces, including both wireless and non-wireless platforms, are well known in the art and require no particular description herein.

[0031] Some or all of the patient-related imaging information described herein can be obtained by a method such as a computed tomography device 106 and / or other imaging device 107 known in the art.

[0032] In this illustrative example, control circuitry 101 is configured to ultimately output an optimized radiotherapy plan 113. This radiotherapy plan 113 typically includes specified values ​​for each of various treatment platform parameters during each of multiple consecutive fields. In this case, the radiotherapy plan 113 is generated through an optimization process. Various automated optimization processes specifically configured to generate such radiotherapy plans are known in the art. Since this teaching is not overly sensitive to any particular choice of these aspects, further elaboration of these aspects is not provided herein except where particularly relevant to the details of this specification.

[0033] In one method, control circuitry 101 is operatively coupled to a radiotherapy platform 114, which is configured to deliver therapeutic radiation 112 to a corresponding patient 104 according to an optimized radiotherapy plan 113. These teachings are generally applicable to any of a wide variety of radiotherapy platforms. In a typical application setup, the radiotherapy platform 114 will include a radiation source 115. The radiation source 115 may include, for example, an X-ray source based on a radio frequency (RF) linear particle accelerator (a linear accelerator (linac)), such as the Varian Linatron M9. A linear accelerator is a particle accelerator that greatly increases the kinetic energy of charged subatomic particles or ions by subjecting charged particles to a series of oscillating potentials along a linear beamline, which can be used to generate ionizing radiation (e.g., X-rays) 116 and high-energy electrons. A typical radiotherapy platform 114 may also include one or more support devices 110 (such as recliners) for supporting the patient 104 during treatment, one or more patient fixation devices 111, a gantry or other movable mechanism that allows selective movement of the radiation source 115, and one or more beamforming devices 117 (such as clamps, multi-leaf collimators, etc.) for providing selective beamforming and / or beam modulation as needed. Since the foregoing elements and systems are well known in the art, detailed descriptions of these aspects are not provided herein except where relevant to the description.

[0034] Now for reference Figure 2 Now, we will present, for example, a process 200 that can be executed by the aforementioned control circuit 101.

[0035] In block 201, control circuitry 101 accesses the aforementioned memory 102, thereby accessing image content 202 concerning the patient. (Those skilled in the art will understand that during the model training phase, a large set of possible heterogeneous image contents from different patients with tumors of different sizes and locations will be accessed and utilized. The described process assumes the use of a trained model.) The image content 202 may, as needed, include image content provided by the aforementioned CT device 106 and / or the aforementioned imaging device 107. In many application settings, it would be beneficial for image content 202 to include one or more computed tomography images, one or more contoured planned target volume images, and one or more contoured images of organs at risk. (Those skilled in the art will know and understand that important volumes, such as the planned target volume of a patient and organs at risk, have their respective visual identifiers surrounding them during the planning process to produce so-called contoured images. By a method, these teachings will be adapted to present contoured planned target volumes (one or more) and organs at risk (one or more) using corresponding organ masks.)

[0036] In block 203, control circuitry 101 accesses the aforementioned memory 102, thereby also accessing field geometry information 204 specific to a particular radiotherapy platform (e.g., platform 114 generally described above). Examples of field geometry information 204 include, for example, the position of the aforementioned radiation source 115 relative to the gantry and / or some patient reference points such as isogonal points corresponding to the planned treatment volume. Other examples include, but are not limited to, gantry angles, collimator angles, clamp positions, treatment bed angles, etc.

[0037] Based on these teachings, at least a portion of the field geometry information is encoded as an image, thus including at least one image describing the corresponding field geometry information. By a method, all provided field geometry information 204 includes one or more such images. In many application settings, it may be advantageous to encode the field geometry information as an image with the same resolution as the computed tomography image provided as part of the image content 202 described above.

[0038] It can be noted that, through one method, the above-mentioned access is to different types of patient data for various patients, where the data shows variations in the location, shape, and size of the planned treatment volume, as well as the type of treatment.

[0039] In block 205, control circuitry 101 then generates a predicted three-dimensional dose map for a specific radiotherapy plan based on both image content 202 and field geometry information 204 about the patient. (As used herein, the term "predicted" in conjunction with the expression "three-dimensional dose map" will be understood to refer to a three-dimensional radiation dose map that is predicted to result when the patient 104 is processed according to the field geometry structure associated with the specific radiotherapy plan.) Therefore, the predicted three-dimensional dose map will provide a predicted level of radiation dose at various locations within the relevant planned treatment volume and organs at risk. This may include, for example, indicating the spatial distribution of varying levels of radiation dose given to such patient structures. In the present case, it is assumed that the dose map quantitatively and carefully identifies different levels of radiation dose within segmented, identified patient structures and throughout the entire structure. This can be accomplished using any visually discretized method, such as by using different colors for different levels of radiation dose and / or isodose lines, as known in the art.

[0040] If necessary, these instructions will be further adapted to present some or all of the predicted three-dimensional dose maps via the aforementioned user interface 103 and / or to present two or more predicted three-dimensional dose maps for different radiotherapy plans in a comparative manner via the user interface 103, so as to facilitate comparison by technicians, oncologists, etc. Of course, these instructions will also be adapted to the use of vector radiotherapy planning with a radiotherapy platform to administer therapeutic radiation to patients according to that plan.

[0041] One method allows control circuit 101 to be configured to use deep learning to generate a predicted three-dimensional dose map for radiotherapy planning. Deep learning (sometimes also called hierarchical learning, deep neural learning, or deep structured learning) is generally defined as a subset of machine learning in artificial intelligence, featuring networks capable of unsupervised learning from unstructured or unlabeled data. That is, deep learning can also be supervised or semi-supervised, if needed. Deep learning architectures include deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks.

[0042] For the purposes of illustrative purposes and not intended to impose any particular limitations on these aspects, it will be assumed here that the control circuit 101 is configured as a convolutional neural network model that receives the aforementioned image content about the patient and field geometry information as input, and processes the field geometry information together with the image content about the patient to generate a predicted three-dimensional dose map. Figure 3 General illustrative examples 300 are provided for these aspects.

[0043] In one method, control circuitry 101 is configured to provide both image content about the patient and field geometry information as input to a convolutional neural network model, as a stack of two-dimensional images and via corresponding channels. (For example, these channels may include at least in part a computed tomography image channel, a contoured planned target volume image channel, a contoured organ at risk image channel, and a field geometry information channel.)

[0044] In this example 300, the input to the convolutional neural network model consists of a stack of two-dimensional images (sometimes referred to in the art as cubes). Each two-dimensional image contains several channels corresponding to the aforementioned computed tomography images (one or more) and a mask for planning the processing volume and organs at risk (one or more). Additional channels of each image carry field geometry information encoded to have the same shape and resolution as the computed tomography images.

[0045] In this illustrative example 300, the network architecture is similar to U-net. U-net is a convolutional neural network developed for biomedical image segmentation. Such networks are based on fully convolutional networks, with their architecture modified and extended to provide accurate segmentation with fewer training images. It should be understood that these teachings are not limited to use with U-net; rather, any convolutional neural network capable of analyzing 3D image data may work. However, in this example 300, the architecture deviates from the traditional U-net, at least because it employs residual blocks at each level. As detailed by reference numeral 301, each residual block consists of a stack of two convolutional layers. The output of each residual block is the sum of the input layer and the second convolution.

[0046] The input stack of the convolutional neural network model has a shape of N×256×256×Ch, where N refers to the number of slices, each with a Ch channel of size 256×256 pixels. (As used in this paper, a "slice" refers to a set containing a CT image, a target image, an image of an organ at risk, and an image with field geometry information.) Here, it is assumed that the first (ch-2) channel contains the set of organ masks, the penultimate channel contains the corresponding computed tomography image, and the last channel includes the field geometry information image. (Note that the order in which these channels are provided is not necessarily important. However, it is useful to provide this information as input to the network.)

[0047] In this illustrative example 300, the network output comprises a stack of shape (n-2)×256×256×1. This stack represents the dose prediction corresponding to the input image stack, except for the first and last slices.

[0048] One method allows for the grouping of organs at risk masks according to their importance level in the clinic where the procedure is performed. The images of organs at risk can be represented as binary masks, where a value of 1 corresponds to a location belonging to a given organ, and a value of 0 corresponds to a location outside the organ.

[0049] One method, for a planned processing volume mask, allows for the use of scaled dose level values ​​for locations corresponding to a given planned processing volume, and a value of 0 for any location outside the planned processing volume. For example, a mask corresponding to a planned processing volume receiving the highest dose level (“PTV_high”) could have a value of 1 for the corresponding pixel located inside an organ. For a mask corresponding to a planned processing volume receiving the next dose level (“PTV_int”), the ratio between the “PTV_high” and “PTV_int” dose levels of pixels located inside that organ could be used.

[0050] As a very specific example provided for illustrative purposes, in the simplest case of intensity-modulated radiotherapy (IMRT) treatment, the field geometry image may consist only of a set of rays starting from an isogonal point and corresponding to a set of field angles for a particular patient. In a more complex representation, the field can be illustrated as a cone beam starting from an effective point source located in a rotating gantry and covering the target. As another example, in volume-modulated arc therapy (VMAT) treatment, the field geometry image may consist of both the corresponding arc and the intensity level.

[0051] These teachings are highly flexible in practice and will be adapted to any modifications to the foregoing. As an example of these aspects, in addition to a stack containing computed tomography images and planning processing volumes and organ masks, field geometry information can be provided to the network as a separate stack. This configuration would allow the two stacks to be processed in parallel by the first few layers of the network. The extracted features can then be pulled together (e.g., by juxtaposition) and further processed by the rest of the network.

[0052] As another example, when training a model using coplanar field geometry, an alternative way to present field geometry information to a convolutional neural network model is as a 360-dimensional vector. For IMRT processing, non-zero values ​​can be set at locations corresponding to gantry angles, and zero values ​​elsewhere. For VMAT processing, non-zero blocks corresponding to processing arcs can be present. This vector can be merged with low-level features extracted from the image cube, such as at the bottom layer of the aforementioned U-net, and this union will be processed together by the rest of the network.

[0053] Because this method for predicting 3D dose maps receives field geometry information as input, it can be successfully and even efficiently trained on heterogeneous datasets that exhibit variations in the location, size, and shape of the planned treatment volume. These differences, in turn, lead to the use of different treatment field geometry types (e.g., lateral or full-arc treatments) and different treatment locations. Furthermore, the described prediction model can be trained using coplanar and non-coplanar treatment planning. Therefore, these teachings can be used for rapid comparisons between 3D dose maps predicted with different field geometries, and thus help planners select the most appropriate field geometry for a given patient using a specific radiotherapy platform.

[0054] Of particular note is that when training and / or using dose prediction models, these teachings do not require defining a complete radiotherapy plan (e.g., a plan specifying a complete leaf sequence for an expected multi-leaf collimator).

[0055] Those skilled in the art will recognize that various modifications, alterations, and combinations can be made to the above embodiments without departing from the scope of the invention, and such modifications, alterations, and combinations will be considered to be within the scope of the concept of the invention.

Claims

1. An apparatus for facilitating the provision of radiotherapy planning for the administration of therapeutic radiation to a patient via a specific radiotherapy platform, the apparatus comprising: The memory is configured to receive image content about the patient and field geometry information about the specific radiotherapy platform; Control circuitry, operably coupled to the memory, is configured to: - Access the image content related to the patient; - Access the field geometry information; - A predicted three-dimensional dose map is generated based on both the image content and the field geometry information of the patient. The predicted three-dimensional dose map is predicted to obtain a result when the patient is processed according to the field geometry associated with the radiotherapy plan. The predicted three-dimensional dose map quantitatively and carefully identifies different levels of radiation dose within the segmented, identified patient structure and throughout the entire patient structure.

2. The device of claim 1, wherein the image content relating to the patient at least partially comprises at least one organ mask and at least one computed tomography image.

3. The device of claim 2, wherein the at least one organ mask at least partially comprises a contoured planned target volume and at least one contoured organ at risk.

4. The device according to claim 1, 2 or 3, wherein the field geometry information comprises at least one image depicting at least a portion of the field geometry information.

5. The device of claim 1, 2 or 3, wherein the control circuitry is configured to generate the predicted three-dimensional dose map by providing the image content about the patient and the field geometry information as input to a convolutional neural network model, the convolutional neural network model processing the field geometry information together with the image content about the patient to generate the predicted three-dimensional dose map.

6. The device of claim 5, wherein the control circuitry is configured to provide the image content and the field geometry information about the patient as input to the convolutional neural network model as a two-dimensional image stack.

7. The device of claim 6, wherein the control circuitry is configured to provide the two-dimensional image stack via a corresponding channel.

8. The device of claim 7, wherein the channel at least partially comprises: Computed tomography (CT) image channels; Outlined planning of target volume image channels; Outlined images of organs at risk; as well as Field geometry information channel.

9. The device of claim 8, wherein the field geometry information provided via the field geometry information channel is encoded as an image.

10. The apparatus of claim 9, wherein the field geometry information encoded as an image is encoded as an image having the same resolution as the computed tomography image.

11. A method for facilitating the provision of radiotherapy planning for the administration of therapeutic radiation to a patient via a specific radiotherapy platform, the method comprising: A memory is provided, the memory being configured to receive image content about the patient and field geometry information about the specific radiotherapy platform; as well as Through control circuitry operably coupled to the memory: Access the image content related to the patient; Access the field geometry information; as well as A predicted three-dimensional dose map is generated based on both the image content and the field geometry information of the patient. The predicted three-dimensional dose map is predicted to obtain a result when the patient is processed according to the field geometry associated with the radiotherapy plan. The predicted three-dimensional dose map quantitatively and carefully identifies different levels of radiation dose within the segmented, identified patient structure and throughout the entire patient structure.

12. The method of claim 11, wherein the image content relating to the patient comprises at least in part at least one organ mask and at least one computed tomography image.

13. The method of claim 12, wherein the at least one organ mask at least partially comprises a contoured planned target volume and at least one contoured organ at risk.

14. The method of claim 11, 12 or 13, wherein the field geometry information comprises at least one image depicting at least a portion of the field geometry information.

15. The method of claim 11, 12, or 13, wherein generating the predicted three-dimensional dose map comprises providing the image content about the patient and the field geometry information as input to a convolutional neural network model, the convolutional neural network model processing the field geometry information together with the image content about the patient to generate the predicted three-dimensional dose map.

16. The method of claim 15, wherein providing the image content and field geometry information about the patient as input to the convolutional neural network model comprises providing the image content and field geometry information about the patient as input to the convolutional neural network model as a two-dimensional image stack.

17. The method of claim 16, wherein providing the two-dimensional image stack comprises providing the two-dimensional image stack via corresponding channels.

18. The method of claim 17, wherein the channel at least partially comprises: Computed tomography (CT) image channels; Outlined planning of target volume image channels; Outlined images of organs at risk; as well as Field geometry information channel.

19. The method of claim 18, wherein the field geometry information provided via the field geometry information channel is encoded as an image.

20. The method of claim 19, wherein the field geometry information encoded as an image is encoded as an image having the same resolution as the computed tomography image.

Images