Neural network training and use of radiotherapy platform machine control points

Abstract

Embodiments of the present disclosure relate to neural network training and use of radiotherapy platform machine control points. Comprising physical settings for a particular radiotherapy platform, such as a machine control point, forms a training corpus. A neural network is trained using the training corpus. By one approach, the training corpus does not include any information related to any radiation therapy platform other than a particular radiation therapy platform. The training may be repeated as a function of at least one of a lapse of time, a particular number of therapeutic uses of a particular radiotherapy platform, and / or completion of at least one maintenance activity. The radiotherapy plan may be optimized to provide an optimized plan control point. The optimized plan control points may be mapped to corresponding machine control points based at least in part on the foregoing trained neural network.

Description

Technical Field

[0001] This teaching generally involves neural networks, and also involves using energy to treat a patient’s planned target area according to an optimized energy-based treatment plan. Background Technology

[0002] A neural network is a computational model inspired by the structure and function of the human brain, designed to recognize patterns and solve complex problems. It consists of interconnected units called neurons, which are organized in layers: an input layer that receives data, one or more hidden layers that perform computations, and an output layer that delivers the final result. Each neuron applies a mathematical transformation to its received input and forwards the result through the network. The strength of the connections between neurons (called weights) is adjusted during the training process using a training corpus to allow the network to learn by iteratively improving its predictions.

[0003] The use of energy to treat medical diseases 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 harmful tumors. Unfortunately, the applied energy does not inherently distinguish between harmful substances and adjacent tissues, organs, etc., which are desirable or even essential for the patient's continued survival. Consequently, energy, such as radiation, is typically applied in a carefully controlled manner to at least attempt to confine the energy to a given target area. So-called radiation therapy programs often serve this purpose.

[0004] A radiotherapy plan typically includes specified values ​​for each of the various treatment platform parameters during each of multiple sequential fields. Treatment plans used in radiotherapy procedures are often generated automatically through a process known as optimization. As used herein, "optimization" will be understood as improving candidate treatment plans, without necessarily ensuring that the optimized result is actually a single optimal solution. Such optimization often involves automatically adjusting one or more physical therapy parameters (often while observing one or more corresponding limitations in these aspects) and mathematically calculating the corresponding possible treatment outcomes (such as dose levels) to identify a given set of treatment parameters that represents a good trade-off between desired therapeutic outcomes and the avoidance of undesirable side effects. Attached Figure Description

[0005] By providing the development of a neural network-based training corpus, the corresponding neural network training, and the use of machine control points for the radiotherapy platform as described in the following detailed description, various needs are at least partially met, especially when studied in conjunction with the accompanying drawings, wherein:

[0006] Figure 1 Includes block diagrams configured according to various embodiments of this teaching;

[0007] Figure 2 Including flowcharts configured according to various embodiments of this teaching; and

[0008] Figure 3 This includes schematic representations configured according to various embodiments of this teaching.

[0009] The elements in the figures 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 figures may be exaggerated relative to other elements to aid in understanding the various embodiments of this teaching. Furthermore, common but well-understood elements that are useful or necessary in commercially viable embodiments are often not depicted to facilitate a clearer view of these various embodiments of this teaching. 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 order is not actually necessary. The terms and expressions used herein have the same general technical meaning as those used by those skilled in the art set forth above, unless otherwise set forth herein with a different specific meaning. The word “or” as used herein should be interpreted as having a separating construction rather than a connecting construction, unless otherwise specifically indicated. Detailed Implementation

[0010] An optimized radiotherapy plan can be defined, at least in part, by planning control points. These control points are often presented in DICOM format. DICOM stands for Digital Imaging and Communications in Medicine, and is a standard for storing, transmitting, retrieving, processing, printing, and displaying medical imaging information. DICOM files are represented by the .dcm extension and the information that makes up the dataset.

[0011] One method transforms the aforementioned planned control points from the radiotherapy platform into timing trajectories or machine control points. Typically, understanding the machine control point parameters is important for estimating delivery time, determining the acceleration / deceleration of the radiotherapy platform's mechanical axes, and adjusting the delivery rate of the monitor unit.

[0012] Accurate estimation of machine control point parameters is a challenging task, typically relying on a fast-responding, accurate machine simulator that takes into account the higher-order gradients of all movement axes. At least one problem in this regard is that the actual delivery of the optimized radiotherapy plan requires continuous monitoring of machine performance during the delivery of the machine control point, and not all mechanical characteristics of the machine are usually considered during the creation of the machine control point. Furthermore, the foregoing is subject to change due to aging and / or maintenance practices of the drive mechanisms, as well as events. As a result, the simulator may fail to accurately predict how the radiotherapy plan will actually be executed. When actual machine performance deviates from expectations, one or more relevant mechanical axes may deviate from their planned positions. While these deviations are often monitored to prevent exceeding set tolerances, such deviations can still lead to inaccuracies in treatment.

[0013] In some existing methods, machine controllers use linear approximators to estimate velocity changes in the upcoming milliseconds during the treatment phase (i.e., during the delivery of the optimized radiotherapy plan by the radiotherapy platform). Unfortunately, this linear modeling of velocity changes does not account for higher-order derivatives of the motion of the mechanical axes. For example, the third and fourth derivatives are also important in these aspects in addition to acceleration and velocity. When an understanding of treatment delivery time is desired, the result is often calculated based on simplified models in the treatment planning system, and thus suffers from inherent errors due to at least the aforementioned problems of such models.

[0014] Generally, this teaching is intended to at least partially solve at least some of the aforementioned problems. According to these various embodiments, a neural network training corpus can be developed and then used to train the neural network. Optimized plan control points can then be mapped to corresponding machine control points, at least in part, based on this trained neural network.

[0015] This teaching provides a method for accessing information, including physical settings for a specific radiotherapy platform, to form a training corpus. The aforementioned physical settings may include, for example, machine control points, such as, but not limited to, gantry angles, multi-leaf collimator positions, dose rates, and patient support positions.

[0016] The neural network can then be trained using this training corpus. In one method, the training corpus does not include any information related to any radiotherapy platform other than the specific radiotherapy platform. In another method, the aforementioned neural network may include a recurrent neural network. Training itself may include, for example, supervised training. Training can be repeated based on at least one of the following: the passage of time, a specific number of therapeutic uses of the specific radiotherapy platform, and / or the completion of at least one maintenance activity.

[0017] This teaching will be applied to optimizing a radiotherapy plan for a specific patient using the aforementioned specific radiotherapy platform, wherein optimization includes optimizing the plan control points to provide optimized plan control points. This teaching will then be applied to mapping the optimized plan control points to corresponding machine control points, at least in part, based on the aforementioned trained neural network.

[0018] This teaching can be further applied to the output of estimated machine control points from a trained neural network prior to administering therapeutic radiation to a specific patient using a specific radiotherapy platform. In this case, and during the administration of therapeutic radiation to the specific patient during the treatment phase, this teaching can be used to compare the estimated machine control points with real-time physical parameter values ​​of the specific radiotherapy platform, and, upon detecting a triggering difference, replace the corresponding machine control points in the estimated machine control points with at least some of the real-time physical parameter values ​​while continuing to administer therapeutic radiation to the specific patient.

[0019] With this configuration, this teaching can support, facilitate, and / or enable model-based predictions of actual treatment delivery parameters, as well as provide estimates of treatment delivery time.

[0020] These and other benefits become clearer after a thorough review and study of the following detailed description. Refer now to the accompanying drawings, and specifically to... Figure 1 First, an illustrative device 100 compatible with many of the teachings in this teaching will be presented.

[0021] In this particular example, the enabling device 100 includes a control circuit 101. As a “circuit”, the structure included by the control circuit 101 thus includes at least one (typically many) conductive path (such as a path made of a conductive metal (such as copper or silver)) that transmits power in an ordered manner, and this path will also typically include corresponding electrical components (both passive (such as resistors and capacitors) and active (such as any of various semiconductor-based devices, as appropriate)) to allow the circuit to implement the control aspects of this teaching.

[0022] Such control circuitry 101 may include a fixed-purpose hardwired hardware platform (including, but not limited to, application-specific integrated circuits (ASICs) (i.e., integrated circuits designed for a specific purpose rather than for general purposes), field-programmable gate arrays (FPGAs), etc.), or may include a partially or fully programmable hardware platform (including, but not limited to, microcontrollers, microprocessors, etc.). These architectural choices for such a configuration are well known and fully understood in the art and need not be further described herein. The control circuitry 101 is configured (e.g., by using corresponding programming that will be fully understood by those skilled in the art) to perform one or more steps, actions, and / or functions described herein.

[0023] It will be understood that the control circuit 101 may include a single integrated platform, or may include multiple such circuits that work together with each other.

[0024] Control circuitry 101 is operatively coupled to memory 102. Depending on the requirements, memory 102 may be integrated with or physically separate from control circuitry 101 ( wholly or partially). 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 fully remote to control circuitry 101 (where, for example, memory 102 is physically located in another facility, metropolitan area, or even country compared to control circuitry 101). Like control circuitry 101, memory 102 may comprise a single structure or may comprise multiple memory platforms collectively constituting the “memory” of device 100.

[0025] In addition to information such as optimization information for a specific patient and information about a specific radiotherapy platform, as described herein, the memory 102 can be used, for example, to non-transitory store computer instructions that, when executed by the control circuitry 101, cause the control circuitry 101 to operate as described herein. (As used herein, the reference to "non-transitory" will be understood to mean the non-transitory state of the stored content (thus excluding the case where the stored content merely constitutes a signal or wave), rather than the volatility of the storage medium itself, and therefore includes both non-volatile memory (such as read-only memory (ROM)) and volatile memory (such as dynamic random access memory (DRAM)).)

[0026] Alternatively, the control circuitry 101 may also be operatively coupled to the user interface 103. The user interface 103 may include any of a variety of user input mechanisms (such as, but not limited to, keyboards and keypads, cursor controllers, touch-sensitive displays, voice recognition interfaces, gesture recognition interfaces, etc.) and / or user output mechanisms (such as, but not limited to, visual displays, audio converters, printers, etc.) to facilitate receiving information and / or instructions from and / or providing information to the user.

[0027] If needed, 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 internal to the device 100 and external to the device 100) via the network interface. Network interfaces (including both wireless and non-wireless platforms) are well understood in the art and do not require particular detail here.

[0028] By means of a method, a computed tomography device 106 and / or other imaging device 107 known in the art can acquire some or all of any desired patient-related imaging information.

[0029] In this illustrative example, control circuitry 101 is configured to ultimately output an optimized energy-based treatment plan (e.g., an optimized radiotherapy plan 113). This energy-based treatment plan typically includes specified values ​​for each of a variety of treatment platform parameters during each of a plurality of sequential exposure fields. In this case, the energy-based treatment plan is generated through an optimization process, examples of which are further provided herein.

[0030] In one method, control circuitry 101 can be operatively coupled to an energy-based treatment platform 114, which is configured to direct energy towards a target area 105 and one or more organs at risk (in accordance with an optimized energy-based treatment plan 113). Figure 1 Therapeutic energy 112 is delivered to the corresponding patient 104 (represented by the first risk organ 108 to the nth risk organ 109). This teaching is generally applicable to any of a wide variety of energy-based therapeutic platforms / devices. In a typical application setting, the energy-based therapeutic platform 114 will include an energy source, such as a radiation source 115 of ionizing radiation 116.

[0031] In one method, the radiation source 115 can be selectively moved along an arcuate pathway via a gantry (wherein the pathway at least partially encompasses the patient during the administration of treatment). The arcuate pathway may, as needed, comprise a complete or near-complete circle. In another method, control circuitry 101 controls the movement of the radiation source 115 along the arcuate pathway and can accordingly control when the radiation source 115 begins to move, stops moving, accelerates, decelerates, and / or the speed at which the radiation source 115 travels along the arcuate pathway.

[0032] As an illustrative example, radiation source 115 may include, for example, an X-ray source based on a radio frequency (RF) linear particle accelerator (linac-based). A linac 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.

[0033] A typical energy-based treatment platform 114 may also include one or more support devices 110 (such as examination tables) for supporting the patient 104 during treatment, one or more patient fixation devices 111, a gantry or other movable mechanism for allowing selective movement of the radiation source 115, and one or more energy-shaping devices (e.g., beamforming devices 117, such as jaws, multi-leaf collimators, etc.) for providing selective energy shaping and / or energy modulation as needed.

[0034] In a typical application setting, it is assumed here that the patient support device 110 can be selectively controlled by the control circuitry 101 to move in any direction (i.e., any X, Y, or Z direction) during an energy-based treatment session. Since the foregoing elements and systems are well understood in the art, further details in these respects are not provided herein, except where otherwise relevant to the description.

[0035] Now for reference Figure 2 The following describes a process 200 that can be implemented, for example, in conjunction with the application setup described above (more specifically, via the aforementioned control circuit 101). Generally, this process 200 is used to facilitate the generation of an optimized radiotherapy plan 113, thereby facilitating the treatment of a specific patient with therapeutic radiation according to the optimized radiotherapy plan, using a specific radiotherapy platform.

[0036] At block 201, process 200 is used to access information including physical settings for a specific radiotherapy platform to form a training corpus. These physical settings may include, for example, machine control points. Generally, machine control points refer to specific parameters or settings on a radiotherapy platform that are adjusted to control the delivery of a prescribed radiation dose to the patient's tumor while minimizing exposure to surrounding healthy tissue. These machine control points include variables such as beam energy, intensity, shape, and orientation, which are configured based on a treatment plan developed for the patient. Examples of specific machine control points include: gantry angles, which determine the direction of radiation beam delivery; multi-leaf collimator positions, which shape the beam to conform to the geometry of the tumor and / or modulate the beam; dose rates, which affect the intensity of radiation delivered per unit time; and one or more patient support positions, which adjust the patient's positioning relative to the beam. By precisely controlling these points, the radiotherapy machine can deliver treatment according to a complex plan to target the tumor with high precision while protecting healthy tissue.

[0037] In one method, the accessed information includes at least two of the aforementioned illustrative machine control points. In another method, the accessed physical settings include prior physical settings for a particular radiotherapy platform, such as prior machine control points. In yet another method, the resulting training corpus does not include any information related to any radiotherapy platform other than the specific radiotherapy platform.

[0038] At block 202, control circuitry 101 trains a neural network (e.g., a recurrent neural network) using the aforementioned training corpus, thereby providing a trained neural network. This training may include (if desired) supervised training of the neural network. As an illustrative example in these respects, for some or all of the treated patients, a clinic or other medical system may store previously optimized control points (OCPs) and their corresponding machine control points (MCPs) in DICOM format and machine log files (e.g., using .txt format), respectively. In supervised learning, this historical data may be divided into two sets: a training set and a test set. The neural network can be trained on the training set to learn the mapping from the optimized control points to the machine control points (i.e., MCP = f(OCP)). Periodically, the quality of the mapping f(.) can be tested on the test set, which has not been previously revealed to the neural network. Training can continue until the neural network can accurately estimate the machine control points based on the optimized control points given in the test set.

[0039] In addition to the initial training, this teaching will apply to retraining a neural network based on at least one of the following: the passage of time (such as a specific number of days, weeks, or months), therapeutic use of a specific radiotherapy platform (such as a specific number of therapeutic uses that have occurred), and / or the completion of at least one maintenance activity (such as the replacement and / or adjustment or recalibration of one or more components of the radiotherapy platform (such as one or more motors).

[0040] The final results of blocks 201 and 202 include a trained neural network, specifically, a neural network that has been trained relative to the physical setup (such as machine control points) for a particular radiotherapy platform.

[0041] At optional block 203, control circuit 101 can then optimize a radiotherapy plan for a specific patient using a specific radiotherapy platform, wherein optimization includes optimizing the plan control point to provide an optimized plan control point.

[0042] At optional block 204, the control circuit can then map those optimized planned control points to corresponding machine control points, at least in part, based on the trained neural network. Specifically, using the planned control points as input, the trained neural network can output the corresponding machine control points.

[0043] At optional block 205, control circuit 101 can output estimated machine control points from a trained neural network before administering therapeutic radiation to a specific patient using a specific radiotherapy platform.

[0044] Then, at block 206, and during the administration of therapeutic radiation to a specific patient during the treatment process, control circuit 101 can compare the aforementioned estimated machine control point with the real-time physical parameter values ​​of the specific radiotherapy platform, and when a trigger difference is detected (e.g., when the estimated position of the multi-leaf collimator blade deviates from the real-time position of the blade), replace the corresponding machine control point in the estimated machine control point with at least some of the real-time physical parameter values, while continuing to administer therapeutic radiation to the specific patient.

[0045] With this configuration, this teaching can employ a deep neural network model as a function approximator to map between optimized planning control points and machine control points for a specific radiotherapy platform, so that the machine control points better (or best) match the actual machine performance.

[0046] The dimensionality of the problem makes a simple recurrent neural network architecture that trains quickly sufficient for accurate estimation in many application settings. With this configuration, treatment clinics can periodically (e.g., monthly) retrain the model to ensure it accounts for variations in the mechanical characteristics of the radiotherapy platform. Recurrent neural networks can capture the velocity dynamics (such as higher-order gradients) of machine control point sequences via recurrent connections, where the recurrent connections are expanded across time steps (or sequence steps) while applying the same underlying parameters to each step.

[0047] While standard connections can be applied synchronously to propagate activations from each layer to the next layer at the same time step, cyclic connections can be dynamic to pass information across adjacent time steps. This approach allows the model to estimate the parameters of the current machine control point while taking into account the higher-order derivatives of the machine axes at previous and upcoming control points.

[0048] Further details consistent with this teaching will now be presented. It will be understood that the specific details of these examples are intended for illustrative purposes and not to impose any particular limitation on this teaching.

[0049] Figure 3 A schematic representation of a recurrent neural network 300 configured according to this teaching is presented. A recurrent neural network can be thought of as a feedforward neural network, where the parameters of each layer (both regular and recurrent parameters) are shared across time steps. In this example, training such a model involves collecting real machine control point parameters from the logs of the radiotherapy platform after each treatment. The accumulated parameters can represent any point in several to hundreds of treatments performed by the same platform.

[0050] These parameters are then used as a baseline (label) in the supervised training of the model. The input to the recurrent neural network includes (or only (if needed) optimized planning control point (PCP) parameters.

[0051] The resulting trained model can be used to accurately estimate machine control point (MCP) parameters just before treatment is administered to a specific patient. The correctness of the parameter estimation can be verified based on the real-time position and velocity of the movement axis at the treatment platform. In the event of a discrepancy, the estimated parameters can be swapped with the corresponding real-time values ​​to prevent error propagation to the upcoming machine control point estimation.

[0052] It will be understood that this teaching can also be applied to situations where multiple similar radiotherapy platforms are similarly tuned so that they can be considered as the same group of machines with equivalent doses. A recurrent neural network model can be trained to satisfy constraints on the least capable machine in such a group, ensuring that treatment time estimation is effective for all machines in that group.

[0053] With this configuration, the teachings enable more accurate timing of treatment planning because the evaluation is based on realistic machine trajectories that take into account the actual performance of the machine. Another advantage is that, at least in part, recurrent neural networks tend to provide more accurate estimates of machine control point parameters compared to current methods, because they can capture the velocity dynamics (i.e., higher-order gradients) of the machine control point sequence via recurrent connections. This could allow, for example, the use of machine control points generated by trained neural networks to control a radiotherapy platform.

[0054] This teaching will also apply to combining the described model-based machine control point generation with more traditional methods to achieve improved performance. For example, one of the slower methods may be used and / or there may be a maximum permissible difference between the two methods.

[0055] Further aspects of the invention are provided by way of the subject matter of the following provisions (wherein it will be understood that any one of these provisions may be combined with any one or more other provisions as appropriate):

[0056] Clause 1: A method includes: via control circuitry: accessing information including physical setup for a specific radiotherapy platform to form a training corpus; and using the training corpus to train a neural network to provide the trained neural network.

[0057] Clause 2: The method described in Clause 1, wherein the physical setup includes machine control points.

[0058] Clause 3: The method described in Clause 2, wherein the machine control points include at least two of the following: gantry angle, multi-leaf collimator position, dose rate, and patient support position.

[0059] Clause 4: The method according to Clause 1, wherein training the neural network includes training the neural network based on at least one of the following: the passage of time, therapeutic use of a particular radiotherapy platform, and the completion of at least one maintenance activity.

[0060] Clause 5: The method described in Clause 1, wherein the training corpus does not include any information relating to any radiotherapy platform other than a specific radiotherapy platform.

[0061] Clause 6: The method described in Clause 1, wherein the neural network includes a recurrent neural network.

[0062] Clause 7: The method described in Clause 1, wherein training the neural network includes supervised training of the neural network.

[0063] Clause 8: The method described in Clause 1 further includes: optimizing a radiotherapy plan for a specific patient using a specific radiotherapy platform, wherein optimization includes optimizing the plan control points to provide optimized plan control points; and mapping the optimized plan control points to corresponding machine control points, at least in part, based on a trained neural network.

[0064] Clause 9: The method described in Clause 1 further includes: outputting estimated machine control points from a trained neural network before administering therapeutic radiation to a specific patient using a specific radiotherapy platform; comparing the estimated machine control points with real-time physical parameter values ​​of the specific radiotherapy platform during the administration of therapeutic radiation to the specific patient; and, upon detecting a trigger difference, replacing the corresponding machine control points in the estimated machine control points with at least some of the real-time physical parameter values ​​while continuing to administer therapeutic radiation to the specific patient.

[0065] Clause 10: An apparatus comprising: a control circuit configured as a neural network, the neural network having been trained with a training corpus including physical settings for a particular radiotherapy platform.

[0066] Clause 11: The apparatus according to Clause 10, wherein the physical setup includes a machine control point.

[0067] Clause 12: The apparatus according to Clause 11, wherein the machine control points include at least two of the following: gantry angle, multi-leaf collimator position, dose rate, and patient support position.

[0068] Clause 13: The apparatus according to Clause 10, wherein the neural network is repeatedly trained based on at least one of the following: the passage of time, therapeutic use of a particular radiotherapy platform, and the completion of at least one maintenance activity.

[0069] Clause 14: The apparatus described in Clause 10, wherein the training corpus does not include any information relating to any radiotherapy platform other than a specific radiotherapy platform.

[0070] Clause 15: The apparatus according to Clause 10, wherein the neural network includes a recurrent neural network.

[0071] Clause 16: The apparatus according to Clause 10, wherein the neural network is trained via supervised training.

[0072] Clause 17: The apparatus according to Clause 10, wherein the control circuitry is further configured to: optimize a radiotherapy plan for a particular patient using a specific radiotherapy platform, wherein optimization includes optimizing the plan control points to provide optimized plan control points; and mapping the optimized plan control points to corresponding machine control points, at least in part according to a neural network.

[0073] Clause 18: The apparatus according to Clause 10, wherein the control circuitry is further configured to: output an estimated machine control point from a neural network before administering therapeutic radiation to a specific patient using a specific radiotherapy platform; during the administration of therapeutic radiation to the specific patient, compare the estimated machine control point with real-time physical parameter values ​​of the specific radiotherapy platform, and, upon detecting a trigger difference, replace the corresponding machine control point in the estimated machine control point with at least some of the real-time physical parameter values ​​while continuing to administer therapeutic radiation to the specific patient.

[0074] Clause 19: A method comprising: optimizing a radiotherapy plan for a specific patient using a specific radiotherapy platform, wherein optimization includes optimizing the plan control points to provide optimized plan control points; mapping the optimized plan control points to corresponding machine control points, at least in part, based on a trained neural network, wherein the trained neural network is trained using a training corpus including prior physical settings for the specific radiotherapy platform.

[0075] Clause 20: The method described in accordance with Clause 19, wherein the prior physical setup includes the prior machine control point.

[0076] Those skilled in the art will recognize that various modifications, alterations, and combinations can be made with respect 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 inventive concept.

Claims

1. A method comprising: By controlling the circuit: Access includes information on the physical setup of a specific radiotherapy platform to form a training corpus; The neural network is trained using the training corpus to provide a trained neural network.

2. The method of claim 1, wherein the physical setup includes machine control points.

3. The method of claim 2, wherein the machine control point includes at least two of the following: gantry angle, multi-leaf collimator position, dose rate, and patient support position.

4. The method of claim 1, wherein training the neural network comprises training the neural network according to at least one of the following: the passage of time, therapeutic use of the particular radiotherapy platform, and completion of at least one maintenance activity.

5. The method of claim 1, wherein the training corpus does not include any information relating to any radiotherapy platform other than the specific radiotherapy platform.

6. The method of claim 1, wherein the neural network comprises a recurrent neural network.

7. The method of claim 1, wherein training the neural network includes supervised training of the neural network.

8. The method according to claim 1, further comprising: The specific radiotherapy platform is used to optimize radiotherapy plans for specific patients, wherein the optimization includes optimizing the plan control points to provide optimized plan control points; The optimized planned control points are mapped to corresponding machine control points, at least in part, based on the trained neural network.

9. The method according to claim 1, further comprising: Before administering therapeutic radiation to a specific patient using the specific radiotherapy platform, estimated machine control points are output from the trained neural network. During the administration of the therapeutic radiation to the specific patient, the estimated machine control point is compared with the real-time physical parameter values ​​of the specific radiotherapy platform, and when a trigger difference is detected, the corresponding machine control point in the estimated machine control point is replaced with at least some of the real-time physical parameter values, while the therapeutic radiation continues to be administered to the specific patient.

10. An apparatus comprising: A control circuit configured as a neural network, which has been trained using a training corpus that includes physical settings specific to a particular radiotherapy platform.

11. The apparatus of claim 10, wherein the physical setup includes a machine control point.

12. The apparatus of claim 11, wherein the machine control point includes at least two of the following: gantry angle, multi-leaf collimator position, dose rate, and patient support position.

13. The apparatus of claim 10, wherein the neural network is repeatedly trained according to at least one of the following: the passage of time, therapeutic use of the particular radiotherapy platform, and completion of at least one maintenance activity.

14. The apparatus of claim 10, wherein the training corpus does not include any information relating to any radiotherapy platform other than the specific radiotherapy platform.

15. The apparatus of claim 10, wherein the neural network comprises a recurrent neural network.

16. The apparatus of claim 10, wherein the neural network is trained via supervised training.

17. The apparatus of claim 10, wherein the control circuit is further configured to: The specific radiotherapy platform is used to optimize radiotherapy plans for specific patients, wherein the optimization includes optimizing the plan control points to provide optimized plan control points; The optimized planned control points are mapped to corresponding machine control points, at least in part, based on the neural network.

18. The apparatus of claim 10, wherein the control circuit is further configured to: Before administering therapeutic radiation to a specific patient using the specific radiotherapy platform, estimated machine control points are output from the neural network; During the administration of the therapeutic radiation to the specific patient, the estimated machine control point is compared with the real-time physical parameter values ​​of the specific radiotherapy platform, and when a trigger difference is detected, the corresponding machine control point in the estimated machine control point is replaced with at least some of the real-time physical parameter values, while the therapeutic radiation continues to be administered to the specific patient.

19. A method comprising: Optimization of radiotherapy plans for specific patients using a specific radiotherapy platform, wherein the optimization includes optimizing the plan control points to provide optimized plan control points; The optimized planned control points are mapped to corresponding machine control points, at least in part, based on a trained neural network trained using a training corpus including prior physical settings for the particular radiotherapy platform.

20. The method of claim 19, wherein the prior physical setup includes a prior machine control point.

Images