Two-step beam geometry optimization and non-isocenter beam incidence angle

Abstract

Embodiments of the present disclosure relate to two-step beam geometry optimization and non-isocenter beam incidence angles. Systems and methods are disclosed for optimizing a processing plan using all degrees of freedom, including degrees of freedom associated with beam geometry parameters, the optimization including the step of limiting a search space for the beam geometry parameters using a trained machine learning model, also disclosed are systems and methods for obtaining beam geometry parameters for processing plans that do not require decentration of the beam delivery device isocenter.

Description

Technical Field

[0001] The present disclosure generally relates to a treatment plan optimization process and generating a beam geometry (BG) solution for a treatment plan, and more particularly, to systems, methods and apparatus for automating a treatment plan optimization process with all degrees of freedom, including those related to beam geometry (BG) parameters, and to systems, methods and apparatus for generating a beam geometry (BG) solution without defining an isocenter of an external radiation beam delivery system. Background Art

[0002] Radiation therapy involves medical procedures that use external beam radiation to treat pathological anatomy (tumors, lesions, vascular malformations, neurological disorders, etc.) by delivering a prescribed dose of radiation (X-rays, gamma rays, electrons, protons, and / or ions) to the pathological anatomy while minimizing radiation exposure to surrounding tissue and critical anatomical structures.

[0003] In general, a complete radiotherapy planning and treatment workflow includes several phases: a treatment planning phase, a treatment delivery phase, and a monitoring and evaluation phase, in which treatment progress (e.g., dose accumulation) is monitored. The treatment delivery phase typically overlaps with the monitoring and evaluation phase. Treatment is typically delivered in many sessions, each lasting several weeks. Patients are monitored throughout treatment to assess treatment progress and whether treatment replanning or adaptation is needed.

[0004] In the treatment planning phase, an accurate three-dimensional (3D) map of the anatomical structures in the region of interest (head, body, etc.) is first constructed using any of the following techniques (or a combination thereof): computed tomography (CT), cone beam computed tomography (CBCT), magnetic resonance imaging (MRI), positron emission tomography (PET), 3D rotational angiography (3DRA), or ultrasound. This determines the exact coordinates of the target within the anatomical structure, i.e., the tumor or abnormality is located in the body and its exact shape and size are defined. On these images, the organs at risk (OARs) in the region of interest are also delineated. Next comes the prescription step, in which the level of radiation that should be delivered to the target (tumor) is specified, as well as the level of OAR protection that needs to be achieved to avoid side effects for the patient. In the prescription step, the movement path of the radiation beam is also calculated to deliver a dose distribution to the target within the treatment volume that the radiation oncologist deems acceptable, taking into account various medical constraints. A team of experts then uses special computer software to develop a treatment plan by designing the radiation beam to converge on the target area from different angles and planes to optimally irradiate the tumor and minimize the dose to the surrounding normal tissue.

[0005] In the treatment delivery phase, the radiation treatment plan is executed. In this phase, the radiation dose is delivered to the patient according to the prescribed treatment plan. Typically, the treatment plan is delivered to the patient through a series of radiation treatments (called fractions). However, there are many factors, such as different patient placement positions, changes that may occur in the case of tumor regression or weight loss during treatment, and uncertainties introduced by movement, which, for example, can cause differences between the prescribed radiation dose distribution and the actual delivered dose (i.e., the actual dose delivered to the target during radiation treatment). These anatomical and physiological changes can cause the target volume and the surrounding anatomical structures and organs to move and change size and shape during treatment. In this way, executing or continuing to execute the initial treatment plan may result in the actual received dose distribution being different from the planned distribution, thereby reducing the dose to the target volume and / or increasing the dose to the organs at risk (OAR). Therefore, in the treatment delivery phase, the treatment plan may adapt to the images of the day to better reflect the current situation. This involves modifying the initial treatment plan based on subsequently acquired image data to match the new location and shape of the target volume and surrounding anatomical structures.

[0006] Generating an optimal treatment plan (whether it is an initial treatment plan generated during the treatment planning phase or an adjusted plan generated during the treatment delivery phase of an adaptive treatment workflow) can be time consuming and cumbersome, particularly in the field of intensity modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) where complex matrix manipulations are required to generate and optimize treatment plans. The complexity and duration of the treatment planning phase also arises from the fact that the optimization step may need to be repeated multiple times after modifying one or more parameters including field geometry parameters (i.e., isocenter location, number of fields, field location, field size of VMAT arcs, etc.) and / or optimization parameters (i.e., objective, priority, etc.).

[0007] Optimization is an iterative process in which the user attempts to specify planning goals in the form of dose or biological objectives to create an ideal dose to the target (tumor) and minimize the dose to critical structures (i.e., organs at risk OARs). During the optimization process, the dose distribution within the treatment volume is evaluated at each iteration until a satisfactory dose distribution is achieved.

[0008] Typically, in IMRT and VMAT, creating an optimal treatment plan involves solving the inverse problem of determining the best combination of angles, radiation doses, and multileaf collimator (MLC) leaf movements to deliver the desired total radiation dose to the target while minimizing irradiation of healthy tissue. This can be simplified to an objective function that takes into account the relative importance of the various trade-offs inherent in such planning and the constraints that must be satisfied for the plan to be medically acceptable or physically feasible. To create a deliverable treatment plan, treatment planning algorithms must take into account the capabilities of the specific radiation therapy system they are used with.

[0009] Current optimization algorithms use dose-volume objectives to guide the search for system parameters, since the quality of the generated plan depends on the choice of these objectives. In general, the treatment planning algorithm proceeds by calculating the radiation dose received by each voxel in the treatment volume, adjusting one or more variable system parameters (such as the radiation angle or the position of the MLC blades), and then recalculating the dose received by each voxel. This process is performed iteratively until an optimized plan is reached.

[0010] However, the search space of system parameters is usually too large for optimizers to cope with. Therefore, in current IMRT and VMAT optimization algorithms, beam geometry (BG) (i.e., machine instructions such as gantry angles, collimator angles, and other variable system parameters associated with the radiotherapy system, excluding MLC leaf and secondary collimator positions) is either not considered as an optimization target (i.e., BG parameters are not optimized) or, when it is considered as an optimization target, the beam geometry produced by the optimization algorithm is unacceptable (i.e., the obtained BG parameters are too complex to be implemented in practical applications, or the BG parameters are clinically unacceptable).

[0011] For example, in typical IMRT and VMAT optimization algorithms, rather than iteratively searching for the best beam geometry, the beam geometry (BG) parameters are fixed prior to treatment plan optimization. This is typically accomplished by using a clinical template that specifies the BG parameters or by having the user manually enter the BG parameters. Since the BG parameters are fixed prior to the optimization process, there are no beam geometry-related quantities to optimize during the optimization process.

[0012] In a knowledge-based approach to the treatment planning process, machine learning or other inference methods are used to predict beam geometry based on patient geometry relative to previous similar patient geometries. However, the resulting beam geometry is one that may be useful for manual entry of user-selected beam geometry, rather than being optimized in an optimization process.

[0013] Among other approaches to treatment planning that include dose aspects, optimization algorithms that include beam trajectory or other beam geometry as optimization targets either require extensive manually coded rule sets to avoid obtaining beam geometries that are too complex or infeasible, or the optimization process takes much longer than standard IMRT optimization and is therefore not applicable, or the optimization process produces beam geometries that are clinically unacceptable.

[0014] In other approaches, where beam geometry (such as collimator angle) is considered the optimization target, the optimization algorithm does not include dose aspects.

[0015] Therefore, beam geometry (BG) parameters in current treatment planning optimization processes are not optimized to produce dose-optimal plans for individual patients.

[0016] Therefore, there is a need for a system and method for automated treatment plan optimization that can be performed with all degrees of freedom including those related to beam geometry (BG) parameters.

[0017] Furthermore, the current process for determining the beam geometry (BG) parameters to be used in treatment planning requires prior knowledge of the system isocenter location. Traditionally, the isocenter is fixed by a manual planner or, in some cases, by an automated algorithm. Only after the isocenter is set can the beam geometry (BG) parameters, especially the gantry angle, be determined by an automated solution. Therefore, the results of automated beam geometry setup have a strong dependency on the value of the isocenter, especially when using a machine learning model as an automated solution.

[0018] This traditional approach forces the evaluator of the automated beam geometry (BG) generation model to evaluate the radiation field setting values ​​relative to the system rather than relative to the patient. However, a patient-centric approach would be a more natural approach because it is consistent with the way treatment planners consider patient geometry when setting the radiation field. In current beam geometry (BG) parameter generation models, the dependence of automated beam geometry (BG) parameters on isocenters and their values ​​is complex and often unnecessarily narrow.

[0019]

[0006] Therefore, there is also a need for a system and method that allows for automated beam geometry (BG) parameter determination (ie, finding a BG solution) without explicitly determining the system isocenter location. Summary of the invention

[0020] Systems and methods are disclosed for automating a treatment planning optimization process that can be performed in all degrees of freedom including those related to beam geometry (BG) parameters.

[0021] In an embodiment, the treatment plan optimization process includes: a first step, which is configured to limit the search space of the optimization target of the optimization algorithm, the optimization target being the beam geometry (BG) parameters of the radiation beam delivery system; and a second step, which is configured to optimize the beam geometry (BG) parameters based on the limited search space obtained in the first step.

[0022] In an embodiment, the first step includes executing a trained machine learning model to predict an initial selection of beam geometry (BG) parameters based on patient geometry as input.

[0023] In an embodiment, the first step further comprises predicting a feasible range of beam geometry (BG) parameters and / or a maximum acceptable deviation from an initial selection of beam geometry (BG) parameters.

[0024] Additionally or alternatively, in an embodiment, the first step includes initially setting beam geometry (BG) parameters using a template and applying a rule set to define a search space around the template.

[0025] Additionally or alternatively, the first step includes manually entering an initial selection of beam geometry (BG) parameters and associated search boundaries.

[0026] In an embodiment, the second optimization step includes: receiving information about the desired dose distribution within the treatment volume of the patient and treatment parameters including beam geometry (BG) parameters as input to a treatment plan optimization algorithm; calculating the dose distribution within the treatment volume by executing the treatment plan optimization algorithm; determining whether the calculated dose distribution is within an acceptable threshold of the desired dose distribution; and iteratively modifying one or more treatment parameters including the beam geometry (BG) parameters until the calculated dose distribution is within the acceptable threshold or reaches an endpoint.

[0027] In an embodiment, beam geometry (BG) parameters are modified based on the search space obtained in the first step.

[0028] Also disclosed are systems and methods for obtaining beam geometry (BG) parameters for a treatment plan without explicitly determining the isocenter of a radiation beam delivery system that delivers the treatment plan to a target structure within a patient.

[0029] In an embodiment, the method comprises: determining a reference point within a target structure; obtaining a predicted beam incidence angle relative to the reference point by executing a trained machine learning model; displaying the predicted beam incidence angle superimposed on the patient geometry for evaluation; allowing a user to accept the predicted beam incidence angle, or interactively modify the predicted beam incidence angle until an acceptable beam incidence angle is obtained; and allowing the user to accept the reference point as the isocenter or manually enter the isocenter.

[0030] In an embodiment, when a reference point is accepted as the isocenter, the accepted beam incidence angle is selected as the beam geometry (BG) parameter, and when the manually entered isocenter is in close proximity to the reference point, the beam geometry (BG) parameter is calculated from the accepted beam incidence angle.

[0031] Also disclosed is a system comprising a computer processing device configured to execute sequences of programmed instructions embodied on a computer-readable storage medium, the execution of which causes the system to perform the method steps disclosed herein.

[0032] Also disclosed is a non-transitory computer-readable storage medium having a sequence of programming instructions embodied thereon, optimizing a radiation treatment plan by executing a first step configured to limit a search space of an optimization target of an optimization algorithm, the optimization target being a beam geometry (BG) parameter of a radiation beam delivery system, and executing a second step configured to optimize the beam geometry (BG) parameter based on the limited search space obtained in the first step, and a computer processing system that executes the sequence of programming instructions embodied on the computer-readable storage medium. Executing the sequence of programming instructions may cause the computer processing system to perform the treatment planning and optimization process described herein.

[0033] Also disclosed is a non-transitory computer readable storage medium having a sequence of programming instructions embodied thereon for obtaining beam geometry (BG) parameters for a treatment plan without explicitly determining an isocenter of a radiation beam delivery system that delivers the treatment plan to a target structure in a patient, and a computer processing system that executes the sequence of programming instructions embodied on the computer readable storage medium. Execution of the sequence of programming instructions may cause the computer processing system to perform an automated process to obtain a machine isocenter-independent beam geometry (BG) solution as described herein.

[0034] Objects and advantages of embodiments of the disclosed subject matter will become apparent from the following description considered in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a simplified schematic diagram of a radiation therapy system in accordance with various embodiments of the disclosed subject matter.

[0036] Figure 2 is a schematic diagram of a treatment volume on a planning image according to various embodiments of the disclosed subject matter.

[0037] Figure 3 is a simplified flow chart of a process plan optimization procedure in accordance with various embodiments of the disclosed subject matter.

[0038] Figure 4-Figure 5 is a schematic flow chart of a process plan optimization procedure in accordance with various embodiments of the disclosed subject matter.

[0039] Figure 6-Figure 7 is a schematic flow chart of a first step of a process plan optimization procedure in accordance with various embodiments of the disclosed subject matter.

[0040] Figure 8 is a schematic flow chart of a second step of a process plan optimization procedure according to various embodiments of the disclosed subject matter.

[0041] Figure 9A-9C is a schematic diagram of a process for optimizing beam geometry parameters in accordance with various embodiments of the disclosed subject matter.

[0042] Fig.10 is a flow chart of a two-step process plan optimization procedure in accordance with various embodiments of the disclosed subject matter.

[0043] Figure 11-Figure 12 is a schematic flow chart of an automated process for predicting beam geometry without explicitly defining the system isocenter.

[0044] Fig.13 is an illustration of the patient geometry and patient-centered quantities that are used to train a machine learning model to predict the beam incidence angle without explicitly defining the system isocenter.

[0045] Fig.14 is a display of the gantry angles predicted using an automated process that predicts beam geometry without explicitly defining the system isocenter.

[0046] Fig.15 is a flow chart of an automated process for predicting beam geometry parameters used in treatment planning without explicitly defining the system isocenter. DETAILED DESCRIPTION

[0047] See also Figure 1 , shows an exemplary radiation therapy system 100 that can be used in radiation therapy and that can deliver radiation according to a treatment plan determined using the techniques described herein. The radiation therapy system 100 can provide radiation to a patient 110 positioned on a treatment couch 112 and can allow for a variety of radiation treatment regimens to be implemented. The radiation therapy can include photon-based radiation therapy, particle therapy, electron beam therapy, or any other type of treatment therapy.

[0048] In an embodiment, the radiotherapy system 100 may include a radiation treatment device 101, such as, but not limited to, a LINAC, which is operable to generate one or more megavoltage (MV) X-ray radiation for treatment. The LINAC may also be operable to generate one or more kilovoltage (kV) X-ray radiation, for example, for patient imaging. The system 100 has a gantry 102 that supports a radiation treatment head 114 having one or more radiation sources 106 and various beam modulation elements (such as, but not limited to, a flattening filter 104 and a collimation assembly 108). For example, the collimation assembly 108 may include a multi-leaf collimator (MLC), an upper secondary collimator and a lower secondary collimator and / or other collimation elements. The collimation assembly 108 and / or the flattening filter 104 may be positioned within the radiation beam path by a corresponding actuator (not shown), which may be controlled by the controller 200.

[0049] The gantry 102 can be an annular gantry (i.e., it extends through a full 360° arc to create a complete ring or circle), but other types of mounting arrangements can also be used. For example, a static beam, C-shaped, partial annular gantry, or robotic arm can be used. Any other framework capable of placing the treatment head 114 in various rotational and / or axial positions relative to the patient 110 can also be used.

[0050] In an embodiment, the radiotherapy device is an MV energy intensity modulation radiotherapy (IMRT) device. The intensity distribution in this system is tailored to the treatment requirements of individual patients. The IMRT field is delivered with MLC 108, which can be a computer-controlled mechanical beam shaping device attached to the head 114 and includes a combination of metal fingers or leaves. For each beam direction, an optimized intensity distribution is achieved by sequentially delivering various subfields with optimized shapes and weights. From one subfield to the next, the leaves can move when the radiation beam is turned on (i.e., dynamic multi-leaf collimation (DMLC)) or when the radiation beam is turned off (i.e., segmented multi-leaf collimation (SMLC)). Therefore, MLC 108 can be used for conformal treatment of tumors from all angles and intensity modulated radiotherapy, thereby delivering different radiation doses to different parts of the treatment area. The treatment volume (i.e., the irradiation volume near the isocenter in the path of the X-ray beam) is defined by the secondary collimator, the head 114, and the MLC 108. In IMRT, the leaves of the MLC are moved so that the treatment volume includes the volume that is exposed during treatment.

[0051] Alternatively or additionally, the radiation therapy device 101 can be a tomotherapy device, a helical tomotherapy device, or a simplified intensity modulated arc therapy (SIMAT) device, a volumetric modulated arc therapy (VMAT) device, or a volumetric high definition (or hyper arc) therapy (HDRT). In fact, any type of IMRT device can be adopted as the radiation therapy device 101 of the system 100, and can also include on-board volume imaging, which can be used to generate in-process image data generated during a treatment session.

[0052] Each type of radiation therapy equipment may be accompanied by a corresponding radiation plan and radiation delivery procedure.

[0053] The controller 200 (which may be, but is not limited to, a graphics processing unit (GPU)) may include a computer having appropriate hardware (such as a processor) and an operating system for running various software programs and / or communication applications. The controller 200 may include a software program that operates to communicate with the radiation therapy device 101, which is operable to receive data from external software programs and hardware. The computer may also include any suitable input / output (I / O) device 210, which may be suitable for allowing communication between the controller 200 and a user (e.g., medical staff) of the radiation therapy system 100. For example, the controller 200 may be provided with an I / O interface, a console, a storage device, a memory, a keyboard, a mouse, a monitor, a printer, a scanner, and a department information system (DIS), such as a communication and management interface (DICOM) for storing and transmitting medical imaging information and related data and enabling integration of medical imaging devices (such as scanners, servers, workstations, printers, network hardware, etc.).

[0054] Alternatively or additionally, the I / O device 210 may provide access to a network (not shown) for transmitting data between the controller 200 and a remote system. For example, the controller 200 may be networked with other computers and radiation therapy systems via the I / O 210. The radiation therapy system 100, the radiation treatment device 101, and the controller 200 may communicate with a network as well as databases and servers, such as a dose calculation server (e.g., a distributed dose calculation framework), a treatment planning system 300, a computer processing device for training one or more machine learning models, a computer processing device for applying one or more trained machine learning models, and / or any other data processing computer processing device. The controller 200 may also be configured to transfer medical image related data between different pieces of medical equipment.

[0055] The system 100 may also include multiple modules (e.g., as part of the controller 200, or as separate modules within the system 100, or integrated into other components of the system 100) that contain programmed instructions that, when executed, cause the system 100 to perform various functions related to radiation therapy, including adaptive radiation therapy or other radiation treatments, as discussed herein. For example, the system 100 may include or communicate with the following modules: a treatment planning module 320 of the treatment planning system 300, which is operable to generate a treatment plan for the patient 110 based on multiple data input into the system by medical personnel using the computer 310; a patient positioning module, which is operable to position and align the patient 110 with respect to a desired location (such as the isocenter of the gantry) for a specific radiotherapy treatment; an image acquisition module, which is operable to instruct the radiotherapy system and / or imaging device to acquire images of the patient 110 before the radiotherapy treatment (i.e., pre-treatment / reference images for treatment planning and patient positioning) and / or acquire images of the patient 110 during the radiotherapy (i.e., images during the treatment session), and to instruct the radiotherapy system 100 and / or the imaging device 101 or other imaging devices or systems to acquire images of the patient 110.

[0056] The system 100 may also include: one or more contour generation modules, which are operable to generate contours of target volumes and other structures in pre-treatment (planning, reference) and in-treatment (treatment session) images; contour verification modules, which are operable to verify the generated contours; dose calculation modules, which are operable to calculate cumulative doses; dose assessment modules, which are operable to assess the dose distribution within the desired volume; modules for electron density map generation, isodose distribution generation, dose volume histogram (DVH) generation, image synchronization, image display, treatment plan generation, treatment plan optimization, automatic optimization parameter generation, update and selection, and adaptive instruction and treatment information transfer. The system 100 may also include a treatment delivery module, which is operable to instruct the radiotherapy device 100 to deliver the treatment plan to the patient 110.

[0057] For example, these modules may be written in C or C++ programming languages. Computer program code for implementing the operations described herein may be written in any programming language, for example, C or C++ programming languages.

[0058] The treatment planning system 300 may be used to generate an optimized treatment plan for the radiation therapy system 100. The treatment planning system 300 includes a program memory containing processor-executable instructions that, when executed by the processor 310, generate an optimized treatment plan that may be executed by a processing unit (e.g., the controller 200) of the radiation therapy system 100. The treatment planning system 300 is configured to communicate with a plan image memory containing image data and with a knowledge base, which is a database or other information retrieval system that contains, for example, machine learning models, plan templates including clinical goals (CGs) and priorities for different anatomical structures, and knowledge-based information such as patient records similar to the current patient record (i.e., previous / existing / historical treatment plans), treatment types, and knowledge-based statistical models (such as DVH models).

[0059] Treatment planning usually starts with obtaining one or more planning images / image slices of the patient's part including the tumor. For example, the image / image slice can be an image / image slice of CT, CBCT, MRI, etc. A qualified medical staff (doctor) then determines and outlines the treatment volume on the image / image slice. Outlining the treatment volume involves generating a body contour around the total volume to be irradiated, and outlining the structures contained within the body contour, such as one or more malignant tumors (i.e., target structures / volumes) that will receive a radiotherapy dose and structures that should be limited to irradiation, because exceeding a certain amount of radiation dose may have an adverse effect on them (i.e., organs at risk (OARs)). Outlining is usually done manually by qualified medical staff (doctors) on the planning image / image slice, but contour drawing can also be performed automatically or semi-automatically using any available segmentation algorithm. The planning image used for treatment planning can also be an image previously taken and stored in a planning image memory.

[0060] The typical outline of a malignant tumor includes a gross target volume (GTV), a clinical target volume (CTV), and a planned target volume (PTV). The (GTV) determines the anatomical site where the tumor cell density is the highest and the highest prescription dose is required. The (GTV) is the location and extent of the gross tumor, that is, the extent that can be seen, palpated, or imaged. The (CTV) contains the (GTV), as well as the margins where subclinical disease spreads and therefore cannot be fully imaged. The (CTV) is the volume that must be fully treated to be cured. The (PTV) allows for uncertainty in planning or treatment delivery. It is a geometric concept that is designed to ensure that the radiotherapy dose is actually delivered to the (CTV). Therefore, the (PTV) is used to compensate for the uncertainty of the treatment setup by volume expansion of the (CTV) margin. Although the treatment volume is a three-dimensional structure that includes the target volume and other structures, for ease of illustration, the treatment volume is shown as a two-dimensional cross-sectional structure. However, this is not a limiting feature.

[0061] Figure 2 An exemplary treatment volume 410 is illustrated, including a body contour 420, a target structure / volume (PTV) 430, an OAR 440, and other regions 450 of the treatment volume 410 other than the PTV 430 or the OAR 440 (ie, the remainder of the treatment volume 410, ie, normal tissue).

[0062] After outlining the treatment volume 410, the physician specifies treatment objectives (i.e., treatment prescriptions), such as a preferred / desired dose distribution within the treatment volume 410. This can be expressed as a geometry or template of clinical goals (CGs). For example, these clinical goals (CGs) can be given in the form of an average radiation dose (in Gray) to a target volume 430 and a dose that must not be exceeded to a certain volume of an organ (such as an organ at risk (OAR) 440) or other areas 450 within the treatment volume 410 exposed to radiation. However, clinical goals can also be given in other dimensions that are not in the form of radiation dose to a target volume and dose to an organ volume, such as specifying a dose value for each voxel within the treatment volume. A voxel represents a three-dimensional rectangular element into which the treatment volume is discretized. During treatment, each of these voxels absorbs a certain dose of radiation.

[0063] Each given goal can be further sorted by priority, which describes the importance of achieving one goal relative to another. This set is called a priority set of clinical goals (priority CG). Each clinical goal can be expressed as a quality metric and its associated goal value. An exemplary set of priority clinical goals is:

[0064] Objective 1: The target (PTV) must receive 90 Gy: Priority 1

[0065] Objective 2: Organs at risk (OAR) must receive a maximum dose of less than 25 Gy: Priority 2

[0066] Objective 3: The rest of the treatment volume must receive a mean dose of less than 30 Gy: Priority 3.

[0067] Another way to express the desired dose distribution is by taking a library of clinically approved and delivered plans for previously treated patients with similar medical characteristics in order to find a set of parameters that produces a clinically ideal plan for patient 110. In this approach, an algorithm (e.g., a dose volume histogram (DVH) model) trained from historical patient data (i.e., structure and dose distribution) is used as a starting point to predict achievable dose distributions for a new set of patient structures. The achievable dose distribution is represented by a pair of dose volume histograms (DVHs) that represent the lower and upper limits of the predicted 95% confidence interval. These DVH histograms can then be used as targets for each structure within the treatment volume.

[0068] The physician also prescribes a set of treatment parameters, including beam geometry (BG) parameters, to account for constraints imposed on the treatment process by the radiation therapy system 100 for delivering radiation to the treatment volume 410. For example, the physician may need to consider the number of gantry angles and collimator angles required to deliver the treatment beam in a fixed direction (i.e., radiation field), as well as any other parameters associated with each field, such as, but not limited to, MLC leaf position, beam intensity distribution, field direction, etc. For this application, beam geometry (BG) parameters include parameters related to machine instructions, including gantry angles, collimator angles, treatment couch angles, isocenter position, etc., but do not include MLC leaf and secondary collimator positions.

[0069] A treatment plan is then generated based on the treatment parameters. Generating a treatment plan typically involves using an algorithm to solve the "inverse problem" of designing and optimizing a specific plan for irradiating the treatment volume 410 from a variety of angles, or in arc therapy, delivering a desired radiation dose to the target volume while minimizing irradiation of nearby organs / tissues, etc., as the gantry moves, while taking into account the capabilities and limitations of the radiotherapy system 100. Thus, treatment planning essentially involves designing a plan that provides an optimal dose distribution to the treatment volume. This can be simplified to a cost function (objective function) that takes into account the various trade-offs inherent in such planning and has constraints that must be satisfied in order for the plan to be medically acceptable or physically possible.

[0070] Algorithms that can be used for treatment planning are dose calculation algorithms based on a given set of parameters (such as gantry angle, MLC leaf position, etc.), or search algorithms that use various techniques to adjust treatment parameters between dose calculations to achieve plan optimization. There are many algorithms that can be used for treatment planning. For example, dose calculation algorithms can include, but are not limited to, Monte Carlo techniques, pencil beam convolution, generalized Gaussian pencil beam, folded convolution, and anisotropic analysis algorithms. Search algorithms can include various random, deterministic, and other methods, including, but not limited to, simulated annealing techniques, generalized simulated annealing, dual annealing, group-based algorithms, algebraic inverse treatment plans, simultaneous iterative inverse plans, iterative least squares inverse treatment plans, and stacked convolutions, etc.

[0071] Each of these planning algorithms requires iterative dose calculations for optimization and is typically Figure 3 The process S100 is performed as shown in the process S100. In order to obtain an optimized treatment plan, the process S100 first uses the starting treatment parameters S101 to start the initial or basic dose calculation for planning. Then, the treatment plan generation and optimization algorithm S103 of the treatment plan module 320 calculates the radiation dose received by each voxel in the treatment volume in S104. If the dose distribution does not meet (S105) the desired dose distribution (S102) within a predetermined threshold (preset by the doctor), the plan is not optimal. In this case, one or more treatment parameters can be modified (S106) and the dose distribution can be recalculated in S104. Ideally, the process is iteratively performed until the desired dose distribution is obtained, at which point the plan is called optimized. However, the amount of time required to perform a large number of calculations for each iteration sets a practical limit on the number of iterations that can be performed. In this way, the optimization process is usually terminated at S107 after reaching a predetermined endpoint. The final treatment plan corresponds to the optimized plan.

[0072] Because the search space (i.e., all possible solutions) for system parameters is too large for the optimizer to handle, in current IMRT and VMAT optimization algorithms, beam geometry (BG) parameters are not optimized during the optimization process S100. Instead, the starting (initial) (BG) parameters and any other values ​​of the (BG) parameters used in the optimization process are either preset and presented using a clinical template or manually entered by the physician / planner / user.

[0073] According to the present invention, an optimization process S200 is disclosed, which will allow the optimizer to automatically optimize all optimization targets (ie, optimize all treatment parameters that need to be optimized, including beam geometry (BG) parameters) to obtain a higher quality treatment plan in a short time.

[0074] The exemplary optimization process S200 is Figure 4. The optimization process S200 includes two steps, step 1 and step 2. In the first step (step 1), one or more automated processes (S202, S202') can be performed to determine the initial (starting) beam geometry (BG) parameters (S203) and limit the search space (S204) of the beam geometry (BG) parameters (i.e., automatically find beam geometry (BG) solutions). The automated process S202 for determining the initial / starting beam geometry (BG) parameters can be independent of the automated process S202' for limiting the search space of the beam geometry (BG) parameters. Therefore, it should be understood that the machine learning model used to predict the initial selection of beam geometry (BG) parameters can be independent of / different from the machine learning model used to predict the relevant search range of the beam geometry (BG) parameters.

[0075] Alternatively or additionally, some or all of the initial (starting) beam geometry (BG) parameters may be given as a template. For example, beam geometry (BG) parameters templated based on a typical patient in a given clinic may be used as the starting / initial beam geometry (BG) parameters. Then, in order to limit the search space for the beam geometry (BG) parameters, a set of rules may be formulated to define the search space around the template for the current patient. For example, in order to cover the irradiation of the breast PTV of the current patient, rules may be established for the templated beam geometry (BG) parameters that cover the irradiation of the PTV of the current patient while limiting irradiation of the contralateral breast and / or heart. These rules may include, but are not limited to, rules that limit the gantry and / or collimator angles to angles that avoid irradiating a breast that does not contain a tumor and / or avoid irradiating the heart. These templated (BG) parameters and the associated rules for establishing search boundaries are merely exemplary, and it should be understood that any other appropriate templates and rules may be used to obtain the initial / starting beam geometry (BG) parameters and establish corresponding search boundaries.

[0076] Additionally or alternatively, the medical staff (planner) may manually enter one or more initial beam geometry (BG) parameters (such as starting gantry angle, starting collimator angle, etc.) and / or search boundaries of beam geometry (BG) parameters via the graphical user interface.

[0077] Additionally or alternatively, initial / starting beam geometry (BG) parameters may be templated or manually entered by the planner, and an automated process may be applied to establish search boundaries for the beam geometry (BG) parameters.

[0078] In the second step (step 2), an automatic optimization process is performed to optimize all optimization targets (i.e., based on the information obtained in the first step (step 1), all variable treatment parameters including beam geometry (BG) parameters (S206) are optimized) to generate an optimal treatment plan (S208). Since the search space of beam geometry (BG) parameters is automatically determined in the first step (step 1), and since the optimization is performed with all degrees of freedom in the second step (step 2), the optimal treatment plan (S208) is generated faster and with higher quality than a treatment plan in which the doctor manually sets the (BG) parameters.

[0079] To determine the initial / starting beam geometry (BG) parameters (S203), the automated process of the first step (step 1) uses a trained machine learning model (S202). The machine learning model (S202) is a trained machine learning model that is trained to predict beam geometry (BG) parameters as output based on patient geometry S201 (i.e., input), which includes patient-related information such as planning images / image slices (CT, CBCT, MRI, etc.), body contours, delineated targets (i.e., target structures), and delineated OARs.

[0080] An exemplary training machine learning model (S202) that can be used in step 1 is a knowledge-based supervised machine learning model ( Figure 6 ). Figure 6 The knowledge-based supervised machine learning model S202 shown in can be a machine learning model trained on historical treatment plans, that is, the machine learning model S202 is trained to predict the beam geometry (BG) parameters for a new patient geometry based on the patient geometry and the beam geometry (BG) parameters used in previous treatment plans to achieve the same / similar treatment goals.

[0081] like Figure 4 and Figure 5 As shown in , the patient geometry S201 for which the optimal treatment plan is to be formulated is used as input data in the trained machine learning model S202. By executing the trained machine learning model S202, the initial / starting beam geometry (BG) parameters S203 are obtained as machine learning model predictions S203A. The output of the trained machine learning model S202 can be used as the input of the trained machine learning model S202'. The trained machine learning model S202' is a machine learning model trained to predict the relevant maximum search range of the initial / starting beam geometry (BG) parameters S203. By executing the trained machine learning model S202', the confidence interval (S204A') and the standard deviation of the average prediction (S204B') in the machine learning prediction (S203A) can be determined as the search boundary S204.

[0082] By determining the confidence interval (S204A') and the standard deviation of the average prediction (S204B') in the machine learning prediction (S203A), the feasible parameter range (S204A) of the prediction and the maximum acceptable deviation from the initial prediction (S204B) can be determined respectively. The feasible parameter range (S204A) and the maximum acceptable deviation from the initial prediction (S204B) are expressions of the search boundaries S204 of the predicted beam geometry (BG) parameters S203.

[0083] Alternatively or additionally, the search bounds (ie, search ranges) for some or all beam geometry (BG) parameters may be determined using established rules used in clinical practice and / or by a physician.

[0084] like Figure 6 As shown in , when the machine learning model S202 is trained to predict beam geometry (BG) parameters (such as gantry angle, collimator angle, medial angle, etc.), the machine learning model outputs the initial / starting gantry angle, collimator angle, medial angle, etc. (S203) as a prediction S203A, whereby the corresponding feasible angle range S204A and the maximum acceptable deviation from the corresponding initial / starting angle S204B can be calculated and / or determined using an independent machine learning model S202', or using established rules used in clinical practice, and / or determined by a physician.

[0085] The initial / starting beam geometry (BG) parameters (S203) and associated search boundaries (S204) obtained in the first step (step 1), whether expressed as a feasible parameter range (S204A) or as a maximum acceptable deviation (S204B), are input into the treatment plan optimizer S205, and the second step (step 2) begins to search for the optimal treatment plan in the treatment plan optimizer S205.

[0086] Typically, in order to obtain an optimized treatment plan S208, the optimization process of step 2 begins with an initial or basic dose calculation using the starting treatment parameters S206 of the treatment plan. The initial / starting beam geometry (BG) parameters S203 obtained in the first step (step 1) are used as the starting beam geometry (BG) parameters. After the initial calculation, the treatment plan generation and optimization module S205 determines whether the treatment goal S207 is achieved. If the treatment goal is not achieved, one or more treatment parameters including the beam geometry (BG) parameters are iteratively modified until the goal S207 is achieved or the endpoint is reached. The final plan is an optimized treatment plan S208. The search for the beam geometry (BG) parameters is limited to the corresponding search boundary S204 determined in the first step (step 1).

[0087] exist Figure 8An exemplary optimization step 2 is illustrated in FIG. 1 , where the treatment target is a desired dose distribution within the treatment volume. Optimization step 2 begins with an initial or basic dose calculation using the starting treatment parameters of the treatment plan S206 by a dose distribution calculation algorithm S208. The initial / starting beam geometry (BG) parameters S203 obtained in the first step (step 1) are used as the starting beam geometry (BG) parameters.

[0088] If the dose distribution does not meet the desired dose distribution S207 (i.e., target) within a predetermined threshold (preset by the physician), the treatment plan is not optimal. In this case, one or more treatment parameters including beam geometry (BG) parameters S206 may be modified, and the dose distribution is recalculated at S208. Ideally, this process is iteratively performed until the desired dose distribution S207 or a preset endpoint is reached at S209, at which point the optimization process of step 2 ends at S213, and the resulting treatment plan S214 is called optimized.

[0089] In each iteration of the optimization process, the beam geometry (BG) parameters are automatically modified at S210 using a beam geometry (BG) parameter optimization algorithm S211 to obtain updated / revised beam geometry (BG) parameters ( S212 ) for recalculating the dose distribution at S208 .

[0090] Additionally or alternatively, beam geometry (BG) parameters and optimization objectives may be iteratively modified ( S210 ′) to generate an optimized treatment plan.

[0091] Figure 9A-9C Exemplary BG parameter optimization algorithms are illustrated, such as simulated annealing method S211A, simplex method S211B, and training reinforcement learning method S211C, which can be used to update beam geometry (BG) parameters in each iteration of the optimization process. Each of these methods S211A-S211C uses the beam geometry (BG) parameter values ​​from the previous iteration of the optimization process, loops through a series of iterations, and derives an optimal solution for the beam geometry (BG) parameters within the range / deviation obtained in step 1. For the first optimization iteration, the initial / starting beam geometry (BG) parameter S203 values ​​are used for the beam parameter (BG) optimization algorithms S211A-S211C. These are just exemplary beam parameter (BG) optimization algorithms, and any other optimization algorithm / method can be used to update / modify the beam geometry parameters, such as but not limited to any deterministic, stochastic, or heuristic optimization method.

[0092] Figure 7An exemplary automated process for developing an optimal treatment plan that irradiates a non-extended target of the left breast of a patient with a partial arc VMAT is illustrated. In this example, the body surface is substantially parallel to the chest or sternum. The relevant beam geometry (BG) parameter that one would like to optimize is the medial angle at which the arc begins. The medial angle is important for adjustment to allow maximum coverage of the target (i.e., target structures) while avoiding excessive exposure of the contralateral breast and lung (i.e., OARs). Other beam geometry (BG) parameters such as the posterior (end angle) and collimator angle can be templated and / or manually fixed by the physician (planner).

[0093] To determine the initial / starting medial angle S303 and search boundaries (S304A, S304B) for the second optimization step, the trained machine learning model S302 is first executed to predict the initial choice of medial angle S303 from the patient geometry data S301 (i.e., step 1). The trained machine learning model S302 is trained on previous treatment plans. The trained machine learning model S302 may also indicate that for similar situations, the maximum search range (S304A) does not exceed ±10 degrees of the initial selection of the inner angle, for example, and / or the maximum acceptable deviation (S304B) does not exceed the initial selection of the inner angle S303. Therefore, for this example case, the predicted medial angle is are used as the initial / starting beam geometry (BG) parameters in the second optimization step (step 2), and and / or is used as the search range for the medial angle within which the optimizer finds the dose-optimal treatment plan.

[0094] Alternatively, a second different machine learning model may be applied that is trained to predict the maximum search range associated with the initial / starting beam geometry (BG) parameters S303. By executing this trained machine learning model, a feasible angle range S304A and a maximum acceptable deviation S304B may be determined. The predicted medial angle is then is used as the initial / starting beam geometry (BG) parameters in the second optimization step (step 2), and and / or Serves as the search range for the medial angle within which the optimizer will find the dose-optimal treatment plan.

[0095] Fig.10 is a flow chart illustrating steps of an embodiment of a treatment plan optimization process S400 of the present invention, wherein the optimization is performed with all degrees of freedom including degrees of freedom associated with (BG) parameters.

[0096] In step S401 , an automated process is performed to determine search boundaries for beam geometry (BG) parameters based on patient geometry data and an initial selection of beam geometry (BG) parameters.

[0097] In one or more embodiments, the automated process includes executing one or more trained machine learning models to determine search boundaries for beam geometry (BG) parameters.

[0098] In an exemplary embodiment, an initial selection of beam geometry (BG) parameters is automatically determined using a trained supervised machine learning model that receives patient geometry data as input and outputs an initial selection of beam geometry (BG) parameters.

[0099] In an alternative embodiment, the initial selection of beam geometry (BG) parameters is from a template or is manually entered by a planner. In such an embodiment, the automated process includes executing one or more trained machine learning models that receive as input the patient geometry data and the templated or manually entered initial selection of beam geometry (BG) parameters and output a search boundary for the beam geometry (BG) parameters. Additionally or alternatively, the automated process includes executing one or more rules to determine the search boundary for the beam geometry (BG) parameters. Additionally or alternatively, the search boundary for the beam geometry (BG) parameters may also be manually entered by a user.

[0100] In step S402, at a treatment plan optimization module, treatment parameters, treatment plan objectives, and initial beam geometry (BG) parameters and associated search boundaries are received.

[0101] In step S403, the treatment plan optimization algorithm of the treatment plan optimization module is iteratively executed to optimize the treatment parameters by automatically modifying / revising / changing the treatment parameters including the beam geometry (BG) parameters, wherein the search for the beam geometry (BG) parameters is limited to the determined corresponding search boundaries. In each iteration of the optimization process, the treatment parameters including the beam geometry (BG) parameters are automatically modified / revised / changed until it is determined in S406 that the treatment plan objectives evaluated in step S404 are met. The resulting treatment plan is the optimal treatment plan.

[0102] The beam geometry (BG) parameters are modified / revised / changed in each iteration of the optimization process by executing an optimization algorithm in S405, which receives the beam geometry (BG) parameters and corresponding search boundaries as input data and outputs updated beam geometry (BG) parameters within the determined corresponding search boundaries.

[0103] Therefore, by applying a two-step optimization process (step 1 and step 2), beam geometry (BG) parameter predictions and acceptable variations (step 1) can be automatically determined and the optimizer can be allowed to focus the search on acceptable initial conditions, which improves optimization algorithm performance.

[0104] Fig.11 and Fig.12 An example of an automated process S500 to obtain beam geometry (BG) parameters that do not require explicit determination of system / machine isocenter locations to be used in a treatment plan optimization process is illustrated. The process S500 may include a training phase S600, an application phase S700, and a transformation phase S800.

[0105] In the training phase S600, a machine learning model S620 is trained using a training dataset S610 to predict a specific output S622. The training dataset S610 includes a plurality of images S611 containing patient geometry (target structures (PTV) and OARs) and patient-centric quantities (PCQs) as ground truth values.

[0106] exist Fig.13 An exemplary patient image S611 is illustrated in FIG. In this example, assuming that the patient is in a supine position, the image can be taken from the axial direction of the patient, but other orientations and views are also possible. Image S611 includes a target structure (PTV), an organ at risk (OAR), and patient-centered quantities (PCQs), such as an auxiliary reference point (ARP), a body surface angle (BSA), a beam incidence angle (BEA), and a relative beam incidence angle (RBEA). The auxiliary reference point (ARP) is a reference point in the target structure and can be defined as the center of mass of the target. In general, the "body surface" can be defined using a portion of the body contour close to the clinical site in question. For example, if the patient is in a supine position and the clinical site is located in the upper body (excluding the head and neck), the orientation of the patient's chest can be used to define the direction of the body surface. The body surface angle (BSA) can be used to track the slight axial tilt that may occur in the patient's body (i.e., a body part close to the clinical site) relative to its ideal position on the treatment bed (e.g., horizontal, depending on the clinical site). The beam incidence angle (BEA) may be defined as the angle at which the radiation beam enters the patient relative to the patient body and target structures, and the relative beam incidence angle (RBEA) is the difference between the beam incidence angle (BEA) and the body surface angle (BSA).

[0107] A training data set S610 may be generated by defining a patient-centered quantity (PCQ) S602 on a plurality of medical images / image slices (e.g., CT images) containing target structures and OARs (S601). The medical images / image slices may be images / image slices (e.g., CT images) previously acquired for the same patient or a previous patient, and the patient-centered quantity (PCQ) may be generated by a user on the images / image slices through observation and experience to facilitate supervised learning. The user may be a planner / doctor and / or qualified medical personnel. Alternatively or additionally, both the patient geometry and the patient-centered quantity (PCQ) may be generated by a user on the medical images / image slices. Alternatively or additionally, both the patient geometry and the patient-centered quantity (PCQ) may be automatically generated on the medical images / image slices using a contour / segmentation algorithm suitable for the patient geometry and a graphical algorithm suitable for the patient-centered quantity. For example, the training data set S610 may be stored in a planning image memory.

[0108] The machine learning model S621 can be a neural network model running on a neural network engine, which refers to any suitable hardware and / or software component of a computer system that can execute an algorithm according to any suitable deep learning model. The deep learning model can be based on any existing or later developed neural network or a combination thereof, such as but not limited to convolutional neural network (ConvNet or CNN), residual neural network (ResNet), recurrent neural network (RNN), long short-term memory (LSTM) neural network, recursive neural network, generative adversarial network (GAN), and deep belief network (DBN). A deep neural network refers to a class of computer-based machine learning algorithms that utilize many layers or stages (i.e., hidden layers) of data processing for feature learning, pattern analysis, and / or classification. Typically, a deep neural network is formed by a hierarchical network of processing elements interconnected by connections. The node layer is trained from end to end (i.e., from the input layer to the output layer) to extract features from the input and classify (multiple) features to produce outputs. For example, a node in a hidden layer may include a filter or kernel, whose parameters (e.g., kernel weights, size, shape, or structure) may be adjusted during the training process.

[0109] Training of the machine learning model S620 refers to determining one or more parameters of nodes in the hidden layer of the deep neural network through an iterative process S623, which changes the parameters S624 so that the neural network model output S622 more closely matches the corresponding true value. In this example, the machine learning model S621 is trained to output a beam incidence angle (BEA) S622 that closely matches the true beam incidence angle (BEA) of the training data set S610.

[0110] In operation, a training data set S610 is input to a machine learning model S621, for example, to its input layer. The machine learning model S621 processes the training data by propagating through the nodes of its hidden layer to produce a result (i.e., beam incidence angle (BEA) S622) at its corresponding output layer. The output beam incidence angle BEA (S622) is compared with the true beam angle (BEA) via a loss function. The loss function can be a mean square error, a dice loss, a cross-entropy-based loss, or any other loss function known in the art. During training S600, the loss function provides feedback to the machine learning model S621, indicating the degree of match between its output S622 and the correct output (true BEA). Once the iteration criteria S623 are met (i.e., the loss function meets a predetermined threshold, a threshold number of iterations has been reached, or no further improvement is seen between iterations), the machine learning model is fixed as a trained machine learning model S710. Otherwise, training S600 continues by adjusting S624 the parameters of the hidden layer nodes to improve the match between the output (S622) and the desired output (ie, the true value BEA). The training process S600 may be iterated repeatedly until the desired iteration criterion is met at S623.

[0111] Once the training phase S600 is completed, the process S500 can proceed to the application phase S700, which uses the trained machine learning model S710 to process the patient's images to automatically predict the beam incidence angle (BEA) relative to the patient. The application phase S700 can begin, where a patient dataset S720 is provided to the trained machine learning model S710. The patient dataset S720 includes one or more images / image slices (e.g., CT) obtained for the patient, each image / image slice containing a target structure (PTV) and OARs (i.e., patient geometry) and auxiliary reference points (ARPs) defined for the target structure (PTV). The auxiliary reference point (ARP) can be the center of mass of the target and can be determined by the user or automatically calculated using a center of mass calculation algorithm. Any suitable center of mass calculation algorithm can be used to find the center of mass of the target.

[0112] The trained machine learning model S710 takes as input a patient dataset S720 including patient geometry and associated auxiliary reference points (ARPs), processes the dataset S720, and outputs a beam incidence angle (BEA) S730 based on its training.

[0113] In the transformation phase S800, the beam incidence angle (BEA) S730 obtained in the application phase S700 can be superimposed S820 on the anatomical structure contained in the CT image used in the application phase S700 using a computer processing device S810 to allow a user to visualize the predicted beam incidence angle (BEA) S730 relative to the patient geometry. The predicted beam incidence angle (BEA) S730 superimposed on the patient geometry including the body contour, target structure, and OAR can be displayed (S820) on the display of the computer processing device S810. The computer processing device S810 can be the same as or different from the computer processing device 310 used in the treatment planning, machine learning model training, or computer processing device 200 of the processing system 100. When different computer processing systems are used, the computer processing systems are networked with each other via I / O 210 and with the controller 200.

[0114] Next, the beam incidence angle (BEA) superimposed on the patient geometry is evaluated (S821) by the user. The user may interactively manually adjust / tune (S823) the beam incidence angle (BEA) S730 until the user determines in S821 that the beam incidence angle (BEA) is acceptable (S822). The determination may be based on visual observation of predicted beam incidence angles (BEA) relative to target structures, OARs, and auxiliary reference points (ARPs) in the target structure. The evaluation may be guided by the user's clinical experience with acceptable beam incidence angles relative to the patient geometry. The computer processing device S810 may include a graphical interface to allow interactive beam incidence angle (BEA) adjustment / tune.

[0115] Once the beam incidence angle (BEA) is accepted, the user can continue to define the isocenter in S830 by accepting the auxiliary reference point (ARP) of S820 as the isocenter (ISO-ARP) in S831, or manually entering the isocenter location (ISO-MAN) based on the observed ARP location (S832). Instead of manually selecting the isocenter, the isocenter can be picked using a computer processing algorithm that is programmed to define the best possible location of the isocenter based on the patient geometry and the auxiliary reference point (ARP).

[0116] If the user selects an auxiliary reference point (ARP) (S831) as the isocenter, the beam incidence angle (BEA) accepted in S822 is determined to be the beam geometry (BG) parameter 900. In a specific embodiment, the beam geometry (BG) parameter 900 is the gantry angle, and thus, the beam incidence angle (BEA) accepted in S822 is the same as the gantry angle.

[0117] If the user manually (or using an algorithm) selects the isocenter (ISO-MAN), the isocenter is further evaluated in S833 to determine whether the selected (or calculated) isocenter location is close to the auxiliary reference point (ARP). This evaluation can be done by displaying the isocenter location on the patient geometry and visually determining whether the isocenter location is close to the auxiliary reference point (ARP), or by using the computer processing system S810 to calculate the difference between the isocenter location and the auxiliary reference point (ARP) location.

[0118] If it is determined in S833 that the ISO-MAN (S832) is close to the auxiliary reference point (ARP) (i.e., within a predetermined threshold), the beam geometry (BG) parameters 900' are determined from the beam incidence angle (BEA). For example, the beam incidence angle (BEA) accepted in S822 may be further modified (S823) to account for the difference between the location of the auxiliary reference point (ARP) and the location of the ISO-MAN (S832). The modified beam incidence angle (BEA) is then accepted as the beam geometry (BG) parameters 900'. Therefore, when the ISO-MAN (S832) is close to the auxiliary reference point (ARP) (i.e., within a predetermined threshold), the gantry angle 900' may also be determined from the beam incidence angle (BEA).

[0119] On the other hand, if it is determined in S833 that the ISO-MAN (S832) is far from the auxiliary reference point (ARP) (i.e., beyond a predetermined threshold), and is therefore significantly beyond the target, the user may need to re-evaluate whether the (BEA) predicted in S730 is actually an acceptable approximation of the allowed beam geometry (BG) parameters (i.e., gantry angle) given the specific isocenter selection (ISO-MAN). This can be done by manually re-evaluating the beam geometry in S821 and repeating the process until an acceptable (BEA) and subsequently acceptable beam geometry (BG) parameters are obtained (900').

[0120] Optionally, once the isocenter (ISO-ARP S831 or ISO-MAN S832) and gantry angle (900, 900') are defined, one can Fig.13 The patient and machine settings can be re-visualized and displayed to the user. Fig.14 As shown in , the body contours and target structures can be appropriately shifted relative to the determined isocenter and the gantry orientation drawn on the display. The user can accept, reject, or manually adjust the gantry angle, just as done for the beam incidence angle (BEA). After the final inspection, the machine-specific beam geometry (BG) parameters (i.e., gantry angle) can be accepted as the beam geometry (BG) parameters of the treatment plan.

[0121] The beam geometry (BG) parameters 900, 900' are machine specific parameters that may be used in treatment planning. In an exemplary embodiment, the beam geometry (BG) parameters 900, 900' obtained in process S500 may be used as initial / starting beam geometry (BG) parameters in the two-step optimization process S200.

[0122] Fig.15 An exemplary flow chart is illustrated showing steps of a process S1000 that allows for automatic prediction of beam geometry (BG) parameters that can be evaluated and selectively adjusted by a user without the need to define an explicit isocenter at the beginning of the process.

[0123] In step S1001, a computer processing system receives training patient geometry data (a CT image with delineated target structures and OARs). In step S1002, patient-centric quantities are defined for the training patient data, including auxiliary reference points inside the target structure, body surface angles, and beam incidence angles. In step S1003, a machine learning model is trained using the training patient geometry data and the patient-centric quantities to predict a beam incidence angle (BEA). In step S1004, a computer processing system receives patient geometry data including a medical image (e.g., CT) including body contours, target structures, and OARs, and determines auxiliary reference points (ARPs) within the target structure. In step S1005, the patient geometry and associated ARPs are used as input to a trained machine learning model, which predicts a beam incidence angle (BEA) as output. In step S1006, the predicted beam incidence angle (BEA) is superimposed on the patient geometry and displayed to a user. In step S1007, the user accepts, rejects, or adjusts the beam incidence angle (BEA). In step S1008, the user determines the isocenter by accepting the ARP as the isocenter or setting the isocenter manually (or using an algorithm). In step S1009, machine-specific beam geometry (BG) parameters (gantry angle) are determined based on the selected isocenter. Optionally, in step S1010, the body contour and target structure are shifted relative to the selected isocenter, and the gantry angle is drawn on the display screen, followed by S1011, where the user can again accept, reject, or adjust the gantry angle.

[0124] The accepted gantry angle may optionally be used in step S1012 as initial beam geometry (BG) parameters for treatment planning and optimization in step S1013.

[0125] The software stored in the planning image memory and / or computer 310 of the treatment planning system 300 and / or the computer used in S810 is configured to be loaded and processed in any conventional manner and is configured to be executed to optimize the treatment plan, train, and apply the trained machine learning model, determine patient-centric quantities, and predict, evaluate, and adjust beam incidence angles without specifically defining the system / machine isocenter, as described herein. The software is also configured to optimize the treatment plan to irradiate a target volume using a radiotherapy system having a multi-leaf collimator and capable of irradiating the treatment volume from one or more angles.

[0126] The treatment planning software also includes software that converts the results of the optimized treatment plan into instructions for operating the radiotherapy system 100 for controlling elements of the radiotherapy system 100 that define beam geometry and irradiation angles (treatment bed, gantry, MLC, etc.) in order to provide an optimized radiation treatment plan to the patient 110.

[0127] It is therefore apparent that methods and systems are disclosed herein for an automated treatment planning optimization process that can be performed in all degrees of freedom including those related to beam geometry (BG) parameters.

[0128] It is also apparent that methods and systems are disclosed herein for determining automated beam geometry (BG) parameters (ie, finding a BG solution) that do not require explicit determination of the system isocenter position.

[0129] It is also apparent that a system comprising a computer processing device configured to execute a sequence of programming instructions embodied on a computer-readable storage medium is also disclosed herein, the execution of which causes the system to perform any method steps disclosed herein or alternatively any combination of method steps.

[0130] Also disclosed is a non-transitory computer-readable storage medium on which a sequence of programming instructions for process plan optimization is embodied, and a computer processing system that executes the sequence of programming instructions embodied on the computer-readable storage medium is disclosed. Execution of the sequence of programming instructions may cause the computer processing system to perform the process of process plan optimization described herein.

[0131] Also disclosed is a non-transitory computer-readable storage medium having embodied thereon a sequence of programmed instructions for determining automated beam geometry (BG) parameters (i.e., finding a BG solution) without explicitly determining a system isocenter location, and a computer processing system for executing the sequence of programmed instructions embodied on the computer-readable storage medium.

[0132] It should be appreciated that aspects of the disclosed subject matter may be implemented in whole or in part in hardware, hardware programmed by software, software instructions stored on a computer-readable medium (eg, a non-transitory computer-readable medium), or any combination thereof.

[0133] For example, components of the disclosed subject matter (including components such as controllers, processes, or any other features) may include, but are not limited to, a personal computer or workstation or other such computing system including a processor, microprocessor, microcontroller device, or comprised of control logic including integrated circuits, such as, for example, an application specific integrated circuit (ASIC).

[0134] Features discussed herein may be performed by components distributed across multiple computers or systems, or by components co-located in a single processor or system on a single or distributed processor (single core and / or multi-core). For example, aspects of the disclosed subject matter may be implemented via a programmed general purpose computer, an integrated circuit device (e.g., ASIC), a digital signal processor (DSP), an electronic device programmed with microcode (e.g., a microprocessor or microcontroller), a hardwired electronic or logic circuit, a programmable logic circuit (e.g., a programmable logic device (PLD), a programmable logic array (PLA), a field programmable gate array (FPGA), a programmable array logic (PAL)), software stored on a computer readable medium or signal, an optical computing device, a networked system of electronic and / or optical devices, a special purpose computing device, a semiconductor chip, a software module or object stored on a computer readable medium or signal.

[0135] When implemented in software, functions can be stored on or transmitted by computer-readable media as one or more instructions or codes. The steps of the method or algorithm disclosed herein can be embodied in a software module executable by a processor, which module can reside on a computer-readable medium. Instructions can be compiled from source code instructions provided according to a programming language. Programming instruction sequences and data associated therewith can be stored in a computer-readable medium (e.g., a non-transient computer-readable medium), such as a computer memory or a storage device, which can be any suitable memory device, such as, but not limited to, a read-only memory (ROM), a programmable read-only memory (PROM), an electrically erasable programmable read-only memory (EEPROM), a random access memory (RAM), a flash memory, a disk drive, etc.

[0136] As used herein, computer-readable media include computer storage media and communication media, including any medium that facilitates the transfer of computer programs from one place to another. Thus, storage media can be any available medium that is accessible to a computer. By way of example, but not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.

[0137] Furthermore, any connection may be properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a transmission medium (e.g., coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave), the transmission medium is included in the definition of computer-readable medium. Furthermore, the operations of a method or algorithm may reside as one of a set of codes and / or instructions (or any combination thereof) on a machine-readable medium and / or computer-readable medium, which may be incorporated into a computer program product.

[0138] Those skilled in the art will readily appreciate that the above description is not exhaustive, and that various aspects of the disclosed subject matter may be implemented in ways other than those specifically disclosed above. In fact, from the functional descriptions provided herein, embodiments of the disclosed subject matter may be implemented in hardware and / or software by those skilled in the art using any known or later developed system, structure, device, and / or software.

[0139] In this application, unless expressly stated otherwise, the use of the singular includes the plural, and the individual use of "or" and "and" includes the other thereof, i.e., "and / or". In addition, the use of the terms "including" or "having", and other forms such as "including", "comprising", "including", "having" or "having" are intended to have the same effect as "comprising" and therefore should not be construed as limiting.

[0140] Any range described herein will be understood to include the endpoints and all values ​​between the endpoints. Whenever "substantially", "approximately", "essentially", "closely" or similar language is used in conjunction with a specific value, unless otherwise expressly stated, it means that the value does not vary by more than 10%.

[0141] The terms "system," "device," and "module" are used interchangeably herein, and the use of one term in the description of an embodiment does not prevent the application of the other term to that or any other embodiment.

[0142] The present disclosure enables many alternatives, modifications, and variations. Although specific examples have been shown and described in detail to illustrate the application of the principles of the present invention, it should be understood that the present invention can be embodied in other ways without departing from these principles. For example, the disclosed features can be combined, rearranged, omitted, etc. to produce additional embodiments, and some disclosed features can sometimes be utilized without corresponding use of other features. Therefore, the applicant intends to cover all such alternatives, modifications, equivalents, and variations, as long as they are within the spirit and scope of the present invention.

Claims

1. A method for optimizing a treatment plan for a radiation beam delivery system, the method comprising: A first step, the first step is configured to restrict the search space of an optimization target of an optimization algorithm, the optimization target being a beam geometry (BG) parameter of the radiation beam delivery system; as well as The second step is configured to optimize the beam geometry (BG) parameters based on the restricted search space obtained in the first step.

2. The method of claim 1 , wherein the first step comprises one of: predicting an initial selection of the beam geometry (BG) parameters; inputting the initial selection of the beam geometry (BG) parameters by a user; or entering the initial selection of the beam geometry (BG) parameters using a template, wherein the prediction of the initial selection of the beam geometry (BG) parameters comprises executing a machine learning model trained to predict the initial selection of the beam geometry (BG) parameters based on patient geometry.

3. The method according to claim 2, wherein the first step further includes one of the following: predicting a feasible range of the beam geometry (BG) parameters and / or a maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters; inputting the feasible range of the beam geometry (BG) parameters and / or the maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters by the user; or templating the feasible range of the beam geometry (BG) parameters and / or the maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters. 4 . The method of claim 3 , wherein the beam geometry (BG) parameter comprises one of: a gantry angle, a collimator angle, and a radiation field direction of the radiation beam delivery system.

5. The method according to claim 3, wherein the second step comprises: receiving information regarding a desired dose distribution and treatment parameters within a treatment volume of a patient as input to the treatment plan optimization algorithm; calculating a dose distribution within the treatment volume by executing the treatment plan optimization algorithm; determining whether the calculated dose distribution is within an acceptable threshold of the desired dose distribution; as well as iteratively modifying one or more of the treatment parameters including the beam geometry (BG) parameters until the calculated dose distribution is within the acceptance threshold or an endpoint has been reached, Wherein the beam geometry (BG) parameters are modified based on the search space obtained in the first step.

6. The method of claim 5, wherein in each iteration of the optimization process, the beam geometry (BG) parameters are modified using one of the following: a trained reinforcement learning agent, a simplex method, simulated annealing, or any other deterministic, stochastic, or heuristic optimization method.

7. The method of claim 2, wherein the machine learning model is a knowledge-based supervised machine learning model trained using historical processing plans.

8. The method of claim 1, wherein the optimizing step comprises executing an automated process to predict an initial selection of the beam geometry (BG) parameters without explicitly determining the radiation beam delivery system isocenter.

9. The method of claim 8, wherein the automated process comprises: executing a trained machine learning model to predict a beam incidence angle relative to a reference point within the patient based on patient geometry as an input; as well as The initial beam geometry (BG) parameters are determined based on the predicted beam incidence angle.

10. The method of claim 9, wherein the determining of the initial beam geometry (BG) parameters comprises: displaying the predicted beam incidence angle superimposed on the patient geometry on a display for user evaluation; allowing the user to accept the predicted beam incidence angle or to interactively modify the predicted beam incidence angle on the display until an acceptable beam incidence angle is obtained; and allowing the user to accept the reference point as the isocenter, or to manually enter the isocenter, Wherein when the reference point is accepted as the isocenter, the accepted beam incident angle is selected as the initial beam geometry (BG) parameters, and when the manually entered isocenter is closely adjacent to the reference point, the initial beam geometry (BG) parameters are calculated based on the accepted beam incident angle.

11. The method of claim 10, wherein the patient geometry includes a body contour and a target structure, the reference point is a center of mass of the target structure, and the beam geometry (BG) parameter is a gantry angle.

12. The method according to claim 11, further comprising: shifting the body contour and the target structure relative to the determined isocenter; and The gantry angle is further adjusted based on the displacement.

13. An automated method for obtaining beam geometry (BG) parameters for a treatment plan to a patient without explicitly determining the isocenter of a radiation beam delivery system for delivering the treatment plan to the patient, the method comprising: determining a reference point within the target; obtaining a predicted beam incident angle relative to the reference point by executing a machine learning model trained to predict the beam incident angle relative to the reference point based on patient geometry; displaying the predicted beam incidence angle superimposed on the patient geometry for user evaluation; allowing the user to accept the predicted beam incidence angle or to interactively modify the predicted beam incidence angle until an acceptable beam incidence angle is obtained; as well as allowing the user to accept the reference point as the isocenter or to manually enter an isocenter, Wherein when the reference point is accepted as the isocenter, the accepted beam incidence angle is selected as the beam geometry (BG) parameter, and when the manually entered isocenter is in close proximity to the reference point, the beam geometry (BG) parameter is calculated based on the accepted beam incidence angle.

14. The method of claim 13, wherein the patient geometry includes a body contour and a target structure, the reference point is a center of mass of the target structure, and the beam geometry (BG) parameter is a gantry angle.

15. The method according to claim 14, further comprising: shifting the body contour and the target structure relative to the determined isocenter; and The gantry angle is further adjusted based on the displacement.

16. A system for developing a treatment plan for delivering a prescribed radiation dose to a treatment volume within a patient, the system comprising: processor; as well as a memory coupled to the processor, the memory comprising instructions that, when executed by the processor, cause the processor to: receiving information related to the prescribed radiation dose, the treatment volume, and a plurality of parameters associated with the radiation beam delivery system, the plurality of parameters including beam geometry (BG) parameters; formulating a treatment plan optimization model based on the received information, the treatment plan optimization model being configured to find an optimal dose distribution within the treatment volume; as well as generating an optimal treatment plan based on the treatment plan optimization model, Wherein the generating of the optimal treatment plan comprises: A first step, wherein the first step is configured to limit a search range of the beam geometry (BG) parameter; as well as A second step is configured to optimize the processing parameters, wherein the beam geometry (BG) parameters are optimized based on the restricted search field obtained in the first step.

17. The system of claim 16, wherein the first step comprises one of: predicting an initial selection of the beam geometry (BG) parameters, and a feasible range of the beam geometry (BG) parameters, and / or a maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters; inputting the initial selection of the beam geometry (BG) parameters, and a feasible range of the beam geometry (BG) parameters, and / or a maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters by a user; or using a template to determine the initial selection of the beam geometry (BG) parameters, and a feasible range of the beam geometry (BG) parameters, and / or a maximum acceptable deviation from the initial selection of the beam geometry (BG) parameters, Wherein the predicting comprises executing a trained machine learning model to predict the initial selection of the beam geometry (BG) parameters based on patient geometry as input.

18. The system according to claim 17, wherein the second step comprises: calculating a dose distribution within the treatment volume; determining whether the calculated dose distribution is within an acceptable threshold of a desired dose distribution; as well as iteratively modifying one or more of the treatment parameters including the beam geometry (BG) parameters until the calculated dose distribution is within the acceptance threshold or an endpoint has been reached, Wherein the beam geometry (BG) parameters are modified based on the search range obtained in the first step.

19. The system of claim 18, wherein the beam geometry (BG) parameters are modified using one of: a trained reinforcement learning agent, a simplex method, simulated annealing, or any other deterministic, stochastic, or heuristic optimization method.

20. A system for obtaining beam geometry (BG) parameters for a treatment plan to a patient without explicitly determining the isocenter of a radiation beam delivery system for delivering the treatment plan to the patient, the system comprising: processor; as well as a memory coupled to the processor, the memory comprising instructions that, when executed by the processor, cause the processor to: determining a reference point within the target; obtaining a predicted beam incident angle relative to the reference point by executing a machine learning model trained to predict the beam incident angle relative to the reference point based on patient geometry; displaying the predicted beam incidence angle superimposed on the patient geometry for user evaluation; allowing the user to accept the predicted beam incidence angle or to interactively modify the predicted beam incidence angle until an acceptable beam incidence angle is obtained; as well as allowing the user to accept the reference point as the isocenter or to manually enter an isocenter, wherein when the reference point is accepted as the isocenter, the accepted beam incidence angle is selected as the beam geometry (BG) parameter, and when the manually entered isocenter is in close proximity to the reference point, the beam geometry (BG) parameter is calculated based on the accepted beam incidence angle, The processor is further configured to: shifting the patient geometry relative to the determined isocenter; and The gantry angle is further adjusted based on the displacement.

Images