Method and system for robust radiation therapy planning against biological uncertainty

Abstract

A method for generating a robust radiotherapy plan for a treatment volume of an object, the treatment volume being defined using a plurality of voxels, the method comprising the following steps: defining (S100) an optimization problem using at least one optimization function for a biological endpoint associated with the radiotherapy; defining (S102) a set of scenarios, the set of scenarios comprising at least a first scenario and a second scenario, wherein at least two of the scenarios in the set of scenarios represent different biological models quantifying the same biological endpoint; calculating (S104) an optimization function value for each scenario in the set of scenarios; generating (S106) a radiotherapy plan by robustly optimizing the optimization function values ​​evaluated on the set of scenarios.

Description

Technical Field

[0001] The present disclosure relates generally to the field of radiation therapy, and more particularly to the generation, optimization, and evaluation of radiation therapy plans. Background Art

[0002] Cancer is a widespread disease that expresses itself through the erroneous growth of abnormal cells. If the uncontrolled growth of these cells is not stopped, it can be fatal. The global trend of cancer fatality is steadily increasing along with the total estimated cost of cancer management. This has led to an increase in public demand for the development of more effective tools and technologies for treating and curing the disease. With the rapid development of medical imaging, tumors are diagnosed in the early stages when they are still local or regional. Different types of radiation therapy (including brachytherapy and particle therapy, surgery and systemic therapy including chemotherapy) are effective in treating localized cancer cells or tumors.

[0003] Radiation therapy is a form of cancer treatment that uses ionizing radiation to damage DNA and cause cell death in the irradiated area. The primary goal of radiation therapy is to eradicate cancer cells by delivering a sufficiently high radiation dose to kill all targeted tumor cells while avoiding unacceptable damage to healthy structures. According to the International System of Units (SI), the physical quantity of energy damaged per unit mass (the so-called absorbed dose) is expressed in grays (Gy), where 1 Gy is equal to one joule of energy absorbed per kilogram of material.

[0004] Modern radiotherapy planning typically involves the following set of steps: patient imaging, target definition (i.e., structure contouring), dose prescription, machine and seed type selection, parameter definition, beam configuration selection, plan generation (usually in the form of plan optimization), and quality assurance and / or quality control.

[0005] In the field of radiotherapy planning, treatment planners can generate different types of radiotherapy plans for external beam radiotherapy. To generate a treatment plan, experts need to identify the area to be treated in the patient and assess the organs at risk (OARs) using traditional scanning techniques such as computed tomography (CT). The target volume is then defined as the volume within the patient's body (e.g., the prostate) affected by the tumor. Studies have shown that the quality of the treatment plan is closely related to the experience of the person performing the treatment, the quality of the imaging and treatment equipment, and many technical parameters such as beam quality. This suggests that many treatment plans have room for improvement, especially if they are developed by inexperienced personnel. In addition, using inappropriate planning methods, terminating the optimization process prematurely, or measuring plan quality in an inappropriate manner, as well as misunderstanding the parameter weightings, can significantly damage the quality of treatment.

[0006] Once the area to be treated is imaged, the physician delineates the tumor and OARs and prescribes the desired dose to treat the tumor. Medical physicists then begin plan generation and create a plan with one or more fields and beams to treat the patient. Significant manual effort is spent on plan optimization and quality control to minimize adverse effects on surrounding tissue and maximize plan effectiveness. This imposes additional costs on hospitals and society, and creates challenges in delivering radiation therapy on time.

[0007] Radiotherapy can use, for example, high-energy photons (X-rays) or proton beams or heavy ion beams (such as helium or carbon). Each type of radiation has its own characteristic distribution of energy deposition in matter, which distributions influence the physical, chemical and biological effects. For photons, cell damage is proportional to the energy absorbed (= dose). In addition, for protons and heavy ions, cell damage depends on the distribution of absorbed energy events, which is characterized by, for example, the ionization density of each particle along the ion trajectory. The ionization density depends in particular on the energy of the particle. The ionization density is usually quantified in terms of linear energy transfer (LET), i.e. the energy loss per distance (usually in units of MeV / cm or keV / pm). LET is low, for example, at high proton energies. LET increases towards lower proton energies, reaches a maximum and then decreases. The LET maximum is reached when the residual range of the proton is small and therefore appears in the Bragg peak.

[0008] Because tumor cells need to be killed as efficiently as possible, a high LET fraction is desirable inside the tumor. In healthy tissue outside the tumor, it is desirable to have the lowest possible dose and low LET.

[0009] The absorbed dose and spatial distribution of energy deposition are important parameters for the biological effects of radiation. However, many factors influence biological responses, including the inherent radiosensitivity of biological systems (cells, tissues, organs, etc.), the degree of oxygenation, the dose distribution within the irradiated volume, and the dose fractionation scheme (dose per fraction, time between fractions, total treatment time, etc.).

[0010] The goal of curative radiotherapy is to achieve local tumor control with a high probability without an unacceptably high risk of side effects. However, in treatment planning, clinical goals are often expressed in terms of physical quantities (such as the target prescribed dose and the dose-volume constraints of the region of interest), and the quality of a radiotherapy plan is often judged by its dose consistency, achievement of physical clinical goals, and treatment delivery time. Dose consistency describes how well the high radiation dose area conforms to the target tumor and spares surrounding healthy tissue, while treatment delivery time describes how long the treatment takes and the efficiency of the treatment machine used.

[0011] These quantities are surrogates for biological outcomes in the patient being treated, rather than directly estimating the probabilities of cure and side effects. Mathematical radiobiological models aim to bridge this gap by explicitly estimating the probability of tumor control (TCP) and the probability of normal tissue complications (NTCP), allowing the desired dose distribution to be determined based on the radiosensitivity of the tumor and organs at risk. In general, radiobiological models can be used, for example, to describe the relationship between radiation quantities and biological effects, to interpolate and extrapolate from known outcomes, to estimate the outcomes of new treatment techniques, to compensate for treatment interruptions, overdoses, or underdoses, and to assist in decision making. Unlike direct estimates of outcomes, biological models are often used in practice to convert non-standard dose distributions into conventional dose distributions. Extensive experience with the relationship between absorbed dose and clinical outcomes has been gained through long-term treatment using uniform intensity photon beams and fractionated doses of approximately 2 Gy. To leverage this experience in new treatment technologies, biological models have been used to convert proton and ion doses to equivalent photon doses using the relative biological effectiveness factor (RBE model), conversion of inhomogeneous dose distribution to equivalent uniform distribution (EUD-model), and conversion of specific fractionation schemes to equivalent standard fractionated doses (EQD model and BED model). These models can be used for plan optimization and assessment to include biological aspects.

[0012] The standard for proton and ion therapy planning is to scale the absorbed physical dose using an RBE factor derived from some RBE model to obtain a corresponding dose distribution for a reference radiation mass with the same level of biological damage, typically a photon energy spectrum. In proton therapy, simple scaling of the absorbed dose using a constant RBE factor of 1.1 is the clinical standard. Range uncertainty due to the increased RBE at the end of the proton range (along with other range uncertainties) is indirectly addressed by the choice of beam angle and margin. Other RBE models, not currently in clinical use, aim to more accurately estimate the RBE by including factors that influence the proton RBE, such as LET, dose per fraction, cell type / tissue type, and biological endpoints. There is active discussion about the appropriateness of using the simple 1.1 approach, but there is no consensus on which model to use, given the uncertainty surrounding how exactly the RBE depends on dose, LET, and tissue parameters. To circumvent the uncertainties of RBE models, other approaches that need to be considered are LET optimization and track-end optimization, which use purely physical quantities to redirect high-RBE protons from the OAR to the target(s) or elsewhere.

[0013] For carbon ions, the RBE variation across the beam is so high that a variable RBE model is required. Two models dominate the carbon ion field: the local effects model and the microdose kinetic model. These two different models are used clinically and result in different dose distributions.

[0014] To estimate the probability of curing the disease and the risk of normal tissue toxicity, other radiobiological models known as TCP and NTCP models are used. They are usually based on follow-up data of patient cohorts and describe mathematically the relationship between dose (possibly in combination with several other input parameters) and tumor control or normal tissue toxicity. The clinical use of TCP models is limited, while the use of NTCP models is slightly more common in the evaluation of treatment plans. For example, NTCP has been proposed to be used to select patient cohorts that are most suitable for proton therapy rather than traditional photon therapy. In order to utilize patient cohorts treated with various treatment techniques and different fractionation schedules, the non-uniform dose input in TCP and NTCP models can be converted to an equivalent uniform distribution (using the EUD model) and corrected for fractionation effects (using the EQD model).

[0015] Biological models and their parameter values ​​are subject to significant uncertainty, which limits their use in clinical practice. An important source of uncertainty is the variability of the experimental data on which the models are based. A general approach to dealing with uncertainty is scenario-based robust optimization. This technique has been used to mitigate other sources of uncertainty in radiotherapy, including the position of the target relative to the beam, the location of cancer cells, organ motion, and uncertainty in patient density data. The possible errors caused by the uncertainty are then discretized into different scenarios, each representing a specific configuration for one or more sources of uncertainty.

[0016] Then, an optimization problem is formulated, which includes an objective function and possible constraints, which describe non-negotiable conditions. The optimization algorithm aims to minimize (or maximize) the objective function while satisfying the constraints. The function used as the objective, a component of the objective, or a constraint in the optimization problem is denoted as an "optimization function."

[0017] For example, the optimization can minimize the objective function evaluated under the worst-case scenario, minimize the expected value of the objective function over all scenarios, or use another approach. Alternatively, the functions evaluated over different scenarios can be included as constraints in the optimization. Alternatively, the functions evaluated over different scenarios can be included as components of the objective function. For example, different robust optimization techniques are compared and discussed in the articles by Fredriksson, A. (2012). Characterization of robust radiotherapy planning methods from expected value to worst-case optimization. Medical Physics, 39(8), 5169-5181, and Fredriksson, A. and Bokrantz, R. (2014). Critical evaluation of worst-case optimization methods for robust intensity-modulated proton therapy planning. Medical Physics, 41. In this way, plans that are less sensitive to errors can be obtained compared to optimization conditioned on only one scenario. One object of the present invention is to use a scenario-based robust optimization framework to handle uncertainties arising from radiobiological models. Summary of the Invention

[0018] The object of the present invention is to provide an improved solution in which a similar robust optimization method is provided for handling uncertainties in biological models, wherein uncertainties in both model selection and parameter values ​​can be handled simultaneously. This object is achieved in a first aspect of the present invention in which a method for generating a robust radiotherapy plan for a volume of an object, the volume being defined using a plurality of voxels, is provided, the method comprising the following steps:

[0019] - defining an optimization problem using at least one optimization function for a biological endpoint related to radiation therapy;

[0020] - defining a scenario set, the scenario set comprising at least a first scenario and a second scenario, wherein at least two scenarios in the scenario set represent different biological models quantifying the same biological endpoint;

[0021] - Calculate the optimization function value for each scenario in the scenario set;

[0022] -Generates radiation therapy plans by robustly optimizing the value of an optimization function evaluated on a collection of scenarios.

[0023] By adding different radiobiological models and / or different parameter values ​​as different scenarios in a robust optimization setup, a treatment plan that is robust to the choice of both models and parameter values ​​can be obtained.

[0024] This disclosure proposes performing robust optimization using at least two different biological models, each with a different set of parameter values ​​as different scenarios, so as to achieve the best possible outcome for all configurations. In this way, the resulting radiation treatment plan is less sensitive to the choice of model and parameter values, but more robust to errors caused by model and / or parameter inaccuracies. These scenarios can also be combined with other scenarios defined for uncertainty, such as range uncertainty, setup errors, organ motion, and so on.

[0025] This approach offers distinct advantages over known methods that only consider physical uncertainties, namely, better exploitation of biological models by explicitly acknowledging the uncertainties involved, and the ability to include a combination of models and parameter values ​​in robust optimization and evaluation.

[0026] The concept includes robustness not only to parameter values ​​but also to model selection. Similar to how uncertainties in, for example, patient placement, density, and organ motion are included in robust optimization and evaluation, this concept could include several biological models, each with a range of parameter values ​​being treated as different scenarios in robust optimization and / or evaluation.

[0027] In a preferred embodiment, biological models for quantifying biological endpoints include equivalent uniform distribution (EUD), equivalent standard fractionated dose (EQD), biologically equivalent dose (BED), relative biological effectiveness (RBE), RBE-weighted dose, tumor control probability (TCP), normal tissue complication probability (NTCP), complication-free cure, secondary cancers, and / or overall survival. Any model designed to estimate the biological response to treatment can be used in accordance with the present disclosure.

[0028] In other preferred embodiments, the optimization problem includes constraints defining parameters that are maintained during optimization. Constraints can be, for example, in the form of predetermined doses in defined subvolumes (e.g., targets) that do not change during optimization. In this way, the target dose is maintained to ensure a certain dose distribution, but the rest of the radiotherapy plan is robustly optimized.

[0029] In advantageous embodiments, the optimization problem includes biological or physical objectives. Preferably, physical objectives include dose constraints to targets and organs at risk (OARs) in the treatment volume, dose volume histogram (DVH) constraints, LET constraints, particle stopping locations, and / or homogeneity and consistency indices. In this way, biological uncertainty can also be combined with other (physical) objectives. Plan optimization and evaluation should be able to be used in conjunction with different biological models and with physical optimization functions and objectives.

[0030] The optimization problem may be a combination of physical objectives (such as minimum and maximum dose to target and organs at risk, respectively, and DVH constraints) and biological objectives (such as EUD, TCP, and NTCP).

[0031] In a preferred embodiment, robust optimization comprises: a stochastic programming approach, in which the expected value of the optimization function is minimized; a minimax approach, in which the maximum value of the optimization function over error scenarios is minimized; or any combination of the two, commonly referred to as minimax stochastic programming; or a voxel worst-case approach, in which the worst-case dose for each voxel considered individually is optimized.

[0032] In an advantageous embodiment, the set of scenarios further comprises at least a third scenario, wherein the third scenario represents a specific realization of uncertainty in one or more parameters related to the treatment plan, including particle range, spatial position of the treatment volume, radiotherapy device settings, density of irradiated tissue, interaction effects, organ motion, and / or biological model parameter values. In this way, biological uncertainty can also be combined with other uncertainties associated with parameters related to the treatment plan.

[0033] In an alternative embodiment, the step of generating a radiation treatment plan includes adjusting a pre-existing radiation treatment. In this way, a "hot start" can be achieved by starting with an existing radiation treatment plan to arrive at a robustly optimized radiation treatment plan more quickly and with less computational load than starting from scratch. The existing radiation treatment plan can be any radiation treatment plan previously generated for the patient (or a standard radiation treatment plan generated using automatically derived default parameter values).

[0034] According to another aspect, there is provided a computer program product comprising computer readable code means which, when run in a computer, causes the computer to perform the method according to the first aspect.

[0035] According to a further aspect, there is provided a computer system comprising a processor coupled to a memory having computer readable instructions stored thereon, which when executed by the processor cause the processor to perform the method according to the first aspect.

[0036] According to another aspect, there is provided a treatment planning system comprising a computer system as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] These and other features, aspects and advantages of the present disclosure will be further explained in the following description with reference to the accompanying drawings, in which:

[0038] Figure 1shows a flowchart representing the steps of a computer-based method for generating a robust radiation therapy plan according to one embodiment of the present disclosure; and

[0039] Figure 2 A computer-based system for evaluating, visualizing, generating, and improving radiation therapy treatment plans according to one embodiment of the present disclosure is schematically illustrated.

[0040] Herein, where possible, identical reference numerals are used to designate identical elements that are common to the figures. Likewise, the images in the figures are simplified for illustrative purposes and are not necessarily drawn to scale. DETAILED DESCRIPTION

[0041] Figure 1 is a flow chart of an embodiment of a method according to the present invention, which can be used in conjunction with generating a radiation therapy treatment plan. In one embodiment, the starting point is an initial treatment plan and a plurality of scenarios to be considered, and the method aims to obtain an improved treatment plan based on the initial treatment plan, to modify the initial plan under some constraints, or to obtain a deliverable treatment plan if the initial treatment plan does not meet all machine limitations. Depending on the type of data contained in the plan, other input data (e.g., patient-related data) may be required for dose calculations. The initial treatment plan can be obtained in any manner known in the art, including scenario-based and non-scenario-based methods. Typically, it will be a previous plan prepared for the same patient (corresponding to a "hot start"), but can also be automatically obtained from a standard plan library as described above.

[0042] The purpose of generating a treatment plan is to provide radiation therapy to a treatment volume of a subject (patient), which may be an organ and includes a target, which may be a tumor or a cluster of tumor cells. As is known in the art, a plurality of voxels is used to define the treatment volume.

[0043] In a first step S100, an optimization function for biological endpoints associated with radiotherapy is defined. As described above, the biological endpoints are quantified based on a biological model to estimate the biological effect of radiation and may include, for example, maximum and / or minimum constraints or targets imposed on one or more of equivalent uniform distribution (EUD), equivalent standard fractionated dose (EQD), biologically equivalent dose (BED), relative biological effectiveness (RBE), RBE-weighted dose, tumor control probability (TCP), normal tissue complication probability (NTCP), complication-free cure, secondary cancers, and / or overall survival.

[0044] The optimization function can be included as a constraint in the optimization. Alternatively, the optimization function can be included as an objective function component in the optimization. Typically, goals are set for the treatment, and these goals are used to define objective function components, constraints, or a combination of these. Objective function components are desired goals that should be strived towards or optimized to the best of their ability, while constraints are strict goals or conditions that must be met precisely, such as a minimum dose to the tumor or a maximum dose to an OAR, or limits on variables that control the objective function.

[0045] In general, one or more scenarios can be used to define a first radiobiological target. For example, the first scenario can be based on a first radiobiological model, and the second scenario can be based on a second radiobiological model. In the event that one or both of the first and second radiobiological models have more than one set of parameter values, each set of parameter values ​​for each radiobiological model can generate a different scenario to be used in the methods according to the present disclosure. This principle can be further extended using additional radiobiological targets, thereby generating more scenarios based on different radiobiological models and parameter sets, as well as physical targets having one or more scenarios.

[0046] As an example, in proton therapy planning, one may choose to optimize a treatment plan with the goal of achieving some stated target based on RBE-weighted dose using multiple RBE models (including a constant RBE model), each with a different set of parameter values ​​for a different scenario, such that the target is achieved for all configurations as closely as possible. In this way, the plan is less sensitive to the choices made of the models and parameter values, but more robust to errors caused by model and / or parameter inaccuracies.

[0047] In robust optimization of proton plans, RBE-weighted doses can then be calculated using both a standard constant RBE model (RBE = 1.1) and, for example, RBE models based on different variable LETs, where each model includes a range of parameter values. In this way, the plan will not be strongly dependent on one model with nominal parameter values, but will incorporate uncertainties using, for example, worst-case optimization. This is described in more detail in Example 1 below.

[0048] Another example is to use different TCP and NTCP models for the same endpoint, where each model can have a different set of parameter values, and use another model and set of parameter values ​​for another endpoint. These biological models can be combined with other physical objectives. This is described in more detail in Example 2 below.

[0049] In step S102, a scenario set is defined that includes at least a first scenario and a second scenario. These scenarios represent the uncertainty of the biological model when quantifying biological endpoints related to radiation therapy. Scenarios can be defined manually or automatically. Several semi-automatic approaches to scenario definition are also contemplated. In a preferred embodiment, the user is allowed to set the uncertainty level as an input to the system, which then calculates an appropriate scenario set based on the uncertainty.

[0050] In step S104, the optimization function value is calculated for each scenario in the scenario set. In step S106, the optimization function value is robustly optimized and evaluated on the scenario set to generate a radiation therapy plan.

[0051] Various types of optimization methods for achieving robustness can be used in conjunction with the methods according to the present disclosure. For example, minimax (or "composite worst-case") optimization can be used, in which the worst-case scenario over a composite objective function is optimized. The optimization problem is then formulated as

[0052]

[0053] where X is the set of feasible optimization variables (e.g., the set of allowed point weights, MLC leaf locations, etc.), S is the set of scenarios enumerating different biological models, and

[0054] f(x;s)

[0055] is a composite objective as a function of the optimization variable x under scenario s. For example, f(x; s) can be given by g(d(x; s)), where g is a function related to the dose d(x; s) produced by the optimization variable x under scenario s. Here, s is a parameter that can completely change the function in question. For example, f(x; s1) may be the NTCP produced by the first NTCP model, and f(x; s2) may be the NTCP produced by the second NTCP model. Similarly, d(x; s1) may be the RBE-weighted dose produced by the first RBE model, and d(x; s2) may be the RBE-weighted dose produced by the second RBE model.

[0056] Another type of optimization method to achieve robustness is expected value optimization, in which the expected value of the uncertainty is optimized. The optimization problem is formulated as

[0057]

[0058] where E is the expectation operator and Y is a random variable taking values ​​from the set of scenarios S.

[0059] The third alternative is the voxel-wise worst-case optimization approach. In this approach, two artificial worst-case dose distributions, dhigh and d low Here, d high is calculated as the highest dose of the scene for each voxel considered individually, and d low is calculated as the minimum dose of the scene for each voxel considered individually, i.e.,

[0060]

[0061]

[0062] Among them, d i represents the dose to voxel i, and N is the number of voxels.

[0063] Then, the optimization problem is formulated as

[0064]

[0065] Among them, f high is a composite objective function whose components are used to avoid overdose (e.g., targets for organs at risk, OARs), and f low is a composite objective function whose components are used to avoid underdosing (eg, minimum dose requirement to a target).

[0066] Another alternative is to minimize an objective function h(x) that is not necessarily (but possibly) related to the complete set S of scenarios, and includes constraints on the function f(x;s) for all s in S, i.e.,

[0067]

[0068] subject to f(x;s)≤0,s∈S.

[0069] The objective function h(x) can be formulated according to any of the above methods, but can also be formulated to consider only nominal scenarios corresponding to no errors.

[0070] Other methods may also be used, such as stochastic minimax methods, which are a combination of compound worst-case optimization and expected value optimization, and are known in the art.

[0071] Now turn Figure 2, which shows a simplified schematic diagram of a computer-based system 100 for generating a radiation treatment plan 114 according to the present disclosure. The computer-based system 100 includes a memory or database 110 on which the radiation treatment plan 114 is stored, and a computer program 116 for generating an improved radiation treatment plan 118. The memory 110 can be any volatile or non-volatile memory device, such as a flash drive, a hard drive, an optical drive, a dynamic random access memory (DRAM), a static random access memory (SRAM), or any other suitable device for storing information and subsequently retrieving information for data processing. In addition, the system 100 includes one or more hardware processors 120 for performing data processing, which are capable of accessing the memory 110. The hardware processors 120 can be composed of one or more of a central processing unit (CPU), a digital signal processor (DSP), a reduced instruction set computer (RISC), an application-specific integrated circuit (ASIC), a complex programmable logic device (CPLD), a field programmable gate array (FPGA), a parallel processor system, or a combination of these different hardware processor types.

[0072] The computer program 116 is comprised of computer-readable instructions that can be delivered to and executed by the hardware processor 120. When executed on the hardware processor 120, the computer-readable instructions will perform a method for generating an improved radiation therapy plan 118. The results of the processing performed by the hardware processor 120 when executing the computer program 116 can be stored in the memory 110, such as the improved radiation therapy plan 118 and associated data. The hardware processor 120 can also access the memory 110 via direct memory access (DMA) and can also use cache memory to store temporary processing results. The computer program 116 can also be stored on a non-transitory computer-readable medium 130, such as a universal serial bus (USB) flash drive, an optical data carrier (such as a CD-ROM, DVD-ROM, and Blu-ray Disc), a floppy disk, an interchangeable hard drive, a USB external hard disk drive (HDD), or any other portable information storage device, so that the computer program 116 can be transported to different computing systems and also loaded into the memory 110 of the system 100. This may be accomplished by connecting the computer readable medium 130 to the system 100 via a data reader / writer 140 (eg, an optical drive, a USB port, etc.).

[0073] Furthermore, the system 100 further comprises a display unit 150 having a display driver that allows visualization of the results of the data processing, such as visualization of a three-dimensional (3D) representation of a target volume of a patient containing, for example, a tumor or cancer cells, and healthy organs at risk to which dose delivery must be prevented, 3D contour data, or two-dimensional (2D) slice representations (for various cross-directions and for the LET distribution in both the target volume and organs at risk) and biological effects (e.g., probability of injury / cell death / side effects), etc. For example, a 3D computer rendering of a CT scan can be displayed. Furthermore, the display unit 150 can display a dose volume histogram (DVH) summarizing the 3D dose distribution by using a graphical 2D format. For example, the display unit 150 is configured to show a patient volume (which shows the dose contribution of the radiotherapy plan 114) and a comparative DVH plot of the same volume of the optimized or improved radiotherapy plan 118, so that the LET distribution can be intuitively compared.

[0074] The display unit 150 is used to display a 3D scan of the patient taken before, during or after treatment. For example, a 3D computer reproduction of a CT scan can be displayed. In addition, the display unit 150 can display the LET, dose and / or DVH summarizing the 3D dose distribution, either using a graphical 2D format or using a digital format. For example, the display unit 150 is configured to show a comparative LET diagram of the patient volume, which shows the cancer cell destruction or dose contribution of the radiation therapy plan 114. Optimized or improved radiation therapy plans for the same volume are displayed and compared so that the improvements can be compared intuitively. In addition, it is possible that the display unit 150 is equipped with a touch screen function and can display a graphical user interface for operating the system 100.

[0075] Furthermore, the computer system 100 has a system bus 160 that connects the hardware processor 120, the memory 110, the data reader 140, a touch screen, and various other data input / output interfaces and peripheral devices (not shown). For example, the computer system 100 can be connected to a keyboard 170 for user data entry and to an external radiation therapy planning device 180, such as a powerful dedicated computer, that has created the radiation therapy plan. Furthermore, the system 100 can be connected to a CT scanner (not shown). For example, the external device 180 that creates the radiation therapy plan 114 may be capable of developing a dose and LET distribution calculation algorithm encoded in software, accessing radiation data regarding the prescribed dose distribution, machine calibration data, and patient-specific information regarding the patient's target volume and organs at risk. The external device 180 can then transmit the radiation therapy plan 114 to the computer system 100 for evaluation, visualization, creation of a new plan, and improvement of an existing plan to account for the LET distribution. However, the computer program 116 may also be run on the external device itself, generating not only the radiation therapy plan 114 but also an improved radiation therapy plan 118.

[0076] Furthermore, a computer program product for performing parameter optimization is introduced. The computer program product 130 comprises computer readable code means which, when run in a computer, perform the above method.

[0077] Example 1: Robust biooptimization for RBE-weighted dose

[0078] In radiotherapy using charged particles (proton therapy, carbon ion therapy, etc.), relative biological effectiveness (RBE) must be considered when prescribing doses to tumor(s) and organs at risk. Rather than using a dose, an RBE-weighted dose is used for this purpose: the dose in each voxel multiplied by the local RBE for that voxel. However, the RBE is a complex function of the particle's microscopic energy deposition characteristics, the local dose, tissue properties, the biological endpoint of interest, tissue oxidation, and more. Several models are available for calculating the RBE, and because experimental RBE data have substantial uncertainties and models are more or less influenced by biological mechanisms, the resulting RBE is highly model-dependent.

[0079] question : Which RBE model should be used to calculate RBE-weighted dose?

[0080] Proposed solution : Select at least two RBE models and robustly optimize the RBE-weighted dose to account for RBE uncertainties.

[0081] - Define a new treatment plan or start from a pre-optimized treatment plan using any optimization method.

[0082] - Define at least one objective for the RBE-weighted dose that one wishes to optimize robustly with respect to uncertainties in the RBE. Examples of such an objective could be:

[0083] - Minimum or maximum RBE-weighted dose to the tumor

[0084] -Maximum RBE-weighted dose to OAR

[0085] - Maximum mean RBE-weighted dose to OARs

[0086] - Optionally, add other objectives or constraints to the composite objective function.

[0087] - Select at least two different radiobiological models to calculate RBE.

[0088] - Scenarios are defined for each selected RBE model, where each scenario then represents a scenario for robust optimization.

[0089] - Robustly optimize treatment plans using a preferred robust optimization framework.

[0090] Example 2: Robust biooptimization targeting TCP and / or NTCP

[0091] In radiotherapy, the physical quantity of energy applied per unit mass, the so-called absorbed dose, is often used as a surrogate for biological effect. Therefore, radiotherapy plans are usually optimized in terms of dose, but biological effect is the primary quantity of interest.

[0092] However, direct optimization of biological effects can be achieved through radiobiological models that use tumor control probability (TCP) and normal tissue complication probability (NTCP). However, due to the significant uncertainty in clinical data on TCP and NTCP, several radiobiological models exist to calculate the same biological endpoint (TCP or NTCP for a specific biological effect). In addition, various models also consider the effects of factors such as smoking, diabetes, age, and gender in addition to dose.

[0093] question : Which TCP and / or NTCP model should be used in the biological optimization of radiotherapy treatment plans?

[0094] Proposed solution : A minimum of two radiobiological models were selected for the same biological endpoint and robustly optimized to account for biological uncertainties.

[0095] - Define a new treatment plan or start from a pre-optimized treatment plan using any optimization method.

[0096] - define at least one objective based on the radiobiological model. Examples of such objectives could be:

[0097] ○ The NTCP of a biological endpoint should be minimized or below a certain probability.

[0098] ○ TCP should be maximized or above a certain probability.

[0099] - Optionally, add other objectives or constraints to the composite objective function.

[0100] - Select at least two different radiobiological models to calculate TCP and / or NTCP.

[0101] - Scenarios are defined for each selected radiobiological model, where each scenario then represents a scenario for robust optimization.

[0102] - Robustly optimize treatment plans using a preferred robust optimization framework.

[0103] Preferred embodiments of the method and system for generating a radiation therapy plan have been disclosed above. However, a person skilled in the art will appreciate that this may be varied within the scope of the appended claims without departing from the spirit of the invention.

[0104] Without departing from the concept of the present invention, all the alternative embodiments or parts of the embodiments described above may be freely combined or used separately from each other as long as the combination is not contradictory.

[0105] The following abbreviations are used:

[0106] BED biologically equivalent dose

[0107] CT computed tomography

[0108] CTV clinical tumor volume

[0109] DICOM Digital Imaging and Communications in Medicine

[0110] DVH Dose Volume Histogram

[0111] EHR electronic health record system

[0112] EQD Equivalent standard fractionated dose

[0113] EUD Equivalent Uniform Distribution

[0114] eMIX Electronic Medical Information Exchange System

[0115] GUI Graphical User Interface

[0116] GTV total tumor volume

[0117] HIS Hospital Information System

[0118] HIM Health Information Management System

[0119] IMRT Intensity Modulated Radiation Therapy

[0120] LET Linear Energy Transfer

[0121] MLC Multi-Leaf Collimator

[0122] MRI Magnetic Resonance Imaging System

[0123] MU Monitoring Unit

[0124] NTCP Normal Tissue Complication Probability

[0125] OAR Organs at Risk

[0126] PBS Pencil Beam Scanning

[0127] PET Positron Emission Tomography

[0128] PTV planned tumor volume

[0129] QA Quality Assurance

[0130] QC Quality Control

[0131] Ultrasound examination

[0132] RBE relative bioavailability

[0133] ROI Area of ​​Interest

[0134] RVS Recording and Verification System

[0135] SPECT single-photon positron emission tomography

[0136] TCP tumor control probability

Claims

1. A computer program product (130) comprising computer-readable instructions which, when executed on a computer, cause the computer to perform a method for generating a robust radiation therapy plan for a treatment volume of a subject, the treatment volume being defined using a plurality of voxels, the method comprising the steps of: - defining (S100) an optimization problem using at least one optimization function for a biological endpoint related to said radiation therapy; - defining (S102) a scenario set, the scenario set comprising at least a first scenario and a second scenario, wherein at least two scenarios in the scenario set represent different biological models for quantifying the same biological endpoint, wherein the biological models for quantifying the biological endpoint comprise equivalent uniform distribution (EUD), equivalent standard fractionated dose (EQD), biologically equivalent dose (BED), relative bioeffectiveness (RBE), RBE-weighted dose, tumor control probability (TCP), normal tissue complication probability (NTCP), complication-free cure, secondary cancer and / or overall survival; - calculating (S104) an optimization function value for each scene in the scene set; - generating (S106) a radiation therapy plan by robust optimization of the optimization function value evaluated on the set of scenarios.

2. The computer program product according to claim 1, wherein The optimization problem includes constraints defining parameters that are maintained during the optimization.

3. The computer program product according to claim 1 or 2, wherein: The optimization problem includes a biological objective or a physical objective.

4. The computer program product according to claim 3, wherein: The physical objectives include dose constraints to targets and organs at risk (OARs) in the treatment volume, dose volume histogram (DVH) constraints, linear energy transfer (LET) constraints, locations of particle stopping, and / or homogeneity and uniformity indices.

5. The computer program product according to claim 1 or 2, wherein: The robust optimization includes: a stochastic programming method, in which the expected value of the optimization function is minimized; a minimax method, in which the maximum value of the optimization function over error scenarios is minimized; or any combination of these two, known as minimax stochastic programming; or a voxel worst-case method, in which the worst-case dose of each voxel considered individually is optimized.

6. The computer program product according to claim 1 or 2, wherein: The set of scenarios also includes at least a third scenario, wherein the third scenario represents a specific realization of uncertainty in one or more parameters related to the treatment plan, including particle range, spatial position of the treatment volume, radiotherapy equipment settings, density of irradiated tissue, interaction effects, organ motion and / or biological model parameter values.

7. The computer program product according to claim 1 or 2, wherein: The step of generating a radiation therapy plan includes adjusting a pre-existing radiation therapy.

8. A computer system (100) comprising a processor (120) coupled to a memory (110) having computer-readable instructions stored thereon, the computer-readable instructions, when executed by the processor, causing the processor to perform a method for generating a robust radiation therapy plan for a treatment volume of a subject, the treatment volume being defined using a plurality of voxels, the method comprising the steps of: - defining (S100) an optimization problem using at least one optimization function for a biological endpoint related to said radiation therapy; - defining (S102) a scenario set, the scenario set comprising at least a first scenario and a second scenario, wherein at least two scenarios in the scenario set represent different biological models for quantifying the same biological endpoint, wherein the biological models for quantifying the biological endpoint comprise equivalent uniform distribution (EUD), equivalent standard fractionated dose (EQD), biologically equivalent dose (BED), relative bioeffectiveness (RBE), RBE-weighted dose, tumor control probability (TCP), normal tissue complication probability (NTCP), complication-free cure, secondary cancer and / or overall survival; - calculating (S104) an optimization function value for each scene in the scene set; - generating (S106) a radiation therapy plan by robust optimization of the optimization function value evaluated on the set of scenarios.

9. The computer system according to claim 8, wherein: The optimization problem includes constraints defining parameters that are maintained during the optimization.

10. The computer system according to claim 8 or 9, wherein: The optimization problem includes a biological objective or a physical objective.

11. The computer system according to claim 10, wherein: The physical objectives include dose constraints to targets and organs at risk (OARs) in the treatment volume, dose volume histogram (DVH) constraints, linear energy transfer (LET) constraints, locations of particle stopping, and / or homogeneity and uniformity indices.

12. The computer system according to claim 8 or 9, wherein: The robust optimization includes: a stochastic programming method, in which the expected value of the optimization function is minimized; a minimax method, in which the maximum value of the optimization function over error scenarios is minimized; or any combination of these two, known as minimax stochastic programming; or a voxel worst-case method, in which the worst-case dose of each voxel considered individually is optimized.

13. The computer system according to claim 8 or 9, wherein: The set of scenarios also includes at least a third scenario, wherein the third scenario represents a specific realization of uncertainty in one or more parameters related to the treatment plan, including particle range, spatial position of the treatment volume, radiotherapy equipment settings, density of irradiated tissue, interaction effects, organ motion and / or biological model parameter values.

14. The computer system according to claim 8 or 9, wherein: The step of generating a radiation therapy plan includes adjusting a pre-existing radiation therapy.

15. A radiation therapy planning system comprising the computer system (100) according to any one of claims 8 to 14.

Images