Methods and systems related to radiotherapy treatment planning

Abstract

The present disclosure relates to using machine learning to determine initial machine setting parameters for radiation therapy treatment planning. A machine learning system is trained with a dataset including dose distributions and sets of machine parameter settings that resulted in the dose distributions. The trained system can be used to determine machine parameter settings based on a desired dose distribution, which can be used as initial machine parameter settings for radiation therapy optimization.

Description

TECHNICAL FIELD

[0001] The present invention relates to radiotherapy treatment planning optimization, and in particular to parameter initialization for such optimization processes. BACKGROUND

[0002] In radiotherapy treatment planning optimization, given a set of variables that typically include machine parameters of the radiotherapy treatment device to be used, an optimization problem is set up and a radiotherapy treatment plan is optimized to achieve a desired dose distribution in a patient. The initial values of these parameters can be set up in different ways.

[0003] For photon therapy, the machine parameters typically include MLC leaf sequencing. These initial machine parameter values can be determined by solving a fluence map optimization problem and then performing a conversion from the optimized fluence map dose to feasible machine parameters. This involves target projection, fluence map optimization and conversion to machine parameters. This is a time consuming process and the conversion step is often a source of inaccuracy in the planning process, which leads to long optimization times as multiple iterations are needed.

[0004] For ion therapy, such as proton therapy, the machine parameters include spot placement, spot weights and beam energy. For example, for pencil beam scanning, the initial values of these parameters can be set up in a number of different ways and the implementation varies from one clinic to another. One possible implementation involves calculating a target projection and then using some mathematical formula to decide the initial values of the spot weights.

[0005] The present disclosure aims at making the treatment planning optimization process faster and enabling better treatment plans to be produced by the treatment planning optimization process. SUMMARY

[0006] The present disclosure relates to the use of machine learning for determining initial machine setup parameters for radiotherapy treatment planning. Thus, the present disclosure relates to a computer-based method of training a machine learning system, comprising inputting a plurality of data sets to the machine learning system, each data set comprising one or more dose distributions and a set of machine parameter settings, the set of machine parameter settings comprising at least one machine parameter setting resulting from the one or more dose distributions in a planning process, to train the machine learning system to output at least one set of machine parameter settings based on a reference dose distribution.

[0007] The present disclosure also relates to a machine learning system trained according to the above method. The machine learning system is arranged to take input data in the form of one or more dose distributions and to output at least one machine parameter setting that will be adapted to produce a dose distribution for a particular radiotherapy delivery device.

[0008] The present disclosure also relates to a computer-based method of determining machine parameter settings using such a machine learning system. The method comprises the steps of: inputting one or more reference dose distributions into the machine learning system; performing parameter initialization by the machine learning system; and outputting a set of machine parameter settings from the machine learning system, the set of machine parameter settings comprising at least one machine parameter setting for a radiotherapy delivery device. By basing the machine parameter settings on knowledge about suitable machine parameter settings for similar dose distributions, a better set of machine parameter settings can be obtained.

[0009] Thus, according to the present invention, more correct input data about machine parameters of a radiotherapy delivery device can be obtained in an efficient way by machine learning. This means that the planning optimization time can be shortened, since the initial data will be more correct. The method is particularly useful for machine parameter initialization for use in machine learning based optimization, but also for any other type of optimization process.

[0010] For photon therapy, the set of machine parameter settings can comprise one or more of MLC leaf settings, MU settings, start and stop angles, couch angle and pitch. For intensity modulated radiotherapy (IMRT) applications, the set of machine parameter settings can comprise one or more of segmented MLC (SMLC) or dynamic MLC (DMLC). For ion therapy, e.g. proton therapy, the set of machine parameter settings can comprise one or more of e.g. beam spot placement, beam spot weights and beam energy.

[0011] The present disclosure also relates to a computer-based radiotherapy treatment planning optimization method, comprising: prior to performing the planning optimization, performing the method of determining a set of machine parameter settings according to any of the above outlined embodiments, and using the resulting set of machine parameter settings as initial settings for the corresponding machine parameters in the radiotherapy treatment planning optimization.

[0012] The one or more machine parameter settings determined above can be used in any optimization process that can generate reference doses in order to guide the prediction of initial machine parameters to be used as optimization variables. A few examples of such optimization processes are:

[0013] • a dose simulation optimization implemented in a machine learning planning, where a machine learning model predicts a reference dose, which is then converted into a deliverable dose by a dose simulation optimization.

[0014] • a multi-criteria optimization (MCO), where a reference dose is calculated from fluence maps, which is then optimized using a dose simulation by simultaneously optimizing more than one objective function.

[0015] • In case the plan has been optimized for delivery by one delivery machine, recreate the same dose distribution for delivery by a different delivery machine by using the original plan as a reference dose distribution for dose simulation optimization.

[0016] • Regular inverse planning for volumetric modulated arc therapy (VMAT) where the optimized fluence map is used to calculate the reference dose (in this case the present invention replaces the leaf conversion algorithm).

[0017] The method according to the present disclosure can be used for any treatment modality, including photon-based and ion-based modalities, such as proton therapy, or carbon or helium ion therapy.

[0018] The present disclosure also relates to a computer program product comprising computer readable code means, which, when run in a computer, will cause the computer to execute any of the methods above outlined. The computer program product can comprise a non-transitory storage means having the computer readable code means stored thereon. The present disclosure also relates to a computer comprising a processor and a program memory having the computer program product stored thereon. BRIEF DESCRIPTION OF DRAWINGS

[0019] The present invention is described in more detail below, by way of example and with reference to the drawings.

[0020] Figure 1 The steps of obtaining machine parameter settings according to the present disclosure are generally illustrated.

[0021] Figure 2 The steps of obtaining machine parameter settings for photon-based treatment according to an embodiment are illustrated.

[0022] Figure 3 is a flowchart of a treatment planning optimization method using the method for parameter optimization of the present invention. DETAILED DESCRIPTION

[0023] The present disclosure relates to a method of training a machine learning system to return suitable machine parameter settings for a radiotherapy treatment plan based on one or more reference doses. Such machine parameter settings reflect the capabilities of the radiation delivery device used to deliver the plan. For a photon-based plan, the parameter settings can be related to MLC leaf positions and / or other parameters such as monitor units (MUs), start and stop angles, couch angles and pitch, where pitch is the possible tilt of the patient support. For a proton or other ion-based plan, other parameter settings can be relevant such as beam spot positions, beam spot weights and beam energies. The present disclosure also relates to a method for obtaining machine parameter settings for a radiotherapy treatment plan using a machine learning system trained in this way. The machine learning system can be any type of machine learning system, including neural networks such as deep neural networks.

[0024] Which machine parameter settings are of particular interest depends on the delivery system to be used for delivering the plan. The machine learning system can be adapted to determine all applicable machine parameter settings or only a subset of them. Some machine parameter settings not determined by the machine learning system must be provided by some other method.

[0025] The parameter settings returned from the trained neural network can be used as initial values for a treatment planning process. The treatment planning process can be any suitable planning method such as dose simulation or a traditional optimization process using an optimization problem. Such processes are known in the art and typically use an optimization problem designed to ensure that a suitable dose distribution corresponding to the one or more reference dose distributions is located on the treatment region, possibly together with other objectives such as treatment time. The machine parameter settings that have been determined based on the one or more reference dose distributions are used as initial values for the optimization variables.

[0026] For application in a dose simulation process, the input data to the optimization process comprises one or more reference dose distributions, where each reference dose distribution is a 3D volume with dose values corresponding to a desired dose distribution for a patient. The dose distributions can be produced in any suitable way. For example, the reference doses can be ML predicted doses, fluence map optimized doses or doses obtained from another plan or treatment technique to be simulated.

[0027] The input data set for the training process of the machine learning system comprises a set of machine parameter settings and the dose distribution resulting from each set of machine parameter settings, which can for example be a clinical plan. Thus, the machine parameter settings constitute a solution that is considered correct in view of the dose distribution of the same input data set.

[0028] Figure 1An example of the process by which machine parameter settings are obtained by the machine learning system 11 is shown. As will be appreciated, the nature of the machine parameter settings returned from the process will differ depending on the treatment modality and the type of delivery device to be used in the treatment. On the left is a reference dose distribution 13 which will be used as input data for the machine learning system 11. This is the dose distribution that the patient should ideally receive. In this example, the reference dose distribution is given as a slice of dose. The machine learning system is a Unet model. The functioning of such a model is well known in the art.

[0029] Figure 2 An example for a light-based treatment is shown. As in Figure 1 the machine learning system 21 receives input data in the form of a reference dose distribution 23 which is the desired dose distribution for treating a patient, and returns machine parameter settings. In Figure 2 a subset of the parameter settings is returned in the form of MLC leaf positions 25. In this example, the output from the machine learning system is a leaf opening matrix, shown schematically with reference 15, which defines the leaf positions of the MLC to be used in the treatment planning optimization.

[0030] In this example, the MU settings are required but are not determined by the machine learning system. Thus, a MU optimization 27 is performed based on the machine parameter settings, returning initial values for the MU. Alternatively, the initial MU settings can be predicted or determined in any other suitable way and used for the subsequent optimization.

[0031] Figure 3 The overall flow of the treatment planning optimization incorporating the steps according to embodiments of the application is shown. First, the steps corresponding to the disclosure of Figure 1 are performed. A reference dose S31 corresponding to the desired dose distribution for a patient is input into a machine learning system which has been trained for a parameterization S32 as outlined above. The output S33 from the parameterization process includes machine parameter settings for the delivery device to be used in the treatment. These output machine parameter settings and an optimization problem S34 are used as input for an optimization process S35. The optimization problem S34 and the optimization process S35 can be as conventionally used in the art. Thus, the initial steps S31-S33 replace conventionally performed steps to provide machine parameter settings, including target projection, fluence map optimization and leaf sequencing. As will be appreciated, the use of a machine learning system which is able to provide these parameter settings S33 directly from the reference dose saves time and effort.

[0032] The optimization process in step S35 can rely on one of the following methods:

[0033] • Dose simulation optimization implemented in a machine learning plan, where a machine learning model predicts a reference dose, which is then converted into a deliverable dose by dose simulation optimization.

[0034] • Multi-criteria optimization (MCO), where a reference dose is calculated from fluence maps, which is then optimized by dose simulation by simultaneously optimizing more than one objective function.

[0035] • In case a plan has been optimized for delivery by one delivery machine, recreating the same dose distribution for delivery by a different delivery machine by using the original plan as a reference dose distribution for dose simulation optimization.

[0036] • Regular inverse planning for volumetric modulated arc therapy (VMAT), where the optimized fluence maps are used to calculate a reference dose (in this case, the present invention replaces the leaf conversion algorithm).

[0037] Figure 4 is a schematic diagram of a computer arranged for performing one or more of the methods according to the present disclosure. The computer 41 comprises a processor 43, a data memory 44 and a program memory 45. Preferably, there is also one or more user input devices 47, 48 in the form of a keyboard, a mouse, a joystick, a voice recognition device and / or any other available user input device. The user input devices can also be arranged for receiving data from an external memory unit.

[0038] The program memory 45 holds a computer program arranged for controlling the processor to perform the processes. As with the data memory 44, the program memory can also be implemented as one or more units, if appropriate. The data memory 44 holds input data usable in the respective processes, and outputs data resulting from the planning. The input data for the training process comprises a training data set. The input data for the method of determining machine parameter settings comprises one or more reference dose distributions. The input data for the method of optimizing a plan comprises objective functions derived from, for example, a reference dose distribution and clinical objectives.

Claims

1. A computer-based method for training a machine learning system, comprising inputting a plurality of datasets into the machine learning system, each dataset including one or more dose distributions and a set of machine parameter settings, the set of machine parameter settings including at least one machine parameter setting generated by the one or more dose distributions during a planning process, thereby training the machine learning system to output at least one set of machine parameter settings based on a reference dose distribution.

2. A machine learning system trained according to the method of claim 1, the machine learning system being configured to acquire input data in the form of one or more reference dose distributions and output a set of machine parameter settings, the set of machine parameter settings including at least one machine parameter setting adapted to generate the one or more reference dose distributions for a particular radiotherapy delivery device.

3. A computer-based method for determining machine parameter settings using the machine learning system according to claim 2, comprising the following steps: One or more reference dose distributions are input into the machine learning system, the machine learning system initializes the parameters, and the machine learning system outputs a set of machine parameter settings for the radiotherapy delivery device.

4. The method according to claim 3, wherein, The machine parameter settings set includes MLC leaf settings.

5. The method according to claim 3 or 4, wherein, The machine parameter settings set includes MU settings.

6. The method according to claim 3, wherein, The machine parameter settings set includes one or more of the following: beam placement, beam weight, and beam energy.

7. A computer-based method for optimizing radiotherapy treatment planning, comprising: Before performing the planning optimization, the method for determining machine parameter settings according to any one of claims 3-6 is performed, and at least one of the obtained machine parameter settings is used as the initial settings of the machine parameters in the radiotherapy treatment planning optimization.

8. The method for optimizing radiotherapy treatment planning according to claim 7, wherein, The planning optimization is performed by optimizing the optimization problem.

9. The method for optimizing radiotherapy treatment planning according to claim 8, wherein, The planning optimization is performed through dose simulation.

10. A computer program product comprising a computer-readable code means, which, when executed in a computer, causes the computer to perform the method according to claim 1 or the method according to any one of claims 3-9.

11. A computer program product comprising a non-transitory storage device on which computer-readable code means are stored, the computer-readable code means, when executed in a computer, causing the computer to perform the method according to claim 1 or the method according to any one of claims 3-9.

12. A computer system comprising a processor and a program memory, wherein the program memory stores a computer program product according to claim 10.

Images