Radiation therapy planning system and computer-implemented method for planning radiation therapy

Abstract

The invention relates to a radiation therapy planning system and a computer-implemented method for planning radiation therapy. A radiation therapy planning system includes a processor configured to: construct a training dataset using simulated machine scan data measured by a radiation therapy device programmed with known values of beam model parameters in radiation simulation; modifying the training data set with additive noise that simulates the machine scan data; training a deep learning model using the modified training data set, the deep learning model being trained to establish a relationship between the modified training data set and known values of beam model parameters; and generating a beam model for the radiation therapy plan by applying the trained deep learning model to machine scan data collected from the radiation therapy device and indicative of a configuration or operating state of the radiation therapy device; and a memory circuit configured to store the generated beam model.

Description

[0001] This application is a divisional application of the application with national application number 202080103747.3, international application date September 2, 2020, national entry date March 2, 2023, and invention name “Automatic beam modeling based on deep learning”. Technical Field

[0002] This document relates generally to dose calculations in radiation therapy treatment systems and, more particularly, to systems and methods for automatically generating beam models used in radiation therapy treatment planning. Background Art

[0003] Radiation therapy (or "radiotherapy") can be used to treat cancer or other diseases in mammalian (e.g., human and animal) tissue. One such radiation therapy technique uses a linear accelerator (also called a "linac") to irradiate a tumor with high-energy X-rays or electrons. The goal of radiation therapy is to maximize the radiation dose to the target tissue (e.g., a tumor or other abnormal tissue) while minimizing damage to surrounding healthy tissue. Similar to the prescription of a drug, a physician prescribes a predetermined amount of radiation dose to the target (tumor or other abnormal tissue) and clinical dose limits for surrounding organs. Typically, ionizing radiation in the form of a collimated beam is directed at the patient from an external radiation source. The radiation beam can be precisely controlled to ensure dose delivery.

[0004] Specified or selectable beam energies may be used for delivering a range of diagnostic energy levels or a range of therapeutic energy levels. Modulation of the radiation beam may be provided by one or more attenuators or collimators such as a multi-leaf collimator (MLC) and jaws. The intensity and shape of the radiation beam may be adjusted by the collimator to avoid damaging healthy tissue adjacent to the target tissue, for example by conforming the projected beam to the contours of the target tissue.

[0005] Treatment planning is a process involving the following: determining specific radiotherapy parameters for achieving treatment goals under constraints. Examples of radiotherapy parameters include radiation beam angles, dose intensity levels, dose distributions, etc. The radiation dose can be calculated using a dose calculation algorithm. The outcome of the treatment planning process is a radiation therapy treatment plan (also referred to as a "treatment plan" or simply a "plan"). A treatment plan can be developed using a treatment planning system (TPS). A treatment plan is custom designed for each patient before radiotherapy can be delivered to the patient. In order to create a plan, one or more medical imaging techniques, such as images from X-rays, computed tomography (CT), magnetic resonance (MR), positron emission tomography (PET), single photon emission computed tomography (SPECT), or ultrasound, can be used to provide an image of the target tumor. Clinicians can use images of the patient's anatomy to identify the target tumor and surrounding organs near the tumor, depict targets to receive a prescribed radiation dose, and similarly depict nearby tissues such as organs (OARs) that are at risk of being damaged by radiation therapy. Delineating or outlining the target tumor and OARs can be done manually or by using automated software tools that aid in identifying or outlining the target tumor and OARs. Optimization techniques can then be used to generate a radiation therapy treatment plan based on clinical and dosimetric goals and constraints (e.g., maximum dose, minimum dose, mean dose, and partial dose for a fractional tumor volume or the entire tumor volume, and similar measures for critical organs).

[0006] The dose distribution for a treatment plan can be generated using a dose calculation algorithm. The beam model used for the algorithm includes parameters that describe the energy distribution of radiation emitted from a radiation therapy machine (e.g., a linac). The beam model parameter values ​​vary from one radiation therapy machine to another, and even the same model of radiation therapy machine from the same manufacturer requires different beam model parameters. Mechanical differences (e.g., mechanical dimensions or material characteristics) or differences in component values ​​(e.g., electronic circuit component values) can result in differences in beam model parameter values ​​between different radiation therapy machines. Medical physicists can perform beam modeling by manually adjusting the model parameters so that the beam model output dose distribution closely matches the measurements from the machine. Summary of the invention

[0007] Treatment planning is the process of determining the parameters of radiotherapy for treating a patient via a radiotherapy device. The treatment planning process may include using a three-dimensional (3D) image of the patient to identify a target area (e.g., a tumor) and identifying critical organs near the tumor. The creation of a treatment plan may be a time-consuming process in which the planner attempts to comply with various treatment goals or constraints (e.g., dose volume histogram (DVH), overlap volume histogram (OVH)), consider their respective importance (e.g., weight) in order to produce a clinically acceptable treatment plan. The treatment plan consists of digital parameters that specify the direction, cross-sectional shape, and intensity of one or more radiation beams. Once a treatment plan is generated, the treatment plan may be executed by positioning the patient in a treatment machine and delivering a prescribed radiotherapy guided by the optimized plan parameters. In some examples, a radiotherapy treatment plan may include a dose "division" so that a series of radiotherapy treatments (e.g., there may be 30 to 45 portions during the patient's treatment) may be provided to the patient over a predetermined period of time, wherein each treatment includes a specified portion of the total prescribed dose. During treatment, the position of the patient relative to the treatment machine (eg, linac) and the position of the target tumor relative to the treatment machine (eg, linac) are important considerations to ensure that the target tumor is irradiated and healthy tissue is not irradiated.

[0008] For a treatment planning system (TPS), in order to use a dose calculation algorithm to calculate the dose for a specific linac, a beam model of the dose algorithm needs to be created for the linac. The generation of the beam model is part of the TPS commissioning. The beam model may include parameters that describe, among other things, the energy distribution of the radiation emitted from the radiotherapy machine, which are key inputs for the TPS to calculate the dose distribution. In some methods, after the radiotherapy machine is installed and final adjustments are performed, the customer may use a phantom (e.g., a water phantom) for beam modeling reference to perform measurements of the radiation beam under different settings.

[0009] An important aspect of treatment planning and beam modeling is determining beam model parameter values. Conventional beam modeling typically requires a human modeler (e.g., a modeling physicist) to manually adjust beam model parameters and apply the beam model with the manually adjusted model parameters to a dose engine to calculate dose metrics (e.g., dose distribution or dose profile). In a virtual phantom, a human expert (e.g., the same human modeler, or a different person such as a radiation oncologist or medical physicist) can verify that the calculated dose meets predetermined dosimetry verification criteria. The dosimetry verification criteria can be based on an analysis of a limited number of points in a low dose gradient region, or based on measurements of distances between isodose lines in a high dose gradient region. The human expert can compare the expected dose and the measured results by placing a detector (e.g., an ionization chamber) in a phantom (e.g., a water phantom) to visualize their differences and check for any differences or verify consistency. If the calculated dose does not meet the predetermined dosimetry verification criteria, further manual adjustments can be made to the beam model parameters until the dose metrics calculated based on the beam model meet the predetermined dosimetry verification criteria.

[0010] Manually adjusting beam model parameters and generating corresponding beam models may be time-consuming and inefficient. In some cases, the task of a human modeler is, for example, to generate separate beam models for different energy levels (e.g., 6MV, 10MV, or 15MV) and / or for different types of collimators. Manual adjustment and testing of beam model parameters may take a long time to complete and require a high degree of effort. Additionally, in some cases, manual adjustments may be inconsistent and lead to interpretation errors, particularly for dose plans for more advanced radiotherapy techniques (e.g., intensity modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT)), which involve more complex dose measurements and distribution representations and more complex dose calculations. For example, the integrity and complexity of IMRT dose delivery technology depends on quantifying the consistency of the planned IMRT dose distribution and the delivered IMRT dose distribution. Therefore, manual beam modeling and parameter adjustment may be particularly time-consuming and cumbersome.

[0011] This document discusses systems, devices, and methods for generating beam models using artificial intelligence (AI) based methods. An exemplary system includes: a memory for storing a trained deep learning model; and a processor circuit for generating a beam model. The deep learning model can be trained to establish a relationship between machine scan data and beam model parameter values, and verified to achieve accuracy. The processor circuit can: receive machine scan data indicating a configuration or operating state of a radiation therapy device; apply the machine scan data to the trained deep learning model to determine values ​​of beam model parameters; and generate a beam model based on the determined values ​​of multiple beam model parameters. The beam model can be provided to a user or a treatment planning system.

[0012] The present application discusses systems and methods for generating beam models for radiation therapy treatment plans. An exemplary system includes: a memory for storing a trained deep learning model; and a processor circuit for generating a beam model. The deep learning model can be trained to establish a relationship between machine scan data and beam model parameter values, and verified to achieve accuracy. The processor circuit can: receive machine scan data indicating a configuration or operating state of a radiation therapy device; apply the machine scan data to the trained deep learning model to determine values ​​of beam model parameters; and generate a beam model based on the determined values ​​of multiple beam model parameters. The beam model can be provided to a user or a treatment planning system.

[0013] According to an embodiment of the present disclosure, a radiotherapy planning system is provided, comprising: a processor configured to: construct a training data set using simulated machine scan data measured by a radiotherapy device programmed with known values ​​of beam model parameters in a radiation simulation; modify the training data set with additive noise of the simulated machine scan data; train a deep learning model using the modified training data set, the deep learning model being trained to establish a relationship between the modified training data set and the known values ​​of the beam model parameters; and generate a beam model for a radiotherapy plan by applying the trained deep learning model to machine scan data collected from the radiotherapy device and indicating a configuration or operating state of the radiotherapy device; and a memory circuit configured to store the generated beam model.

[0014] According to an embodiment of the present disclosure, a computer-implemented method for planning radiotherapy is provided, the method comprising: constructing a training data set using simulated machine scan data measured by a radiotherapy device programmed with known values ​​of beam model parameters in a radiation simulation; modifying the training data set with additive noise of the simulated machine scan data; training a deep learning model using the modified training data set, the deep learning model being trained to establish a relationship between the modified training data set and the known values ​​of the beam model parameters; and generating a beam model for a radiotherapy plan by applying the trained deep learning model to machine scan data collected from the radiotherapy device and indicating a configuration or operating state of the radiotherapy device.

[0015] AI-based beam modeling according to various examples discussed in this document can improve the efficiency and quality of beam modeling in advanced radiotherapy. Compared with conventional manual modeling methods, automatic beam modeling based on deep learning (DL) as discussed in this article can shorten the model development and deployment cycle, reduce the workload of human modelers, and improve model accuracy and treatment planning efficiency. Compared with manual dose modeling that requires a large number of dose calculations for each manually adjusted value of the model parameters, using AI-based beam modeling as discussed in various examples in this article, the dose distribution can be calculated only once to determine whether the beam model meets the delivery criteria. Therefore, overall cost savings for beam modeling and treatment planning can be achieved and radiotherapy can be improved.

[0016] The above content is intended to provide an overview of the subject matter of this patent application. The above content is not intended to provide an exclusive or exhaustive explanation of the present invention. This application includes specific embodiments to provide further information about this patent application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are not necessarily drawn to scale, and in the drawings, like reference numerals describe substantially similar components throughout the several views. Like reference numerals with different letter suffixes represent different instances of substantially similar components. The accompanying drawings generally illustrate various examples discussed in this document by way of example and not by way of limitation.

[0018] Figure 1 An exemplary radiation therapy system is shown.

[0019] Figure 2A An exemplary radiation therapy system that can provide a therapy beam is shown.

[0020] Figure 2B An exemplary combined system including a computed tomography (CT) imaging system and a radiation therapy system is shown.

[0021] Figure 3A partial cross-sectional view of an exemplary combined system including a magnetic resonance (MR) imaging system and a radiation therapy system is shown.

[0022] Figure 4 is a diagram illustrating at least a portion of a Monte Carlo (MC) based TPS system and a virtual source model (VSM) that may be used to model clinical bundles.

[0023] Figure 5 is a block diagram illustrating an exemplary process of training a deep learning (DL) model for predicting bundle model parameters in a bundle model.

[0024] Fig. 6A is a block diagram illustrating an exemplary convolutional neural network (CNN) that may be constructed and trained to predict bundle model parameters.

[0025] Figure 6B It is shown Fig. 6A Block diagram of an exemplary structure of the initial module (Inception Module) in the convolutional layer of the CNN in .

[0026] 7A to 7E is a diagram showing an example of dose characteristics generated based on measurement of scan data during radiation simulation.

[0027] Figure 8 is a flow chart illustrating an exemplary method for AI-based modeling and validation of beam models used in radiation therapy treatment planning.

[0028] Fig. 9 A block diagram is generally shown of an example machine on which any one or more of the techniques (eg, methodologies) discussed herein may be performed. DETAILED DESCRIPTION

[0029] In the following detailed description, reference is made to the accompanying drawings that form a part of the detailed description, and the detailed description is shown by way of a diagrammatic specific example in which the present disclosure can be practiced. These examples, also referred to herein as "examples", are described in sufficient detail to enable those skilled in the art to practice the present disclosure, and it should be understood that the examples may be combined or other examples may be utilized and structural, logical, and electrical changes may be made without departing from the scope of the present disclosure. Therefore, the following detailed description should not be considered to have a limiting meaning, and the scope of the present disclosure is limited by the attached aspects and their equivalents.

[0030] Figure 1An exemplary radiation therapy system 100 for providing radiation therapy to a patient is shown. The radiation therapy system 100 includes, among other components, a data processing device 112. The data processing device 112 may be connected to a network 120. The network 120 may be connected to the Internet 122. The network 120 may connect the data processing device 112 with one or more of a database 124, a hospital database 126, an oncology information system (OIS) 128, a radiation therapy device 130, an image acquisition device 132, a display device 134, and a user interface 136. The data processing device 112 may be configured to generate a radiation therapy treatment plan 142 to be used by the radiation therapy device 130.

[0031] The data processing device 112 may include a memory 116, a processor 114, and a communication interface 118. The memory 116 may store computer-executable instructions, such as a radiation therapy treatment plan 142 (e.g., an original treatment plan, an adjusted treatment plan, etc.), an operating system 143, a software program 144, and any other computer-executable instructions to be executed by the processor 114. The memory 116 may additionally store data, such as medical images 146, patient data 145, and other data required to implement the radiation therapy treatment plan 142.

[0032] The software program 144 may include one or more software packages that, when executed by a machine such as the processor 114, may perform specific image processing and generate a radiation therapy plan 142. In an example, the software program 144 may convert a medical image in one format (e.g., MRI) to another format (e.g., CT) by generating a synthetic image such as a pseudo CT image. For example, the software program 144 may include an image processing program that trains a prediction model for converting a medical image from a medical image 146 of one modality (e.g., an MR image) to a synthetic image of a different modality (e.g., a pseudo CT image); alternatively, the trained prediction model may convert a CT image to an MR image. In another example, the software program 144 may register a patient image (e.g., a CT image or an MR image) with a dose distribution (also represented as an image) of the patient so that corresponding image voxels and dose voxels are appropriately associated through a network. In yet another example, the software program 144 may replace a function of a patient image, such as a signed distance function or a processed version of the image that emphasizes some aspect of the image information. Such a function may emphasize edges or differences in voxel texture, or any other structural aspect useful for neural network learning. The software program 144 may replace a function of the dose distribution that emphasizes some aspect of the dose information. Such a function may emphasize steep gradients around the target or any other structural aspect useful for neural network learning.

[0033] In an example, the software program 144 can generate projection images for a two-dimensional (2D) and / or 3D CT or MR image set depicting an anatomical structure (e.g., one or more targets and one or more OARs), the projection images representing different views of the anatomical structure from the perspective of a treatment gantry of a radiotherapy device. For example, the software program 144 can process a CT or MR image set and create a stack of projection images depicting different views of the anatomical structure depicted in the CT or MR image from various viewing angles of the gantry of the radiotherapy device. In particular, one projection image can represent a view of the anatomical structure at 0 degrees from the gantry, a second projection image can represent a view of the anatomical structure at 45 degrees from the gantry, and a third projection image can represent a view of the anatomical structure at 90 degrees from the gantry. The degree can be the direction of the beam relative to a particular axis of the anatomical structure depicted in the CT or MR image. The axis can remain the same for each beam of different degrees.

[0034] In an example, the software program 144 can generate a graphical aperture image representation of the MLC leaf position at various gantry angles. These graphical aperture images are also referred to as aperture images. In particular, the software program 144 can receive a set of control points that are used to control the radiotherapy device to produce a shaped radiotherapy beam. The control points can represent other machine parameters such as beam intensity, gantry angle relative to the patient position, and leaf position of the MLC. Based on these control points, a graphical image can be generated to graphically represent the beam shape and intensity output by the MLC and jaws at a specific gantry angle. The software program 144 can align the graphical image of the aperture at a specific gantry angle with the corresponding projection image generated at that angle. The image is aligned with the projection and scaled so that the projection image pixels are aligned with the corresponding aperture image pixels.

[0035] The software program 144 may include treatment planning software. The treatment planning software, when executed, for example, by a treatment planning system (TPS), may generate a radiation therapy treatment plan 142. In an example, execution of the treatment planning software may generate a graphical aperture image representation of MLC leaf positions at a given gantry angle for an anatomical projection image representing a view of the anatomical structure from a given gantry angle.

[0036] As depicted, the software program 144 may include a beam model 147. The beam model is represented by various characteristics of the radiation beam that accompanies the introduction of a wide radiation field specific to the treatment machine and exits the radiation therapy machine and is applied to the patient. Using an appropriately determined beam model, machine parameters or control points for a given type of machine may be calculated, and the radiation therapy machine may output a beam from the MLC that achieves an estimated graphical aperture image representation of the same or similar MLC leaf positions and intensities. The treatment planning software, when executed, may output an image representing an estimated image of the beam shape and intensity for a given gantry angle and for a given projection image of the gantry at that angle, and the functionality may calculate control points for a given radiotherapy device to achieve that beam shape and intensity.

[0037] The beam model 147 can be represented by a function of one or more beam model types that characterize various characteristics of one or more radiation modalities, such as photons or electrons. Different beam models can differ in the number and / or configuration of radiation sources. Therefore, beam model parameters (e.g., the size, position, energy spectrum, or fluence distribution of the radiation source) may vary for different beam model types. By way of example and not limitation, beam model parameters may include: the size and position of one or more photon sources within a radiotherapy machine, the maximum energy or average energy of a photon spectrum of photons emitted from the radiotherapy machine, a factor describing the shape of a photon spectrum emitted from the radiotherapy machine, the size and position of one or more electron sources within a radiotherapy machine, the maximum energy or average energy of an electron spectrum emitted from the radiotherapy machine, a factor describing the shape of an electron spectrum, or one or more numbers describing how radiation (e.g., electrons or photons) emitted by a radiotherapy machine may vary off-axis, etc.

[0038] The AI-based beam modeling techniques according to various examples discussed in this document can be applied to different beam models. An example of beam model 147 is a complete Monte Carlo (MC) model. A complete MC model can generate accurate dose estimates by simulating primary photons and scattered photons and contamination electrons from a linac head. Therefore, a complete MC model can handle complex beam arrangements associated with modern radiotherapy techniques, such as stereotactic body radiotherapy (SBRT), intensity modulated radiotherapy (IMRT) or volume modulated arc therapy (VMAT). However, a complete MC model may involve a large number of machine components. Detailed modeling of each of the components may require a large phase space file (PSF) to contain a large amount of particle information, such as the energy, position, direction, charge, generated area or interaction of particles at different scoring planes. The large file size of PSF (e.g., as large as several gigabytes) may cause one or more problems related to file storage, reading, deployment and file transfer (e.g., slow speed of reading PSF from a hard disk and transferring it to a local machine). Additionally, the PSF for a complete MC model may rely on precise information about the various components in the linac head, such as detailed geometry and material specifications, which may be proprietary information of the manufacturer that is not readily available to the bundle modeler.

[0039] Another example of a bundle model 147 is a virtual source model (VSM), which can approximate a complete MC model. Figure 4 , which shows at least a portion of a Monte Carlo (MC) based TPS system 400 that uses a VSM to model a clinical beam. MC-based dosimetry algorithms provide the ability to accurately simulate dose distributions within heterogeneous media and are therefore used in clinical situations in radiotherapy. The depicted portion of the MC-based TPS system 400 includes a treatment head (e.g., a linac head) 410 and a multi-leaf collimator (MLC) and a jaw 420. The MLC and jaws 420 can be adjusted to provide personalized radiotherapy to the patient, and are therefore referred to as patient-specific portions of a radiotherapy machine. The radiation beam emitted from the radiation source and transmitted through the treatment head 410 can be modeled by a beam model such as a VSM. The radiation transmission through the MLC and jaws 420 can be modeled by a transmission filter. A medical image 430 (e.g., a CT image) can be generated by the radiation output of the MLC and jaws. Based on the medical image 430, a dose metric and a dose distribution can be calculated based on the electron density of the tissue, for example, using a dose engine 152 in a processor 114. Various dose calculation algorithms based on Monte Carlo techniques may be used to calculate dose metrics or distributions, including, for example, Voxel Monte Carlo (VMC) or X-ray Voxel Monte Carlo (XVMC).

[0040] VSM is a beam model that assumes that the particles emitted from the linac head originate from one or more virtual particle sources with different geometries (e.g., point, disk, ring). The fluence distribution and energy spectrum for each sub-source can be reconstructed from a well-tuned PSF or measurement set. Specifically, VSM parameterizes the photonic and electronic components in the linac head and describes the beam using several fitting parameters. VSM overcomes the problems associated with full MC simulations of accelerator heads, such as long simulation times, cumbersome debugging routines, and dependence on technical information about the accelerator head. Figure 4 As depicted, the VSM can include two virtual photon sources (representing contributions from the target and the flattening filter) and one virtual electron source. Each virtual source of the VSM can generate a particle distribution that is comparable (with the required accuracy) to the dose distribution (e.g., energy and angular distribution) of the original PSF from the MC model. In particular, the VSM model includes a primary photon source 412, a scattered photon source 414 (also known as a secondary photon source), and an electron contamination source 416. The primary photon source 412 can be located in the bremsstrahlung target plane, which is also known as a reference plane with a depth (z) of zero (z=0). The scattered photon source 414 can be located at the base of the primary collimator, in an example at a depth of approximately z=11.9 cm relative to the reference plane. The electron contamination source 416 can be located at the base of the flattening filter, in an example at a depth of approximately z=15.7 cm relative to the reference plane.

[0041] The virtual resources of a VSM can be described by corresponding analytical functions of one or more parameters. By way of example and not limitation, Figure 4 The fluence distributions of the corresponding virtual sources modeled by the corresponding Gaussian distribution models are shown. For example, the primary photon fluence can be represented by a Gaussian model with an energy-dependent standard deviation σ0, and the scattered photon fluence can be represented by a Gaussian model with an energy-dependent standard deviation σ s The Gaussian model represents the electron flux with energy-dependent standard deviation σ eThe parameters describing the virtual sources may include reference parameters fixed based on complete MC data analysis, and open parameters adapted for each individual accelerator during model commissioning. The photon and electron sources in the VSM model can reproduce a variety of effects that exist in a wide range of clinical beams. Correlations can be calculated using these virtual sources, including, for example, off-axis energy softening for primary and secondary photons, absorption / scattering in flattening filters, energy fluence normalization, energy fluence flatness and particle fluence distribution for primary, secondary and electron fluences, energy spectra of primary and secondary photon sources and electron sources, corrections for energy focusing of source distributions for secondary photons and contamination electrons, enhancement of focal particle fluence for contamination electrons, etc. The three virtual sources (i.e., primary photons, secondary photons and electron contamination) can be independent sources of each other, and the joint probability is 1.

[0042] Now return to Figure 1 , the software program 144 may include a trained deep learning (DL) model 148 configured to predict values ​​of model parameters in the beam model 147. The prediction is based on machine scan data indicating a configuration or operating state of the radiation therapy machine. The trained DL model 148 is trained to establish a relationship between machine scan data collected from the radiation therapy machine and a plurality of beam model parameters of the beam model. Machine scan data corresponding to the radiation therapy machine may be received from the user interface 136 or from the database 124 and applied to the trained DL model 148, which may predict beam model parameters of the beam model. Examples of the trained DL model 148 may include a convolutional neural network (CNN), a recurrent neural network (RNN), a deep belief network (DBN), or a hybrid neural network including two or more neural network models of different types or different model configurations.

[0043] CNN networks can automatically learn the characteristics of data from samples, thereby eliminating the complex feature extraction in conventional machine learning models that usually require a lot of expertise. An additional advantage of CNN is that with the help of weight sharing, the scale of CNN parameters can be greatly reduced. Therefore, the complexity of the training process can be reduced, the convergence speed can be increased, and the model generalization ability can be enhanced.

[0044] RNN includes connections between nodes to form a directed graph along a time series. This directed graph can use an internal state (memory) to store past information, and the network decisions are influenced by what it has learned from the past. In an example, a long short-term memory (LSTM) network can be used. LSTM is a type of RNN architecture characterized by feedback connections. A common LSTM unit can be composed of a cell, an input gate that determines how much new information to add to the cell, a forget gate that determines which information to discard or keep in the cell, and an output gate that determines the value to output.

[0045] In the example where the bundle model 147 is a virtual source model (VSM), the DL model 148 may be trained to predict Figure 4 The values ​​of the VSM parameters of one or more virtual sources are depicted. FIG. 6A to FIG. 6B An example of a trained CNN for predicting bundle model parameters is discussed.

[0046] In addition to the memory 116 storing the software program 144, the software program 144 may additionally or alternatively be stored on a removable computer medium, such as a hard drive, a computer disk, a CD-ROM, a DVD, an HD, a Blu-ray DVD, a USB flash drive, an SD card, a memory stick, or any other suitable medium; and the software program 144 may be executed by the processor 114 when the software program 144 is downloaded to the data processing device 112.

[0047] The processor 114 may be communicatively coupled to the memory 116, and the processor 114 may be configured to execute computer executable instructions stored in the memory 116. The processor 114 may send the medical image 146 to the memory 116 or receive the medical image 146 from the memory 116. For example, the processor 114 may receive the medical image 146 from the image acquisition device 132 via the communication interface 118 and the network 120 to store the medical image 146 in the memory 116. The processor 114 may also send the medical image 146 stored in the memory 116 via the communication interface 118 to the network 120 to store the medical image 146 in the database 124 or the hospital database 126.

[0048] The processor 114 may include a training module 151 configured to train a deep learning (DL) model and to generate a trained DL model 148 to predict beam model parameters such as VSM parameters. In order to train the DL model, a training data set may be constructed using machine scan data indicating the configuration or operating state of a radiotherapy machine (e.g., a linac). The training data may include simulation data with known values ​​of beam model parameters and controlled test conditions, such as dose distribution and dose profile for a simulated environment (e.g., in a phantom). The training module 151 may train the DL model using samples taken from the simulation data and corresponding known values ​​of the beam model parameters. Training involves algorithmically adjusting one or more DL model parameters (e.g., network layer node weights and biases) until the trained DL model meets specified training convergence criteria. Refer to the following Figure 5 Discuss an example of training a DL model to predict the values ​​of bundle model parameters.

[0049] The trained DL model 148 can be stored in the memory 116 or the database 124. In an example, the training module 151 and the trained DL model 148 can be archived in a server and can be accessed by one or more clients (e.g., a TPS system). In some examples, multiple DL models can be trained and archived in the server, corresponding to different multi-leaf collimator (MLC) types (e.g., Agility 160-leaf collimator or MLCi2 80-leaf collimator, both from Elekta AB in Stockholm, Sweden) and / or different energy levels (e.g., 4MV, 6MV, 10MV, and 15MV). The DL models archived in the server can be indexed by MLC type and energy level. The appropriate DL model can be queried and retrieved from the server according to the multi-leaf collimator (MLC) type and / or different energy levels.

[0050] The processor 114 can generate a beam model 147 for a specific radiotherapy machine (with a specific collimator type and / or energy level) using the archived trained DL model. In an example, to generate the beam model 147, the processor 114 can retrieve a trained DL model from a plurality of DL models archived and stored in a server according to the collimator type and / or energy level associated with the radiotherapy machine. The processor 114 can then apply machine scan data (e.g., dose curves and dose statistics) acquired from the radiotherapy machine to the retrieved trained DL model, which can generate output values ​​as beam model parameters of the beam model 147.

[0051] The generated beam model 147 can be stored in the software program 144. In an example, the beam model 147 can be presented to the user, such as being displayed on the display device 134. Other information can be presented to the user (e.g., displayed on the display device 134), such as a report containing beam model parameters, geometric information, dose calculation settings, and fitting results, the report showing both the measured dose distribution and the dose distribution calculated based on the beam model. Fitting results. In an example, the beam model 147 can be fed to the TPS for clinical treatment planning.

[0052] In some examples, the processor 114 may validate the bundle model after it is generated, before deploying the bundle model to the TPS for clinical use. Validation may include importing the bundle model into the TPS to execute treatment planning software (e.g., Elekta AB manufactured by Stockholm, Sweden). A treatment planning system) and calculates a dose distribution in a virtual phantom (e.g., a water phantom). The algorithm for calculating the dose distribution in the virtual phantom may include a Monte Carlo dose algorithm, such as an XVMC algorithm. The calculated dose distribution may be compared with measured beam characteristics from a target radiotherapy machine (e.g., a linac) to determine whether the calculated dose meets a dose verification criterion (also referred to as a delivery criterion), such as one or more dose metrics falling within a tolerance range (±x%) of the dose metric relative to the measured value. Examples of delivery criteria may include: a central ray within a ±2% tolerance, a high dose at a low gradient within a 3% tolerance, a high gradient at 3 mm within a 3% tolerance (e.g., 30% / cm), a low dose at a low gradient within a 3% tolerance, an output factor within a 2% tolerance, or any combination thereof. In some examples, the delivery criteria may include manual inspection and verification. For example, a modeling physicist can grade a beam model (e.g., a VSM model) as one of: "Passed" (indicating that the model is ready for delivery), "Failed" (indicating that the model is unacceptable for delivery), or "Needs Improvement" (indicating that further adjustment of beam model parameters is required), for example, in a user interface. The modeling physicist can further add annotations to indicate various characteristics of the dose that need improvement, such as the spectrum of the PDD, the output factor (OF, at the reference depth), the accumulation area (depth range of attenuation for electron contamination), horns (increase in beam intensity away from the central axis, which is mainly determined by the flattening filter design), and the penumbra of the profile (around the geometric beam edge), etc.

[0053] The processor 114 may include a dose engine 152 configured to calculate dose metrics or dose statistics using a beam model, such as a VSM model having model parameters determined by a trained DL model. Various algorithms may be used to calculate dose. In an example, the dose engine may use a Monte Carlo algorithm or a collapsed cone convolution technique (CCC) algorithm (which may be implemented as a software package stored in the software program 144) to calculate dose metrics or dose statistics. Examples of dose engines may include a voxel Monte Carlo (VMC) dose engine, an X-ray voxel Monte Carlo (XVMC) dose engine, or a GPU Monte Carlo dose (GPUMCD).

[0054] The processor 114 may include at least a portion of a treatment planning system (TPS) configured to execute treatment planning software (as part of the software program 144) and generate a radiation therapy treatment plan 142 using the beam model, the medical image 146, and the patient data 145. The medical image 146 may include information such as imaging data associated with a patient anatomical region, organ, or volume of interest segmentation data. The patient data 145 may include information such as: functional organ modeling data (e.g., serial organs vs. parallel organs, appropriate dose response models, etc.); radiation dose data (e.g., DVH information); or other clinical information about the patient and treatment (e.g., other surgeries, chemotherapy, previous radiation therapy, etc.).

[0055] In some examples, the processor 114 may utilize the software program 144 to generate intermediate data, such as, for example, updated parameters to be used by a machine learning model such as a neural network model; or to generate an intermediate 2D or 3D image, which may then be subsequently stored in the memory 116. The processor 114 may then subsequently transmit the executable radiation therapy treatment plan 142 to the radiation therapy device 130 via the communication interface 118 to the network 120, where the radiation therapy plan may be used to treat the patient with radiation. In addition, the processor 114 may execute the software program 144 to implement functions such as image conversion, image segmentation, deep learning, neural networks, and artificial intelligence. For example, the processor 114 may execute a software program 144 that trains or contours medical images; such a software program 144, when executed, may train a boundary detector or utilize a shape dictionary.

[0056] The processor 114 may be a processing device, including: one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), etc. More specifically, the processor 114 may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor that implements other instruction sets, or a processor that implements a combination of instruction sets. The processor 114 may also be implemented by one or more special-purpose processing devices such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), etc. As will be appreciated by those skilled in the art, in some examples, the processor 114 may be a special-purpose processor rather than a general-purpose processor. The processor 114 may include one or more known processing devices, such as those from Intel. TM Pentium TM 、Core TM , Xeon TM or series of microprocessors from AMD TM Made by Turion TM , Athlon TM 、Sempron TM 、Opteron TM FX TM 、Phenom TM series of microprocessors, or any of a variety of processors manufactured by Sun Microsystems. Processor 114 may also include a graphics processing unit, such as a graphics processing unit from Nvidia TM Manufactured series, by Intel TM GMA, Iris TM series or by AMD TM Made by Radeon TM Processor 114 may also include an accelerated processing unit, such as an Intel TM XeonPhi manufactured TMThe disclosed examples are not limited to any type of processor that is otherwise configured to meet the computational requirements of identifying, analyzing, maintaining, generating and / or providing large amounts of data or manipulating such data to perform the methods disclosed herein. In addition, the term "processor" may include more than one processor (e.g., a multi-core design or multiple processors each having a multi-core design). The processor 114 may execute sequences of computer program instructions stored in the memory 116 to perform various operations, processes, and methods that will be described in more detail below.

[0057] In some examples, various functions performed by the processor 114, such as training the DL model by the training module 151, generating and validating the beam model, and calculating the dose based on the beam model using the dose engine 152, can be distributed among two or more processors, or distributed in a client-server architecture. In an example, the training module 151 can be implemented in a server, and the dose engine 152 can be implemented in a client (e.g., a local TPS system). In an example, the trained DL model 148 can be archived in the server.

[0058] The memory 116 may store medical images 146. In some examples, the medical images 146 may include one or more MR images (e.g., 2D MRI, 3D MRI, 2D flow MRI, four-dimensional (4D) MRI, 4D volume MRI, 4D imaging MRI, etc.), functional MR images (e.g., fMRI, DCE-MRI, diffusion MRI), CT images (e.g., 2D CT, cone beam CT, 3D CT, 4D CT), ultrasound images (e.g., 2D ultrasound, 3D ultrasound, 4D ultrasound), one or more projection images representing views of anatomical structures depicted in MRI, synthetic CT (pseudo-CT) and / or CT images at different angles of the gantry relative to the patient axis, PET images, X-ray images, fluoroscopic images, radiotherapy portal images, SPECT images, computer-generated synthetic images (e.g., pseudo-CT images), aperture images, graphical aperture image representations of MLC leaf positions at different gantry angles, etc. In addition, the medical images 146 may also include medical image data, such as training images and ground truth images, contour images, and dose images. In an example, the medical image 146 may be received from the image acquisition device 132. Thus, the image acquisition device 132 may include an MRI imaging device, a CT imaging device, a PET imaging device, an ultrasound imaging device, a fluoroscopic device, a SPECT imaging device, an integrated linac and MRI imaging device, or other medical imaging device for obtaining medical images of a patient. The medical image 146 may be received and stored in any data type or any format type that the data processing device 112 may use to perform operations consistent with the disclosed examples.

[0059] The memory 116 may be a non-transitory computer-readable medium such as a read-only memory (ROM), a phase-change random access memory (PRAM), a static random access memory (SRAM), a flash memory, a random access memory (RAM), a dynamic random access memory (DRAM) such as a synchronous DRAM (SDRAM), an electrically erasable programmable read-only memory (EEPROM), a static memory (e.g., flash memory, a flash disk, a static random access memory), and other types of random access memory, a cache memory, a register, a CD-ROM, a DVD or other optical storage device, a cassette tape, other magnetic storage devices, or any other non-transitory medium that can be used to store information including images, data, or computer-executable instructions (e.g., stored in any format) that can be accessed by the processor 114 or any other type of computer device. The computer program instructions can be accessed by the processor 114, read from the ROM or any other suitable memory location, and loaded into the RAM for execution by the processor 114. For example, the memory 116 may store one or more software applications. The software applications stored in the memory 116 may include, for example, an operating system 143 for common computer systems and for software-controlled devices. In addition, the memory 116 may store an entire software application or only a portion of a software application that may be executed by the processor 114. For example, the memory 116 may store one or more radiation therapy treatment plans 142.

[0060] The data processing device 112 may communicate with the network 120 via a communication interface 118, which may be communicatively coupled to the processor 114 and the memory 116. The communication interface 118 may provide a communication connection between the data processing device 112 and components of the radiation therapy system 100 (e.g., allowing data to be exchanged with external devices). For example, in some examples, the communication interface 118 may have appropriate interface circuitry to connect to a user interface 136, which may be a hardware keyboard, keypad, or touch screen through which a user may input information into the radiation therapy system 100.

[0061] The communication interface 118 may include, for example, a network adapter, a cable connector, a serial connector, a USB connector, a parallel connector, a high-speed data transmission adapter (e.g., such as fiber optic, USB 3.0, thunderbolt, etc.), a wireless network adapter (e.g., such as a WiFi adapter), a telecommunication adapter (e.g., 3G, 4G / LTE, etc.), etc. The communication interface 118 may include one or more digital and / or analog communication devices that allow the data processing device 112 to communicate with other machines and devices, such as remotely located components, via the network 120.

[0062] The network 120 may provide the functionality of a local area network (LAN), a wireless network, a cloud computing environment (e.g., software as a service, platform as a service, infrastructure as a service, etc.), a client-server, a wide area network (WAN), etc. For example, the network 120 may be a LAN or a WAN, which may include other systems S1 (138), S2 (140), and S3 (141). The systems S1, S2, and S3 may be the same as the data processing device 112, or may be different systems. In some examples, one or more of the systems in the network 120 may form a distributed computing / simulation environment that collaboratively performs the examples described herein. In some examples, one or more of the systems S1, S2, and S3 may include a CT scanner that obtains a CT image (e.g., a medical image 146). In addition, the network 120 may be connected to the Internet 122 to communicate with servers and clients that are remotely residing on the Internet.

[0063] Thus, the network 120 may allow data transfer between the data processing device 112 and many different other systems and devices, such as the OIS 128, the radiation therapy device 130, and the image acquisition device 132. In addition, data generated by the OIS 128 and / or the image acquisition device 132 may be stored in the memory 116, the database 124, and / or the hospital database 126. Data may be transmitted / received via the network 120 through the communication interface 118 for access by the processor 114 as desired.

[0064] The data processing device 112 can communicate with the database 124 through the network 120 to send / receive a variety of different types of data stored on the database 124. For example, the database 124 can store machine data associated with the radiotherapy device 130, the image acquisition device 132, or other machines related to radiotherapy. The machine data information may include control points such as radiation beam size, arc placement, duration of beam on and off, machine parameters, segmentation, MLC configuration, gantry speed, MRI pulse sequence, etc. In an example, the database 124 can store training data, which can be used to train the DL model to predict the model parameter values ​​of the beam model. The training data may include machine scan data indicating the configuration or operating state of the corresponding radiotherapy machine (e.g., linac machine), and the corresponding real, measured dose measurement. The database 124 may be a storage device and may be equipped with an appropriate database management software program. Those skilled in the art will understand that the database 124 may include multiple devices located in a central or distributed manner.

[0065] In some examples, database 124 may include a processor-readable storage medium (not shown). Although the processor-readable storage medium in the example may be a single medium, the term "processor-readable storage medium" should be considered to include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) that store one or more sets of computer-executable instructions or data. The term "processor-readable storage medium" should also be considered to include any medium capable of storing or encoding an instruction set that is executed by a processor and causes the processor to perform any one or more of the methods of the present disclosure. Therefore, the term "processor-readable storage medium" should be considered to include, but is not limited to, solid-state memory, optical media, and magnetic media. For example, a processor-readable storage medium may be one or more volatile, non-transient, or non-volatile tangible computer-readable media.

[0066] The processor 114 may communicate with the database 124 to read images into the memory 116 or store images from the memory 116 to the database 124. For example, the database 124 may be configured to store a plurality of images (e.g., 3D MRI, 4D MRI, 2D MRI slice images, CT images, 2D fluoroscopic images, X-ray images, raw data from MR scans or CT scans, Digital Imaging and Communications in Medicine (DICOM) data, projection images, graphic aperture images, etc.) received by the database 124 from the image acquisition device 132. The database 124 may store data to be used by the processor 114 when executing the software program 144 or when creating the radiation therapy treatment plan 142. The data processing device 112 can receive image data, such as medical images 146 (e.g., 2D MRI slice images, CT images, 2D fluoroscopic images, X-ray images, 3D MR images, 4D MR images, projection images, graphic aperture images, etc.) from a database 124, a radiotherapy device 130 (e.g., MR-linac) and / or an image acquisition device 132 to generate a treatment plan 142.

[0067] In an example, the radiotherapy system 100 may include an image acquisition device 132 that is capable of acquiring medical images of a patient (e.g., MR images, 3D MRI, 2D flow MRI, 4D volume MRI, CT images, cone beam CT, PET images, functional MR images (e.g., fMRI, DCE-MRI, and diffusion MRI), X-ray images, fluoroscopic images, ultrasound images, radiotherapy portal images, SPECT images, etc.). The image acquisition device 132 may, for example, be an MRI imaging device, a CT imaging device, a PET imaging device, an ultrasound device, a fluoroscopic device, a SPECT imaging device, or any other suitable medical imaging device for obtaining one or more medical images of a patient. The images acquired by the image acquisition device 132 may be stored in the database 124 as imaging data and / or test data. By way of example, the images acquired by the image acquisition device 132 may also be stored in the memory 116 by the data processing device 112 as medical images 146.

[0068] In an example, for example, the image acquisition device 132 can be integrated into a single device with the radiation therapy device 130. For example, an MR imaging device can be combined with a linear accelerator to form a system known as an "MR-linac". Such an MR-linac can be used, for example, to determine the location of a target organ or target tumor in a patient's body to accurately direct the radiation therapy to a predetermined target according to the radiation therapy treatment plan 142.

[0069] The image acquisition device 132 can be configured to acquire one or more images of the patient's anatomical structure for a region of interest (e.g., a target organ, a target tumor, or both). Each image, typically a 2D image or slice, can include one or more parameters (e.g., 2D slice thickness, orientation, and position, etc.). In an example, the image acquisition device 132 can acquire 2D slices of any orientation. For example, the orientation of a 2D slice can include a sagittal orientation, a coronal orientation, or an axial orientation. The processor 114 can adjust one or more parameters such as the thickness and / or orientation of a 2D slice to include a target organ and / or a target tumor. In an example, a 2D slice can be determined based on information such as a 3D MRI volume. For example, in the case of using a radiotherapy device 130, such a 2D slice can be acquired by the image acquisition device 132 in "real time" while the patient is receiving radiotherapy treatment, where "real time" means acquiring data at least in milliseconds or less.

[0070] The data processing device 112 can generate and store a radiation therapy treatment plan 142 for one or more patients. The radiation therapy treatment plan 142 can provide information about the specific radiation dose to be applied to each patient. The radiation therapy treatment plan 142 can also include other radiotherapy information such as control points, including beam angles, gantry angles, beam intensities, dose-histogram-volume (DVH) information, the number of radiation beams used during the therapy, the dose of each beam, etc.

[0071] Processor 114 may be configured to perform a treatment planning operation using a software program 144 such as treatment planning software (e.g., manufactured by Elekta AB of Sweden). ) to generate a radiation therapy treatment plan 142. To generate the radiation therapy treatment plan 142, the processor 114 can communicate with an image acquisition device 132 (e.g., a CT device, an MRI device, a PET device, an X-ray device, an ultrasound device, etc.) to access images of the patient and depict a target such as a tumor. In some examples, depiction of one or more OARs, such as healthy tissue surrounding or close to the tumor, may be required. Therefore, when the OAR is close to the target tumor, segmentation of the OAR can be performed. In addition, if the target tumor is close to the OAR (e.g., the prostate is close to the bladder and rectum), by segmenting the OAR relative to the tumor, the radiotherapy system 100 can not only study the dose distribution in the target, but also study the dose distribution in the OAR.

[0072] In order to depict the target organ or target tumor relative to the OAR, medical images such as MR images, CT images, PET images, fMR images, X-ray images, ultrasound images, radiotherapy field images, SPECT images, etc. of the patient undergoing radiotherapy can be non-invasively obtained by the image acquisition device 132 to reveal the internal structure of the body part. Based on the information from the medical image, the 3D structure of the relevant anatomical part can be obtained. In addition, during the treatment planning process, many parameters can be considered to achieve a balance between effective treatment of the target tumor (for example, so that the target tumor receives sufficient radiation dose for effective treatment) and low radiation to the OAR (for example, the OAR receives the lowest possible radiation dose). Other parameters that can be considered include the location of the target organ and target tumor, the location of the OAR, and the movement of the target relative to the OAR. For example, the 3D structure can be obtained by outlining the outline of the target in each 2D layer or slice of the MRI or CT image or outlining the outline of the OAR and combining the outlines of each 2D layer or slice. It can be manually (for example, by a physician, dosimetrist or medical staff using a program such as Elekta AB manufactured by Sweden). ) or automatically (e.g., using a program such as the Atlas-based automatic segmentation software ABAS manufactured by Elekta AB of Sweden TM) Generate a contour. In some examples, the 3D structure of the target tumor or OAR can be automatically generated by the treatment planning software.

[0073] After the target tumor and (one or more) OARs have been located and delineated, the dosimetrist, physician or medical staff can determine the radiation dose to be applied to the target tumor and any maximum dose that can be received by the OARs close to the tumor (e.g., left and right parotid glands, optic nerves, eyes, lenses, inner ears, spinal cord, brainstem, etc.). After the radiation dose is determined for each anatomical structure (e.g., target tumor, OAR), a process known as inverse planning can be performed to determine one or more treatment planning parameters that will achieve the desired radiation dose distribution. Examples of treatment planning parameters include volume delineation parameters (e.g., which define target volumes, contour-sensitive structures, etc.), margins around target tumors and OARs, beam angle selection, collimator settings, and beam on time. During the inverse planning process, the physician can define dose constraint parameters that set limits on how much radiation the OARs can receive (e.g., defining a full dose to the tumor target and a zero dose to any OAR; defining a dose to 95% of the target tumor; defining spinal cord, brainstem, and optic nerve structures to receive ≤45Gy, ≤55Gy, and <54Gy, respectively). The results of the inverse planning may constitute a radiation therapy treatment plan 142 that may be stored in the memory 116 or the database 124. Some of these treatment parameters may be related. For example, adjusting one parameter (e.g., weights for different objectives, such as increasing the dose to the target tumor) in an attempt to change the treatment plan may affect at least one other parameter, which in turn may result in the development of a different treatment plan. Thus, the data processing device 112 may generate a customized radiation therapy treatment plan 142 having these parameters so that the radiation therapy device 130 provides the radiation therapy treatment to the patient.

[0074] In addition, the radiation therapy system 100 may include a display device 134 and a user interface 136. The display device 134 may include one or more display screens that display medical images, interface information, treatment plan parameters (e.g., projection images, graphic aperture images, contours, doses, beam angles, etc.), treatment plans, targets, localized target positions, and / or tracked targets, or any related information to the user. The user interface 136 may be a keyboard, keypad, touch screen, or any type of device that a user can use to input information to the radiation therapy system 100. Alternatively, the display device 134 and the user interface 136 may be integrated into a device such as a tablet computer.

[0075] In addition, any and all components of the radiotherapy system 100 can be implemented as virtual machines (e.g., VMWare, Hyper-V, etc.). For example, a virtual machine can be software that acts as hardware. Therefore, a virtual machine can include at least one or more virtual processors, one or more virtual memories, and one or more virtual communication interfaces that act together as hardware. For example, the data processing device 112, OIS 128, and image acquisition device 132 can be implemented as virtual machines. In view of the processing power, memory, and computing power available, the entire radiotherapy system 100 can be implemented as a virtual machine.

[0076] Figure 2A An exemplary radiotherapy device 202 is shown, which may include a radiation source (e.g., an x-ray source or linac), a bed 216, an imaging detector 214, and a radiotherapy output 204. The radiotherapy device 202 may be configured to emit a radiation beam 208 to provide therapy to a patient. The radiotherapy output 204 may include one or more attenuators or collimators, such as MLCs. The patient may be positioned in the region 212 and supported by the bed 216 to receive a radiotherapy dose according to a radiotherapy treatment plan. The radiotherapy output 204 may be mounted or attached to a gantry 206 or other mechanical support. When the bed 216 is inserted into the treatment area, one or more chassis motors (not shown) may rotate the gantry 206 and the radiotherapy output 204 around the bed 216. In an example, when the bed 216 is inserted into the treatment area, the gantry 206 may rotate continuously around the bed 216. In another example, when the bed 216 is inserted into the treatment area, the gantry 206 may rotate to a predetermined position. For example, the gantry 206 can be configured to rotate the therapy output 204 about an axis ("A"). The bed 216 can be independently moved to other positions around the patient, for example, being movable in a transverse direction ("T"), being movable in a lateral direction ("L"), or being rotatable about one or more other axes, such as about a transverse axis (indicated as "R"). A controller communicatively connected to one or more actuators (not shown) can control the bed 216 to move or rotate so as to appropriately position the patient within or outside of the radiation beam 208 according to the radiation therapy treatment plan. Both the bed 216 and the gantry 206 can be moved independently of each other in multiple degrees of freedom, which allows the patient to be positioned so that the radiation beam 208 can target the tumor. The MLC can be integrated with the gantry 206 to deliver a radiation beam 208 of a specific shape.

[0077] Figure 2AThe coordinate system shown (including axes A, T, and L) can have an origin located at an isocenter 210. The isocenter can be defined as the location where the central axis of the radiation beam 208 intersects the origin of the coordinate axes so as to deliver a prescribed radiation dose to a location on or within a patient. Alternatively, the isocenter 210 can be defined as the location where the central axis of the radiation beam 208 intersects the patient for various rotational positions of the radiation therapy output 204 positioned by the gantry 206 about axis A. As discussed herein, the gantry angle corresponds to the position of the gantry 206 relative to axis A, although any other axis or combination of axes can be referenced and used to determine the gantry angle.

[0078] The linac system can have an imaging detector 214 preferably opposite to the radiation therapy output 204. In an example, the imaging detector 214 can be located within the field of the therapy beam 208. The imaging detector 214 can remain aligned with the therapy beam 208. The imaging detector 214 can rotate around the rotation axis as the gantry 206 rotates. In an example, the imaging detector 214 can be a flat panel detector (e.g., a direct detector or a scintillator detector). In this way, the imaging detector 214 can monitor the therapy beam 208, or generate an image of the patient's anatomical structure. The control circuit system of the radiation therapy device 202 can be integrated into the system 100 or remote from the system 100.

[0079] In the illustrative example, one or more of the couch 216, therapy output 204, or gantry 206 can be automatically positioned, and therapy output 204 can establish therapy beam 208 according to a specified dose for a particular therapy delivery instance. A series of therapy deliveries can be specified according to a radiation therapy treatment plan (e.g., using one or more different orientations or positions of gantry 206, couch 216, or therapy output 204). The therapy deliveries can occur sequentially, but can also be staggered in a desired treatment site on or in a patient (e.g., at isocenter 210). Thus, a prescribed dose of radiation therapy can be delivered to the treatment site while reducing or avoiding damage to tissue near the treatment site.

[0080] Figure 2B An exemplary radiation therapy system 202 is shown that combines a radiation system (e.g., a linac) and a CT imaging system. The radiation therapy output 204 may include an MLC (not shown). The CT imaging system may include an imaging X-ray source 218 that provides X-ray energy, for example, in the kiloelectronvolt (keV) energy range. The imaging X-ray source 218 may provide a fan and / or cone beam 208 directed toward an imaging detector 222, such as a flat panel detector. The radiation therapy device 202 may be similar to that described with respect to Figure 2AThe depicted system, for example, includes a radiation therapy output 204, a gantry 206, a couch 216, and another imaging detector 214 (eg, a flat panel detector). An X-ray source 218 may provide a relatively low energy diagnostic X-ray beam for imaging.

[0081] like Figure 2B As shown, the radiation therapy output 204 and the x-ray source 218 can be mounted on the same rotating rotating mechanism, rotated 90 degrees apart from each other. In some examples, two or more x-ray sources can be mounted along the periphery of the rotating mechanism, so that each x-ray source has its own detector arrangement to provide diagnostic imaging at multiple angles simultaneously. Similarly, multiple radiation therapy outputs 204 can be provided.

[0082] Figure 3 An exemplary radiation therapy system 300 that combines a radiation system (e.g., linac) and a nuclear MR imaging system is shown, which is also referred to as an MR-linac system. The system 300 may include a couch 216, an image acquisition device 320, and a radiation delivery device 330. The system 300 may deliver radiation therapy to a patient according to a radiation therapy treatment plan (e.g., a treatment plan 142 generated and stored in the memory 116). In some examples, the image acquisition device 320 may correspond to Figure 1 An image acquisition device 132 in the image acquisition device can acquire an image of a first modality (for example, an MR image) or a target image of a second modality (for example, a CT image).

[0083] The bed 216 can support the patient during the treatment phase. In some implementations, the bed 216 can move along a horizontal translation axis (labeled "I") so that the bed 216 can move a patient lying on the bed 216 into and / or out of the system 300. The bed 216 can also rotate around a central vertical rotation axis that is transverse to the translation axis. To allow such movement or rotation, the bed 216 can have a motor (not shown) that enables the bed 216 to move. A controller (not shown) can control these movements or rotations so as to appropriately position the patient according to the treatment plan.

[0084] In some examples, the image acquisition device 320 may include an MR imaging machine that can acquire 2D or 3D MR images of the patient before, during, and / or after a treatment session. The image acquisition device 320 may include a magnet 321 for generating a main magnetic field for magnetic resonance imaging. The magnetic field lines generated by the operation of the magnet 321 may extend substantially parallel to the central translation axis "I". The magnet 321 may include one or more coils whose axes extend parallel to the translation axis "I". In some examples, one or more coils in the magnet 321 may be spaced apart so that the central window 323 of the magnet 321 is free of coils. In other examples, the coils in the magnet 321 may be thin enough or have a reduced density so that the coils are substantially transmissive to radiation of a wavelength generated by the radiotherapy device 330. In some examples, the image acquisition device 320 may also include one or more shielding coils that may generate magnetic fields of approximately equal amplitude and opposite polarity outside the magnet 321 to eliminate or reduce any magnetic field outside the magnet 321. As described below, the radiation source 331 of the radiotherapy device 330 can be positioned in a region where the magnetic field is at least cancelled to first order or reduced.

[0085] The image acquisition device 320 may also include two gradient coils 325 and 326, which may generate a gradient magnetic field superimposed on the main magnetic field. The coils 325 and 326 may generate a gradient in the magnetic field generated, which allows protons to be spatially encoded so that the position of the protons can be determined. The gradient coils 325 and 326 may be positioned around a common central axis with the magnet 321 and may be displaced along the central axis. The displacement may create a gap or window between the coils 325 and 326. In an example where the magnet 321 includes a central window 323 between the coils, the two windows may be aligned with each other.

[0086] In some examples, the image acquisition device 320 may be an imaging device other than MRI, for example, X-ray, CT, CBCT, spiral CT, PET, SPECT, optical tomography, fluorescence imaging, ultrasound imaging, radiotherapy portal imaging device, etc. As one of ordinary skill in the art will appreciate, the above description of the image acquisition device 320 relates to certain examples and is not intended to be limiting.

[0087] The radiotherapy device 330 may include a radiation source 331 (e.g., an x-ray source or linac) and a collimator such as an MLC 332. The collimator is a beam limiting device that can help shape the radiation beam coming out of the machine and can limit the maximum field size of the beam. The MLC 332 can be used to shape, direct, or modulate the intensity of the radiotherapy beam to a specified target site in the patient. The MLC 332 may include a metal collimator plate (also referred to as an MLC blade) that slides into position to form a desired field shape. The radiotherapy device 330 may be mounted on a chassis 335. When the bed 216 is inserted into the treatment area, one or more chassis motors (not shown) can rotate the chassis 335 around the bed 216. In an example, when the bed 216 is inserted into the treatment area, the chassis 335 can rotate continuously around the bed 216. The chassis 335 may also have an attached radiation detector (not shown) that is preferably located opposite the radiation source 331 and with the rotation axis of the chassis 335 positioned between the radiation source 331 and the detector. In addition, the device 330 may include control circuitry (not shown) for controlling, for example, one or more of the bed 216, the image acquisition device 320, and the radiation therapy device 330. The control circuitry of the radiation therapy device 330 may be integrated within the system 300 or remote from the system 300.

[0088] During a radiation therapy treatment session, a patient may be positioned on the couch 216. The system 300 may then move the couch 216 into a treatment area defined by the magnet 321 and coils 325, 326, and chassis 335. The control circuitry may then control the radiation source 331, MLC 332, and chassis motors to deliver radiation to the patient through the window between coils 325 and 326 according to the radiation therapy treatment plan.

[0089] FIG. 2A to FIG. 2B and Figure 3 The radiation therapy output configurations shown in the drawings - e.g., a configuration in which the radiation therapy output can rotate about a central axis (e.g., axis "A") - are for purposes of illustration and not limitation. Other radiation therapy output configurations may be used. For example, the radiation therapy output may be mounted to a robotic arm or manipulator having multiple degrees of freedom. In yet another example, the therapy output may be fixed (e.g., located in an area laterally separated from the patient), and a platform supporting the patient may be used to align the radiation therapy isocenter with a designated target site within the patient.

[0090] Figure 55 is a diagram illustrating an exemplary process 500 for training a deep learning (DL) model that can predict bundle model parameter values ​​for a bundle model such as bundle model 147. Process 500 can be implemented as computer-readable and executable instructions and performed by training module 151. Input 504 can include a DL model 510 with an initial network architecture and initial parameter settings. Examples of DL model 510 can include a convolutional neural network (CNN), a recurrent neural network (RNN), a deep belief network (DBN), or a hybrid neural network including two or more neural network models of different types or different model configurations. For example, refer to FIG. 6A to FIG. 6B Describes an example of a CNN model used to predict model parameter values.

[0091] The input 504 may include training data 520 that may be used to train the DL model 510. As depicted, the training data 520 may include simulation data 522 and expected results 524. The simulation data 522 is an input to the DL model, while the expected results 524 are the corresponding desired outputs (or "target labels") of the DL model. The simulation data 522 may include simulated machine scan data representing characteristics of a beam from a radiotherapy machine such as a linac head. Examples of machine scan data may include: a percent depth dose (PDD) representing a relative dose amount, which is determined as a ratio between an axial dose at a particular depth (z, e.g., in the range of 0 mm to 300 mm) and an axial dose at a reference dose depth (z0); a dose profile (or dose distribution) representing an off-axis dose distribution such as a dose on a diagonal, a percent radial dose (PRD) distribution representing a relative dose as a function of radial distance; a dose volume histogram; an overlapped volume histogram; or a three-dimensional dose distribution, etc.

[0092] The machine simulation data 522 may be generated using a dose engine in the software program 144 during a radiation simulation in which a radiation therapy machine is configured with beam model parameters {X} having known values ​​X=X*. During the simulation, a radiation beam from the radiation therapy machine may be applied to a phantom (e.g., a virtual water phantom). The scan data (e.g., PPD and dose distribution) may be measured at different depths in the phantom, such as multiple depths between 0 centimeters (cm) and 30 centimeters (cm). At a particular depth (e.g., z=5 cm), the scan data may be measured at one or more region sizes (e.g., 2×2 cm 2 , 3×3cm 2 , 5×5cm 2 , 10×10cm 2 , 20×20cm 2or 40×40cm 2 ) In some examples, machine simulation data may be generated for each of the beam energy levels, such as 4MV, 6MV, 10MV, and 15MV.

[0093] In some examples, the training data 520 may additionally include machine setup data, which includes information about the settings of the radiation therapy machine. Examples of machine setup data may include: linac settings and geometry information, such as the position and thickness of the jaw collimator and MLC; the thickness, position, shape (perimeter and edge), and composition of each applicator; the thickness, shape, and composition of each insert; the position of the gantry; the position of the gantry, etc. The machine setup data may also include: general information about the radiation therapy machine, such as the linac model, serial number, MLC model and modality, etc.

[0094] The expected results 524 may include a beam model parameter {X} having a known value X=X*. Examples of beam model parameters may include the size and position of one or more photon sources within a radiotherapy machine, the maximum energy of a photon spectrum of photons emitted from the radiotherapy machine, a number of factors describing the shape of a photon spectrum emitted from the radiotherapy machine, the size and position of one or more electron sources within the radiotherapy machine, the average energy of an electron spectrum emitted from the radiotherapy machine, or one or more numbers describing how radiation (e.g., electrons or photons) emitted by a radiotherapy machine may vary off-axis, etc.

[0095] Since the beam model parameters {X} are set to corresponding known values, simulation data 522 can be collected during dose simulation under different test conditions. In an example, the radiotherapy machine can deliver radiation without an applicator during simulation and with the collimator jaws open. The machine scan data collected under this test condition is called an open field measurement. Examples of open field measurements may include an absolute dose (reference dose) at a specific point on a depth-dose curve, a dose distribution in air at a specific source-to-detector distance (SDD), and a depth-dose curve at a specific SSD in water. In an example, the SSD is approximately 100 cm. In another example, an applicator can be used during dose simulation. The machine scan data collected under such test conditions is called an applicator measurement. In an example, an applicator measurement can be performed for each energy / applicator combination, and the applicator measurement includes an absolute dose (reference dose) at a specific point on a depth-dose curve, a depth-dose curve at a specific SSD in water, and the like. Machine scan data (eg, PDD and dose distribution) can be collected for each beam energy in both open field and applicator combinations, where measurements are made at multiple depths in air and / or water.

[0096] Reference 7A to 7E ,exist 7A to 7E The graphs in show beam characteristics such as percentage depth dose (PDD) curves or dose distribution, for example, in a phantom (eg, a water phantom) based on scan data measured during a simulation. Fig. 7A An exemplary PDD curve 710 is shown. The amount of radiation deposited as a function of depth in a medium can be measured using a sensor mounted in a phantom. The PDD curve represents the radiation deposited as a function of depth in a medium ... 2 ) is the dose measurement of the point dose at different depths under the condition of Figure 7B An exemplary dose distribution 720 is shown. The dose distribution is an off-axis dose distribution represented by relative dose measurements at locations having different distances from the central axis of the beam and normalized to dose values ​​at a reference depth and a reference field size. For each dose distribution, a dose value may be measured at a specific depth (z) within a specific field size. By way of example, the dose distribution 720 represents a depth of z=10 cm and 10×10 cm on the x-axis of the phantom. 2 The dose distribution corresponding to the field size.

[0097] Dose characteristics (e.g., PDD curves or dose distributions) can be generated using dose measurements or simulations corresponding to multiple different field sizes. The simulated dose data can be used to construct a 2D dose value matrix containing dose characteristics (e.g., PDD curves or dose distributions) that can be used to train the DL model. PDD curves or dose distributions with different types or field sizes can be sampled at their respective sampling rates to form a 2D matrix. In an example, a 2×2 cm field can be sampled at a resolution of 0.2 mm. 2 In the example, the dose distribution corresponding to the field size of 3×3 cm can be sampled with a resolution of 0.3 mm. 2 By way of example and not limitation, Figure 7C An overlay 730 of a plurality of one-dimensional (1D) relative dose arrays sampled according to respective PDD curves, referred to as sampled PDDs 731 to 736 , is shown. Fig.7D An overlay 740 of multiple 1D relative dose arrays sampled from a dose distribution at a particular depth (e.g., 5 cm on the x-axis of the phantom), referred to as sampled dose distributions 741 to 746, is shown. Fig.7D The portion of the dose distribution having only positive positions (ie, distances greater than zero) is shown in FIG. 731 and the dose distribution 741 corresponding to a 2×2 cm 2 The sampled PDD 732 and sampled dose distribution 742 correspond to a field size of 3×3 cm 2The sampled PDD 733 and sampled dose distribution 743 correspond to a field size of 5×5 cm 2 The sampled PDD 734 and sampled dose distribution 744 correspond to a field size of 10×10 cm 2 The sampled PDD 735 and sampled dose distribution 745 correspond to a field size of 20×20 cm 2 The sampled PDD 736 and sampled dose distribution 746 correspond to a field size of 40×40 cm 2 The field size. Figure 7C and Fig.7D In both, the y-axis represents the relative dose value, and the x-axis represents the index of the relative dose array. The relative dose value at the depth represented by the digital index is sampled using interpolation or extrapolation. The relative dose values ​​from multiple sampled PDDs 731 to 736 or from multiple sampled dose distributions 741 to 746 can be constructed into a data array, such as a two-dimensional (2D) dose value matrix. Sampled PDDs and sampled dose distributions with the same field size can be combined into one curve. The 2D dose value matrix has a first dimension representing different sampling positions and a second dimension representing different field sizes. The resulting 2D dose value matrix can be used as training data to train the DL model 510.

[0098] Training a DL model may require a large amount of training data. Although more simulations may generate more training data, they also extend the simulation time and increase the computational complexity. In addition, this document describes a method for generating additional training data without performing too many simulations. The method is based on interpolating and / or extrapolating the simulation data obtained from the simulation and organized in a 2D dose value matrix. Fig. 7EA surface map 750 representing relative dose values ​​within a field size range and within a position range is shown. The z-axis of the surface map 750 represents the relative dose values, the y-axis represents the sampling index of the field size, and the x-axis represents the sampling index of the position. The relative dose values ​​can be represented by a 2D dose matrix. As depicted, the surface map 750 includes a first portion 751 and a second portion 752. The dose values ​​on the first portion 751 can be represented by a first 2D dose matrix, which is created by interpolating and / or extrapolating the relative dose values ​​between at least two PDD curves. The dose values ​​on the second portion 752 can be represented by a second 2D dose matrix, which is created by interpolating and / or extrapolating the relative dose values ​​between at least two dose distributions. Examples of interpolation can include linear interpolation, nearest neighbor interpolation, polynomial interpolation, spline interpolation, etc. The interpolation and / or extrapolation of the relative dose values ​​can be performed on the dose values ​​with a finer depth (z) resolution and / or a finer field size resolution. In some examples, the interpolated first 2D dose matrix and the second 2D dose matrix can each have a non-uniform depth resolution. For example, for depths between 0 mm and 150 mm, the PDD (first dose matrix) can be linearly interpolated with a resolution of 1 mm; and for depths from 151 mm to 300 mm, the PDD (first 2D dose matrix) can be linearly interpolated with a resolution of 3 mm. In an example, the dose distribution (second 2D dose matrix) can be interpolated with a resolution according to its field size. For example, for a 2×2 cm 2 The field size of the field may have a resolution of about 0.2 mm. In this way, the interpolated / extrapolated first 2D dose matrix and the second 2D dose matrix each represent dose values ​​calculated for a position range and a field size range with a corresponding finer resolution. The interpolated / extrapolated first 2D dose matrix and the second 2D dose matrix can be connected. A specific number (e.g., 3000 to 5000) of connected 2D dose matrices can be used as training data to train the DL model. In an example, the training module 151 can sample the dose characteristics to the connected 2D dose matrix at a specified sampling resolution. The connected 2D dose matrix may include interpolated dose values. In an example, the sampling resolution can be selected according to the field size.

[0099] In an example, the bundle model includes a VSM, such as that described above with reference to Figure 4 During dose simulation, the VSM is programmed with known values ​​for several VSM parameters, X VSM =X VSM*. Simulated machine scan data (e.g., PPD and dose distribution) can be generated while varying only a subset of VSM parameters and keeping other VSM parameters at their respective default values ​​for the same type of linac with the same MLC type (i.e., without changing the entire dose simulation). Examples of the subset of VSM parameters that are varied to generate simulated data may include: Output Factor (OF) parameters including Primary Photons, Primary Sigma, Scattering Distance, Scattering Sigma; Spectral parameters including Energy Maximum, Energy Minimum, B Value, Secondary B Value; and Cumulative Area parameters including Charged Particles, Charged Radius, Charged Average Energy E 平均值 、Maximum charged energy E 最大值 Etc. In an example, at least some of the subsets of VSM parameters may be randomly combined, and simulation data 522 corresponding to the randomly combined VSM parameters may be generated. In an example, probability distributions of the VSM parameters may be analyzed, and simulation data 522 corresponding to the VSM parameter sets may be generated based on their respective probability distributions.

[0100] Now return to reference Figure 5 , the DL training process 508 can apply the training data 520 to train the DL model 510. In an example, the DL training process 508 can be implemented in a first device such as a server, and the training data 520 can reside in a different second device such as a local client (e.g., a TPS system). The training data 520 can be compressed and then transmitted to the server via a communication channel.

[0101] Due to the volume effect of the detector (the finite size of the detector provides an average response to the sensitive volume), the penumbra of the dose distribution obtained from the simulation may become larger than the actual beam characteristics. Simulation data obtained based on the penumbra of the measured dose distribution may introduce prediction errors in one or more beam model parameters such as main sigma or scatter distance and other VSM parameters when used to train the DL model. In order to avoid or reduce such prediction errors, additive noise can be applied to the simulation data 522 to more realistically represent the scan data in the clinical environment, so that the deep learning model can be improved by using such "noisy" training data. For example, the difference between the calculated dose distribution and the measured dose distribution (called dose noise) can be added to at least a portion of the simulation data 522 to produce "noisy" training data that represents more realistic actual clinical data. Training the DL model using noisy training data can improve the robustness of the trained DL model.

[0102] The DL training process 508 may include a training data preprocessing 531 operation to preprocess the training data in order to mitigate certain defects of the training data. As described above, noise may be artificially added to the simulated data before the simulated data is used in the DL training process 508. Preprocessing 531 may include a denoising operation (e.g., a computational model) that may remove or attenuate noise from the simulated data. In some examples, the denoising operation may be integrated into the DL model. For example, denoising may be implemented as a fully connected layer in a CNN. The denoising layer in the CNN may be trained separately from the rest of the DL model. The training data for the denoising model may include measured dose distributions and XVMC calculation data for the feed model. In the example of training a denoising model, the measured dose distribution may be fed to a denoising module to generate an estimated XVMC calculated dose distribution. The estimated distribution may be compared with the calculated dose, and the sum of the differences at each dose point is considered as a penalty for optimizing the denoising module. Alternatively, the denoising layer may be trained together with the rest of the DL model. The process of denoising the training data may improve the model parameter prediction performance of the trained DL model. In some examples, the training data preprocessing 531 operation may include converting the training data into a desired data format, such as rearranging the sequence of PDD data and / or dose distribution data in a 2D dose matrix without changing the values ​​of data points used for feature recognition.

[0103] The pre-processed simulation data may be fed into the DL model 510 to generate estimated results 532, such as values ​​of beam model parameters (e.g., VSM parameters) to be predicted. The estimated results 532 may be compared with expected results 524, such as those beam model parameters {X} (e.g., VSM model parameters {X}) that have been used in the dose simulation to generate the simulation data 522. VSM}) of known value X * . At 533, an estimation error may be calculated, e.g., the difference between the estimation result 532 and the expected result 524. At 534, the estimation error may be compared to a model convergence or training stopping criterion (e.g., a minimum value that persists for a specified number of training iterations). If the convergence or training stopping criterion is not met, the estimation error may be used to update the DL model parameters (θ, e.g., layer node weights and biases), e.g., by backpropagation, to reduce or minimize the estimation error or error in the machine parameters during subsequent training trials. Another batch of training data may then be selected from the training data 520 and the expected results for another iteration of DL model training. In an example, model parameter updates using the estimation error may be performed to minimize or reduce a loss function (or objective function, or cost function). As given in equation (1), an example of a loss function is the square of the estimation error:

[0104] J(θ* )=arg min Θ ||YY * || 2 (1) where Y may represent an estimated value of a bundle model parameter X (estimated result 532), Y* may represent a known value of a bundle model parameter X* (expected result 524), and where Θ* may represent ideal parameters of the DL model that minimize the squared error between Y and Y* (e.g., layer node weights and biases as described above). Other loss functions may be used, such as logcosh, mean absolute error (MAE), mean squared error (MSE), weighted MSE, Huber loss (smoothed MAE), or quantile loss, etc. The values ​​of the DL model parameters (Θ) may be iteratively improved.

[0105] After updating the parameters of the DL model, the iteration index may be incremented by one. The iteration index may correspond to the number of times the parameters of the DL model have been updated. Convergence or training stop criteria may be checked at 534. In an example, the convergence or stop criteria may include comparing the iteration index value with a threshold number of iterations. In an example, the convergence or stop criteria may include comparing an estimated error, such as a cumulative loss (e.g., the difference between an estimated value of the beam model parameter X calculated by the loss function as discussed above and a known value of the beam model parameter X*) in multiple training trials with an error threshold. The error threshold may correspond to an asymptotic minimum of all errors determined. In an example, the error threshold may be represented by an accuracy level, such as a tolerance range, relative to a known value of the beam model parameter X*. The accuracy level may be based on how the parameter affects the calculated dose distribution. By way of example and not limitation, Table 1 shows the accuracy levels of various VSM parameters.

[0106] If it is determined at 534 that a convergence or stopping criterion has been met (e.g., the iteration index exceeds a threshold number of iterations, or the cumulative loss falls below an error threshold), the training process may be stopped. The trained DL model 540 may be saved in the memory 116 of the data processing device 112 or in a server. Additionally or alternatively, a report containing information about the trained DL model 540 may be output to a user via the user interface 136.

[0107] For test data 542 collected from the target radiotherapy system, the trained DL model 540 can be deployed to the DL reasoning process 512 to predict beam model parameters (e.g., VSM parameters). The test data 542 may include machine scan data indicating the configuration or operating state of the radiotherapy machine. By applying the test data 542 to the trained DL model 540, a prediction result 544 (e.g., the value of the beam model parameter) can be determined. The TPS system can use the beam model with the predicted beam model parameter values ​​to create a treatment plan.

[0108] In some examples, multiple DL models can be trained for different MLC types (e.g., Agility collimators or MLCi2 collimators, both from Elekta AB, Stockholm, Sweden) and / or at different energy levels (e.g., 4MV, 6MV, 10MV, and 15MV). The trained DL models can be indexed by MLC type and energy level and archived in the server. During testing, a client (e.g., a TPS system) can request access to the trained DL models in the server, query and retrieve the appropriate DL model based on the MLC type and / or energy level. The test data 542 can then be applied to the retrieved DL model to determine the value of the beam model. The beam model file can be updated by replacing the default values ​​of the beam model parameters with the parameter values ​​determined by the DL model. The beam model file can then be transmitted from the server to the client, where a therapy treatment plan can be generated based on the beam model.

[0109] In some examples, the trained DL model 540 can be validated before being tested in the DL reasoning process 512. The validation data can be generated using a method similar to that used to generate the training data 520. In an example, the validation data can include samples randomly taken from a 2D dose matrix corresponding to the dose surface map 750. The samples included in the validation data may be different from the samples used in the simulation data 522 for the DL training process 508. Applying the validation data to the trained model 540 can produce predicted values ​​for the beam model parameters X. If the difference between the predicted beam model parameter value X and the known beam model parameter value X* falls within a specified accuracy level, the DL model is considered to have passed the validation check. Examples of accuracy levels for various parameters of the VSM beam model for training the DL model to make predictions are shown in Table 1.

[0110] In some examples, the validation data used to validate the DL model 540 may include measured clinical dose data (as opposed to simulated data obtained from the simulations discussed above). Clinical dose data such as PPD and dose distribution data collected from multiple sites (e.g., 5 to 10 sites) with corresponding field sizes can be applied to the DL model 540. The batch validation process can predict beam model parameters, and the dose engine can calculate dose metrics using the beam model with the predicted beam model parameters. If the calculated dose metrics are substantially equivalent to the clinical dose metrics, the DL model is considered to have passed the validation check. The DL model can then be published to the beam modeling process, where the beam model can be generated and provided to the TPS to calculate dose metrics and generate a radiotherapy treatment plan.

[0111]

[0112]

[0113] Table 1: Accuracy levels of VSM parameters

[0114] Fig. 6A and Figure 6B A convolutional neural network (CNN) for predicting VSM parameters is shown. CNN networks can automatically learn the characteristics of data from samples, thereby eliminating the complex feature extraction in conventional machine learning models that usually require a lot of expertise. CNN has been used to analyze images, especially for identifying hierarchical patterns in data. An additional advantage of CNN is that with the help of weight sharing, the size of CNN parameters can be greatly reduced. Therefore, the complexity of the training process can be reduced, the convergence speed can be increased, and the model generalization ability can be enhanced.

[0115] Fig. 6A 610. LeNet CNN 610 includes an input layer 611, an intermediate learning layer 612, and an output layer 613. In the example shown, the intermediate learning layer 612 includes two convolutional layers and two fully connected layers. Each convolutional layer is followed by a maximum pooling layer. Maximum pooling is a form of nonlinear downsampling. Maximum pooling divides the input into a set of non-overlapping sub-regions and outputs the maximum value for each such sub-region. Maximum pooling can help reduce the dimensionality of the intermediate representation, reduce the overall computational effort, and improve the robustness of the CNN.

[0116] In order to detect changes in beam characteristics with several field sizes, the standard convolutional layers can be replaced by inception modules. Inception modules are incorporated into CNNs as a way to reduce computational overhead. Figure 6B It is shown Fig. 6A Figure 6 is a diagram of an exemplary architecture of an initial module 620 (also referred to as an initial V1 module) in a convolutional layer of a LeNet CNN 610 in FIG. The initial module 620 may include multiple parallel convolutional filters with corresponding kernel sizes on the same layer. The kernel size may be modified based on a single CNN. As depicted, the initial module 620 performs convolutions (e.g., in convolutional layer 1) using three different sized filters (1×1, 3×3, 5×5). 1×1 convolutions, 3×3 convolutions, and 5×5 convolutions may be performed on the same level, as shown in FIG. Figure 6BAs shown. As a result, the network becomes progressively wider. Max pooling (e.g., 3×3 max pooling) can be performed in the initial module. The resulting output can be concatenated and sent to the next layer. To make this process computationally cheaper, an additional 1×1 convolution can be added before each of the 3×3 convolution layer and the 5×5 convolution layer. Therefore, the number of input channels is limited, and 1×1 convolution is cheaper than 3×3 convolution and 5×5 convolution. A 1×1 convolution is added after the 3×3 max pooling layer. In some examples, variations of the Initial V1 module such as Initial V2, Initial V3, Initial V4, or Initial-ResNet can be used to replace the convolution layers in LeNet 610.

[0117] Figure 8 8 is a flow chart illustrating an exemplary method 800 of AI-based modeling and validation of a beam model used in radiotherapy treatment planning. The method 800 can use a trained deep learning (DL) model to predict parameter values ​​of a beam model, such as a virtual source model (VSM). In an example, the method 800 can be implemented in and performed by a radiotherapy system 100, which can include a radiotherapy machine for providing radiotherapy to a subject according to a treatment plan.

[0118] At 810, machine scan data may be received or otherwise measured from a radiotherapy machine. The machine scan data may include information about settings of the radiotherapy machine, dose distribution under different test conditions, and dose profiles. In an example, the machine scan data represents beam characteristics of a linac head and may be measured by one or more radiation detectors (e.g., an ion chamber) in a water phantom. Examples of scan data may include: a percentage depth dose (PDD) characterizing a relative dose amount, which is determined as a ratio between an axial dose at a particular depth (z, e.g., in the range of 0 mm to 300 mm) and an axial dose at a reference dose depth (z0); and a dose profile characterizing an off-axis dose distribution, such as a dose on a diagonal, or a percentage radial dose (PRD) distribution representing a relative dose as a function of radial distance.

[0119] At 820, the received machine scan data may be applied to a trained deep learning (DL) model to determine values ​​for beam model parameters for a beam model, including, for example, the size and location of one or more photon sources within a radiotherapy machine, the maximum energy of a photon spectrum of photons emitted from the radiotherapy machine, a number of factors describing the shape of a photon spectrum emitted from the radiotherapy machine, the size and location of one or more electron sources within the radiotherapy machine, the average energy of an electron spectrum emitted from the radiotherapy machine, or one or more numbers describing how radiation (e.g., electrons or photons) emitted by the radiotherapy machine may vary off-axis, etc. The DL model may be pre-trained, for example, using training module 151. As discussed above, the training of the DL model may be performed using exemplary process 500. Examples of DL models may include convolutional neural networks (CNNs), recurrent neural networks (RNNs), long and short term memory (LSTM) networks, deep belief networks (DBNs), generative adversarial networks, machine learning models, transfer learning modules, hybrid neural networks including two or more neural network models of different types or different model configurations, or any type of artificial intelligence based model. The architecture and parameters of the DL model can vary depending on the bundle model to be predicted. In the example, the bundle model is Figure 4 The VSM model described in the above reference FIG. 6A to FIG. 6B An exemplary CNN model for predicting model parameter values ​​of a VSM is described. In some examples, the trained DL model can include a trained denoising module (e.g., a computational model) to preprocess received machine scan data. The preprocessing module can be implemented as a fully connected layer in the denoising module of the CNN and trained to remove or attenuate noise from the received machine scan data. The denoising module can be trained separately from the rest of the DL model. Including a preprocessing module in the DL model can improve the robustness of beam model parameter predictions. The preprocessed data can be converted into a desired data format before being used for prediction of beam model parameter values. For example, as described above with reference to Fig. 7E The described sequence of PDD data and / or dose distribution data in the 2D dose matrix may be rearranged without changing the values ​​of the data points used for feature identification.

[0120] At 830, dose metrics in a virtual phantom (e.g., a water phantom) may be calculated using, for example, a dose engine using a beam model having predicted model parameter values ​​obtained from the trained DL model at 820. Examples of a dose engine may include a voxel Monte Carlo (VMC) dose engine or an X-ray voxel Monte Carlo (XVMC) dose engine. Dose may be calculated using various algorithms including, for example, a Monte Carlo dose algorithm such as an XVMC algorithm. In an example, the beam model may be imported into a TPS (e.g., manufactured by Elekta AB of Stockholm, Sweden). In a treatment planning system), it can use a Monte Carlo algorithm (implemented as a software package stored in software program 144) to calculate the dose distribution in the virtual phantom.

[0121] At 840, the beam model with model parameters determined based on the DL model can be validated before being deployed for clinical treatment planning. The calculated dose from step 830 can be compared with the measured values ​​to determine whether the calculated dose meets the dose verification criteria (also called delivery criteria), such as one or more dose metrics falling within a tolerance range (±x%) of the dose metrics relative to the measured values. Examples of delivery criteria can include any of the following: central ray within ±2% tolerance, high dose at low gradient within 3% tolerance, high gradient (e.g., 30% / cm) at 3mm within 3% tolerance, low dose at low gradient within 3% tolerance, output factor within 2% tolerance. In some examples, the delivery criteria can include manual inspection and verification. For example, a modeling physicist can grade a beam model (e.g., a VSM model) as one of the following: "passed" (indicating that the model is ready for delivery), "failed" (indicating that the model is unacceptable for delivery), or "needs improvement" (indicating that further adjustment of beam model parameters is required), for example in a user interface. Modeling physicists can further add annotations to indicate various characteristics of the dose that need to be improved, such as the spectrum of the PDD, the output factor (OF, at the reference depth), the accumulation area (the depth range of attenuation for electron contamination), the horn (increase in beam intensity away from the central axis, which is mainly determined by the flattening filter design), and the penumbra of the profile (around the geometric beam edge), etc.

[0122] If the beam model meets the delivery criteria at 850, the beam model may be output to a user or deployed to the TPS for clinical treatment planning at 870. The TPS may generate a radiation therapy treatment plan using the beam model, medical images (e.g., imaging data associated with patient anatomical regions, organs, or volume of interest segmentation data), and patient data (e.g., functional organ modeling data, radiation dose data (e.g., DVH information), and other clinical information about the patient and the treatment.

[0123] If at 850, the bundle model does not meet the feed criteria, the bundle model may be modified at 862, such as by manually adjusting one or more bundle model parameters by a human modeler (e.g., a modeling physicist) until it is ready to be deployed to the TPS. Alternatively, in the case of a failed bundle model, the DL model may be retrained or adjusted at 864 and then used to predict a new set of bundle model parameters. The bundle model with the new set of model parameters may be validated at 840.

[0124] Fig. 9 A block diagram of an example of a machine 900 is shown, on which one or more of the methods discussed herein can be implemented. In one or more examples, one or more items of a data processing device 112 can be implemented by the machine 900. In an alternative example, the machine 900 operates as an independent device, or can be connected (e.g., networked) to other machines. In one or more examples, the data processing device 112 can include one or more items of the items of the machine 900. In a networked deployment, the machine 900 can operate with the capacity of a server or client machine in a server-client network environment, or operate as a peer machine in a peer (or distributed) network environment. The machine can be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a network device, a network router, a switch or a bridge, or any machine capable of (sequentially or otherwise) executing instructions, which specify the actions to be taken by the machine. Further, while a single machine is illustrated, the term "machine" shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein.

[0125] The example machine 900 includes a processor 902 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit, a circuit system (e.g., one or more transistors, resistors, capacitors, inductors, diodes, logic gates, multiplexers, buffers, modulators, demodulators, radios (e.g., transmitting or receiving radios or transceivers)), a sensor 921 (e.g., a transducer that converts one form of energy (e.g., light, heat, electricity, mechanical or other energy) to another form of energy), etc. or a combination thereof), a main memory 904 and a static memory 906, which communicate with each other via a bus 908. The machine 900 (e.g., a computer system) may also include a video display unit 910 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)). The machine 900 also includes an alphanumeric input device 912 (eg, a keyboard), a user interface (UI) navigation device 914 (eg, a mouse), a disk drive or mass storage unit 916 , a signal generating device 918 (eg, a speaker), and a network interface device 920 .

[0126] The disk drive unit 916 includes a machine-readable medium 922 on which is stored one or more sets of instructions and data structures (e.g., software) 924 that embody or are utilized by any one or more of the methods or functions described herein. During execution of the instructions 924 by the machine 900, the instructions 924 may also reside, in whole or in part, within the main memory 904 and / or within the processor 902, the main memory 904 and the processor 902 also constituting machine-readable media.

[0127] The machine 900 as shown includes an output controller 928. The output controller 928 manages the flow of data to and from the machine 900. The output controller 928 is sometimes referred to as a device controller, where software that interacts directly with the output controller 928 is referred to as a device driver.

[0128] Although the machine-readable medium 922 is shown as a single medium in the example, the term "machine-readable medium" may include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) storing one or more instructions or data structures. The term "machine-readable medium" should also be considered to include any tangible medium capable of storing, encoding or carrying instructions or capable of storing, encoding or carrying data structures that are executed by a machine and cause the machine to perform any one or more of the methods of the present invention, which data structures are utilized by or associated with such instructions. Therefore, the term "machine-readable medium" should be considered to include, but are not limited to, solid-state memory and optical and magnetic media. Specific examples of machine-readable media include non-volatile memory, which includes, by way of example: semiconductor memory devices, such as erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and flash memory devices; magnetic disks, such as internal hard disks and removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks.

[0129] The instructions 924 may also be sent or received over a communication network 926 using a transmission medium. The instructions 924 may be sent using a network interface device 920 and any of a number of well-known transmission protocols (e.g., HTTP). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), the Internet, a mobile telephone network, a plain old telephone (POTS) network, and wireless data networks (e.g., WiFi and WiMax networks). The term "transmission medium" shall be taken to include any intangible medium that is capable of storing, encoding, or carrying instructions for execution by a machine and includes digital or analog communications signals, or other intangible media that facilitates communication of such software.

[0130] As used herein, "communicatively coupled between" means that entities on either of the couplings must communicate through an item between them, and the entities cannot communicate with each other without communicating through the item.

[0131] Further description of various non-limiting exemplary embodiments is provided below. The exemplary embodiments described below can be practiced in conjunction with one or more other aspects or exemplary embodiments. That is, exemplary embodiments of the present invention, such as those described below, can be implemented, practiced or utilized in any combination (e.g., any combination that is suitable, feasible and / or feasible), and are not limited to those combinations described herein and / or included in the appended claims.

[0132] Example 1 is a system for generating a beam model for a radiation therapy treatment plan for treating a patient via a radiation therapy device. The system includes: a memory for storing a trained deep learning (DL) model and a plurality of beam model parameters; and a processor circuit configured to: receive machine scan data indicating a configuration or operating state of the radiation therapy device; apply the received machine scan data to the trained DL model to determine values ​​of a plurality of beam model parameters; generate a beam model based on the determined values ​​of the plurality of beam model parameters; and store the beam model in a memory accessible to a user or a treatment planning system.

[0133] In Example 2, the subject matter of Example 1 can optionally include a beam model that can include one or more virtual particle sources, the virtual particle sources comprising: a primary photon source; a scattered photon source; or an electron contamination source.

[0134] In Example 3, the subject matter of any one or more of Examples 1 to 2 optionally includes a processor circuit, which may include a training module, which is configured to: access simulated machine scan data measured in a radiation simulation, wherein the simulated machine scan data is measured by a radiation therapy device programmed with known values ​​of beam model parameters; construct a training data set based on the simulated machine scan data; and train the DL model by applying the training data set and the known values ​​of the beam model parameters.

[0135] In Example 4, the subject matter of Example 3 optionally includes simulated machine scan data, which may include at least one dose characteristic, the dose characteristic including: a percentage depth dose (PDD) curve; a dose distribution; a dose volume histogram; an overlapping volume histogram; or a three-dimensional dose distribution.

[0136] In Example 5, the subject matter of any one or more of Examples 3 to 4 optionally includes simulated machine scan data, which may include multiple dose distributions corresponding to multiple field sizes, and the training module is configured to: interpolate dose values ​​between at least two of the multiple dose distributions corresponding to different field sizes; and construct a training data set using samples taken from the multiple dose distributions including the interpolated dose values.

[0137] In Example 6, the subject matter of Example 5 optionally includes a training module, which can be configured to: generate a two-dimensional (2D) dose matrix using multiple dose distributions corresponding to multiple field sizes, the 2D dose matrix representing dose values ​​for multiple positions and corresponding to multiple field sizes; and construct a training data set by sampling the 2D dose matrix at a sampling resolution according to the field size.

[0138] In Example 7, the subject matter of any one or more of Examples 3 to 6 may optionally include a training module that may be configured to apply additive noise to simulated machine scan data to construct a training dataset that enables improvement of the DL model.

[0139] In Example 8, the subject matter of any one or more of Examples 3 to 7 optionally includes: a first device comprising a training module; and a second device configured to receive a trained DL model from the first device via a communication channel and generate a beam model by applying the received machine scan data to the DL model.

[0140] In Example 9, the subject matter of any one or more of Examples 1 to 8 may optionally include a trained DL model, which may include at least one of: a convolutional neural network (CNN); a recurrent neural network (RNN); a long and short-term memory (LSTM) network; a deep belief network (DBN); a generative adversarial network; a machine learning model; a transfer learning module; or an artificial intelligence-based learning model.

[0141] In Example 10, the subject matter of Example 9 optionally includes a trained DL model, which may include a CNN with an initial module.

[0142] In Example 11, the subject matter of any one or more of Examples 1 to 10 optionally includes a trained DL model, which may include a denoising module, and the processor circuit is configured to generate the trained deep learning model, including being configured to train the denoising module to remove or attenuate noise from the training data.

[0143] In Example 12, the subject matter of any one or more of Examples 1 to 11 optionally includes a processor circuit that can be further configured to: calculate a dose distribution using the generated beam model having the determined beam model parameter values; and determine that the beam model passes a validation check if the calculated dose distribution satisfies conditions related to beam characteristics measured from the radiotherapy device.

[0144] In Example 13, the subject matter of any one or more of Examples 1 to 12 optionally includes a treatment planning system configured to generate a radiation therapy treatment plan for a patient based on the beam model.

[0145] In Example 14, the subject matter of any one or more of Examples 1 to 13 may optionally include a user interface configured to present the trained DL model or bundle model to a user.

[0146] Example 15 is a method for generating a beam model for a radiation therapy treatment plan, the radiation therapy treatment plan being used to treat a patient via a radiation therapy device, the method comprising: receiving machine scan data indicating a configuration or operating state of the radiation therapy device; providing a trained deep learning (DL) model associated with a plurality of beam model parameters; applying the received machine scan data to the trained DL model to determine values ​​of the plurality of beam model parameters; generating a beam model based on the determined values ​​of the plurality of beam model parameters; and outputting the beam model to a user or a treatment planning system.

[0147] In Example 16, the subject matter of Example 15 can optionally include a beam model, which can include one or more virtual particle sources, including: a primary photon source; a scattered photon source; or an electron contamination source.

[0148] In Example 17, the subject matter of any one or more of Examples 15 to 16 optionally includes: accessing simulated machine scan data measured in a radiation simulation, wherein the simulated machine scan data is measured by a radiation therapy device programmed with known values ​​of beam model parameters; constructing a training data set based on the simulated machine scan data; and generating a trained DL model by applying the training data set and the known values ​​of the beam model parameters.

[0149] In Example 18, the subject matter of Example 17 optionally includes simulated machine scan data, which may include multiple dose distributions corresponding to multiple field sizes, and wherein constructing the training data set includes: interpolating dose values ​​between at least two of the multiple dose distributions corresponding to different field sizes; and constructing the training data set using samples taken from the multiple dose distributions including the interpolated dose values.

[0150] In Example 19, the subject matter of Example 18 optionally includes constructing a training data set, which may include: generating a two-dimensional (2D) dose matrix using multiple dose distributions corresponding to multiple field sizes, the 2D dose matrix representing dose values ​​for multiple positions and corresponding to multiple field sizes; and sampling the 2D dose matrix at a sampling resolution according to the field size.

[0151] In Example 20, the subject matter of any one or more of Examples 18 to 19 optionally includes applying additive noise to the simulated machine scan data to construct a training dataset that enables improvement of the DL model.

[0152] In Example 21, the subject matter of any one or more of Examples 15 to 20 may optionally include a trained DL model, which may include a convolutional neural network (CNN) having an initial module.

[0153] In Example 22, the subject matter of any one or more of Examples 15 to 21 optionally includes a trained DL model, which may include a denoising module, and generating the trained DL model may include training the denoising module to remove or attenuate noise from the training data set.

[0154] In Example 23, the subject matter of any one or more of Examples 15 to 22 optionally includes: calculating a dose distribution using the generated beam model having the determined beam model parameter values; and determining that the beam model passes a validation check if the calculated dose distribution satisfies conditions related to beam characteristics measured from the radiotherapy device.

[0155] In Example 24, the subject matter of any one or more of Examples 15 to 23 optionally includes generating, via the treatment planning system, a radiation therapy treatment plan for the patient based on the beam model.

[0156] Additional Notes

[0157] The above detailed description includes reference to the accompanying drawings, which form a part of the detailed description. The accompanying drawings show specific examples by way of illustration rather than by way of limitation, in which the present disclosure can be put into practice. These examples are also referred to as "examples" in this article. These examples may include elements other than those shown or described. However, the inventors have also contemplated providing examples of those elements that are only shown or described. In addition, with respect to a specific example (or one or more aspects thereof) or with respect to other examples (or one or more aspects thereof) shown or described in this article, the inventors have also contemplated using examples of any combination or permutation (or one or more aspects thereof) of those elements shown or described.

[0158] All publications, patents, and patent documents referenced in this document are incorporated herein by reference in their entirety, as if individually incorporated by reference. If there are inconsistent usages between this document and those documents incorporated by reference, the usage in the incorporated reference or references should be considered supplementary to that of this document; for conflicting inconsistencies, the usage in this document controls.

[0159] In this document, when introducing elements of various aspects of the present disclosure or examples thereof, the terms "a", "an", "the", and "said" are used, as is common in patent documents, to include one or more than one or more of the elements, independent of any other instances or uses of "at least one" or "one or more". In this document, the term "or" is used to indicate a non-exclusive, or, such that "A or B" includes "A but not B", "B but not A", and "A and B" unless otherwise specified.

[0160] In the appended aspects, the terms "including" and "in which" are used as the plain-English equivalents of the respective terms "comprising" and "wherein." Further, in the following aspects, the terms "comprising," "including," and "having" are intended to be open-ended, meaning that there may be additional elements in addition to the listed elements, such that anything following such a term (e.g., comprising, including, having) in an aspect is still considered to fall within the scope of that aspect. Further, in the following aspects, the terms "first," "second," and "third," etc. are used merely as labels, and are not intended to impose numerical requirements on their objects.

[0161] The examples of the present disclosure can be implemented with computer executable instructions. Computer executable instructions (e.g., software code) can be organized into one or more computer executable components or modules. Any number of such components or modules and any organization of such components or modules can be used to implement the various aspects of the present disclosure. For example, the various aspects of the present disclosure are not limited to the specific computer executable instructions or specific components or modules shown in the drawings and described in this article. Other examples of the present disclosure may include different computer executable instructions or components with more or less functions than those shown and described herein.

[0162] The method examples described herein (e.g., operations and functions) may be at least partially machine or computer implemented (e.g., implemented as software code or instructions). Some examples may include a computer-readable medium or machine-readable medium encoded with instructions that can be operated to configure an electronic device to perform the method as described in the above examples. The implementation of such a method may include software code such as microcode, assembly language code, high-level language code, etc. (e.g., "source code"). Such software code may include computer-readable instructions (e.g., "objects" or "executable code") for performing various methods. The software code may form part of a computer program product. The software implementation of the examples described herein may be provided via an article having codes or instructions stored thereon, or via a method of operating a communication interface to send data via a communication interface (e.g., wirelessly, via the Internet, via satellite communications, etc.).

[0163] In addition, the software code may be tangibly stored on one or more volatile or non-volatile computer-readable storage media during execution or at other times. These computer-readable storage media may include any mechanism for storing information in a form that can be accessed by a machine (e.g., a computing device, an electronic system, etc.), such as, but not limited to, a floppy disk, a hard disk, a removable disk, any form of disk storage media, a CD-ROM, a magneto-optical disk, a removable optical disk (e.g., a compact disc and a digital video disk), a flash memory device, a cassette, a memory card or stick (e.g., a secure digital card), a RAM (e.g., a CMOS RAM, etc.), a recordable / non-recordable medium (e.g., a read-only memory (ROM)), an EPROM, an EEPROM, or any type of medium suitable for storing electronic instructions, etc. Such a computer-readable storage medium is coupled to a computer system bus so that it can be accessed by the processor and other parts of the OIS.

[0164] In an example, the computer readable storage medium may have encoded a data structure for a treatment plan, wherein the treatment plan may be adaptive. The data structure for the computer readable storage medium may be in at least one of a Digital Imaging and Communications in Medicine (DICOM) format, an extended DICOM format, an XML format, and the like. DICOM is an international communication standard that defines a format for transmitting data related to medical images between various types of medical equipment. DICOM RT refers to a communication standard specific to radiation therapy.

[0165] In various examples of the present disclosure, the methods of creating components or modules can be implemented in software, hardware, or a combination thereof. For example, the methods provided by the various examples of the present disclosure can be implemented in software using standard programming languages ​​such as, for example, C, C++, Java, Python, etc., and combinations thereof. As used herein, the terms "software" and "firmware" are interchangeable and include any computer program stored in a memory for execution by a computer.

[0166] A communication interface includes any mechanism that interfaces with any of a hardwired medium, a wireless medium, an optical medium, etc. to communicate with another device, such as a memory bus interface, a processor bus interface, an Internet connection, a disk controller, etc. The communication interface may be configured by providing configuration parameters and / or sending signals to prepare the communication interface to provide data signals describing the software content. The communication interface may be accessed via one or more commands or signals sent to the communication interface.

[0167] The present disclosure also relates to a system for performing the operations herein. The system can be specially constructed for the desired purpose, or can include a general-purpose computer selectively activated or reconfigured by a computer program stored in a computer. Unless otherwise specified, the operation or execution order of the operations in the examples of the present disclosure shown and described herein is not necessary. That is, unless otherwise specified, operations can be performed in any order, and compared with those operations disclosed herein, the examples of the present disclosure may include other operations or fewer operations. For example, it is contemplated that before another operation, simultaneously with another operation, or after another operation, performing or performing a specific operation falls within the scope of the various aspects of the present disclosure.

[0168] In view of the above, it will be seen that several objects of the present disclosure are achieved and other advantageous results are obtained. Having described various aspects of the present disclosure in detail, it will be apparent that modifications and variations are feasible without departing from the scope of the various aspects of the present disclosure as defined in the attached various aspects. Since various changes can be made in the above constructions, products and methods without departing from the scope of the various aspects of the present disclosure, it is intended that all contents contained in the above description and shown in the accompanying drawings should be interpreted as illustrative, and not in a restrictive sense.

[0169] The above description is intended to be illustrative, rather than restrictive. For example, the above examples (or one or more aspects thereof) can be used in combination with each other. In addition, without departing from the scope of the present disclosure, many modifications can be made to adapt specific situations or materials to the teachings of the present disclosure. Although the size and type of the material and coating described herein are intended to limit the parameters of the present disclosure, they are by no means restrictive, but exemplary examples. After reviewing the above description, many other examples will be apparent to those skilled in the art. Therefore, the scope of the present disclosure should be determined with reference to the full range of equivalents to which each of the attached aspects and these aspects are assigned.

[0170] In addition, in the above specific embodiments, various features can be combined together to simplify the present disclosure. This should not be interpreted as an unexpected disclosure feature being necessary for any aspect. Instead, the subject matter of the present invention may be less than all the features of a specific disclosed example. Therefore, the following aspects are thus incorporated into the specific embodiments, with each aspect acting as a separate example alone. The scope of the present disclosure should be determined with reference to the attached various aspects and the full range of equivalents to which these aspects are assigned. In addition, the following limitations of various aspects are not written in a device plus function format, and are not intended to be interpreted based on the sixth paragraph of 35 U.S.C. §112, unless and until the limitation of such aspects explicitly uses the phrase "device for...", "for" is followed by a functional statement without further structure.

[0171] The abstract is provided to allow the reader to quickly ascertain the nature of the technical disclosure. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims.

[0172] According to the above description, the embodiments of the present invention also disclose the following technical solutions, including but not limited to:

[0173] Solution 1. A system for generating a beam model for a radiation therapy treatment plan for treating a patient via a radiation therapy device, the system comprising:

[0174] A memory for storing a trained deep learning model and a plurality of beam model parameters; and

[0175] A processor circuit configured to:

[0176] receiving machine scan data indicating a configuration or operating status of the radiation therapy device;

[0177] applying the received machine scan data to the trained deep learning model to determine values ​​for the plurality of beam model parameters;

[0178] generating a bundle model based on the determined values ​​of the plurality of bundle model parameters; and

[0179] The beam model is stored in the memory accessible by a user or a treatment planning system.

[0180] Solution 2. The system according to solution 1, wherein the beam model includes one or more virtual particle sources, and the one or more virtual particle sources include:

[0181] Primary photon source;

[0182] A source of scattered photons; or

[0183] Sources of electronic pollution.

[0184] Solution 3. A system according to any one of Solution 1 or 2, wherein the processor circuit includes a training module, and the training module is configured to:

[0185] accessing simulated machine scan data measured in a radiation simulation, wherein the simulated machine scan data is measured by the radiation therapy device programmed with known values ​​of beam model parameters;

[0186] constructing a training data set based on the simulated machine scan data; and

[0187] A deep learning model is trained by applying the training data set and known values ​​of the bundle model parameters.

[0188] Embodiment 4. The system of embodiment 3, wherein the simulated machine scan data includes at least one dose characteristic, the at least one dose characteristic comprising:

[0189] Percent Depth Dose (PDD) curve;

[0190] Dose distribution;

[0191] Dose-volume histogram;

[0192] Overlap volume histogram; or

[0193] Three-dimensional dose distribution.

[0194] Embodiment 5. The system of embodiment 3, wherein the simulated machine scan data includes a plurality of dose distributions corresponding to a plurality of field sizes, and the training module is configured to:

[0195] interpolating dose values ​​between at least two of the plurality of dose distributions corresponding to different field sizes; and

[0196] The training data set is constructed using samples taken from the plurality of dose distributions including interpolated dose values.

[0197] Solution 6. The system according to Solution 5, wherein the training module is configured to:

[0198] generating a two-dimensional (2D) dose matrix using the plurality of dose distributions corresponding to the plurality of field sizes, the 2D dose matrix representing dose values ​​calculated for the plurality of locations and the plurality of field sizes; and

[0199] The training data set is constructed by sampling the 2D dose matrix at a sampling resolution according to the field size.

[0200] Option 7. A system according to any one of Options 3 to 6, wherein the training module is configured to apply additive noise to the simulated machine scan data to construct the training data set that enables improvement of the deep learning model.

[0201] Solution 8. A system according to any one of Solution 3 to Solution 7, comprising:

[0202] A first device comprising the training module; and

[0203] A second device is configured to receive the trained deep learning model from the first device via a communication channel and generate the beam model by applying the received machine scan data to the deep learning model.

[0204] Solution 9. The system according to any one of Solution 1 to Solution 8, wherein the trained deep learning model comprises at least one of the following:

[0205] Convolutional Neural Networks (CNN);

[0206] Recurrent Neural Networks (RNNs);

[0207] Long and short-term memory (LSTM) networks;

[0208] Deep Belief Network (DBN);

[0209] Generative Adversarial Networks;

[0210] Machine learning models;

[0211] Transfer Learning Module; or

[0212] Artificial intelligence based learning model.

[0213] Option 10. A system according to Option 9, wherein the trained deep learning model includes a CNN with an initial module.

[0214] Option 11. A system according to any one of Options 1 to 10, wherein the trained deep learning model includes a denoising module, and the processor circuit is configured to generate the trained deep learning model including training the denoising module to remove or attenuate noise from training data.

[0215] Embodiment 12. A system according to any one of embodiments 1 to 11, wherein the processor circuit is further configured to:

[0216] calculating a dose distribution using the generated beam model having the determined beam model parameter values; and

[0217] The beam model is determined to pass the validation check if the calculated dose distribution meets conditions related to beam characteristics measured from the radiotherapy apparatus.

[0218] Embodiment 13. The system of any one of Embodiments 1 to 12, comprising the treatment planning system configured to generate a radiation therapy treatment plan for the patient based on the beam model.

[0219] Option 14. A system according to any one of Options 1 to 13, comprising a user interface configured to present the trained deep learning model or the beam model to the user.

[0220] 15. A method for generating a beam model for a radiation therapy treatment plan for treating a patient via a radiation therapy device, the method comprising:

[0221] receiving machine scan data indicating a configuration or operating status of the radiation therapy device;

[0222] providing a trained deep learning model associated with a plurality of bundle model parameters;

[0223] applying the received machine scan data to the trained deep learning model to determine values ​​for the plurality of beam model parameters;

[0224] generating a bundle model based on the determined values ​​of the plurality of bundle model parameters; and

[0225] The beam model is output to a user or a treatment planning system.

[0226] Embodiment 16. The method according to embodiment 15, wherein the beam model includes one or more virtual particle sources, and the virtual particle sources include:

[0227] Primary photon source;

[0228] A source of scattered photons; or

[0229] Sources of electronic pollution.

[0230] Scheme 17. The method according to any one of Schemes 15 or 16, comprising:

[0231] accessing simulated machine scan data measured in a radiation simulation, wherein the simulated machine scan data is measured by the radiation therapy device programmed with known values ​​of beam model parameters;

[0232] constructing a training data set based on the simulated machine scan data; and

[0233] The trained deep learning model is generated by applying the training data set and known values ​​of the bundle model parameters.

[0234] Embodiment 18. The method of embodiment 17, wherein the simulated machine scan data includes a plurality of dose distributions corresponding to a plurality of field sizes, and wherein constructing the training data set includes:

[0235] interpolating dose values ​​between at least two of the plurality of dose distributions corresponding to different field sizes; and

[0236] The training data set is constructed using samples taken from the plurality of dose distributions including interpolated dose values.

[0237] Solution 19. The method according to Solution 18, wherein constructing the training data set comprises:

[0238] generating a two-dimensional (2D) dose matrix using the plurality of dose distributions corresponding to the plurality of field sizes, the 2D dose matrix representing dose values ​​calculated for the plurality of locations and the plurality of field sizes; and

[0239] The 2D dose matrix is ​​sampled with a sampling resolution that depends on the field size.

[0240] Option 20. The method according to Option 18 also includes applying additive noise to the simulated machine scan data to construct the training data set that enables improvement of the deep learning model.

[0241] Option 21. A method according to any one of Options 15 to 20, wherein the trained deep learning model includes a convolutional neural network (CNN) with an initial module.

[0242] Option 22. A method according to any one of Options 15 to 21, wherein the trained deep learning model includes a denoising module, and generating the trained deep learning model includes training the denoising module to remove or attenuate noise from the training data set.

[0243] Embodiment 23. The method according to any one of embodiments 15 to 22, further comprising:

[0244] calculating a dose distribution using the generated beam model having the determined beam model parameter values; and

[0245] The beam model is determined to pass the validation check if the calculated dose distribution meets conditions related to beam characteristics measured from the radiotherapy apparatus.

[0246] Option 24. The method of any one of Option 15 to Option 23, comprising generating, via the treatment planning system, a radiation therapy treatment plan for the patient based on the beam model.

Claims

1. A radiotherapy planning system, comprising: A processor configured to: constructing a training data set using simulated machine scan data measured by a radiation therapy device programmed with known values ​​of beam model parameters in a radiation simulation; modifying the training data set with additive noise of the simulated machine scan data; training a deep learning model using the modified training data set, the deep learning model being trained to establish a relationship between the modified training data set and known values ​​of the bundle model parameters; as well as generating a beam model for a radiation therapy plan by applying the trained deep learning model to machine scan data collected from the radiation therapy device and indicating a configuration or operating state of the radiation therapy device; as well as A memory circuit is configured to store the generated bundle model.

2. The radiation therapy planning system according to claim 1, wherein: The additive noise includes the difference between the calculated dose distribution and the measured dose distribution.

3. The radiation therapy planning system according to claim 1, wherein: The trained deep learning model includes a denoising module, Wherein, training the deep learning model includes training the denoising module to remove or attenuate additive noise from the modified training data set.

4. The radiation therapy planning system according to claim 1, wherein: The beam model includes one or more virtual particle sources, the one or more virtual particle sources including: Primary photon source; A source of scattered photons; or Sources of electronic pollution.

5. The radiation therapy planning system according to claim 1, wherein: The simulated machine scan data includes at least one dose characteristic, the at least one dose characteristic including: Percent depth dose PDD curve; Dose distribution; Dose-volume histogram; Overlap volume histogram; or Three-dimensional dose distribution.

6. The radiation therapy planning system according to claim 1, wherein: The processor is configured to further construct the training data set using interpolation or extrapolation of the simulated machine scan data.

7. The radiation therapy planning system according to claim 6, wherein: The simulated machine scan data includes a plurality of dose distributions corresponding to a plurality of field sizes, The interpolation or extrapolation includes an interpolated dose value between at least two of the multiple dose distributions corresponding to different field sizes or an extrapolated dose value beyond the at least two of the multiple dose distributions corresponding to different field sizes.

8. The radiation therapy planning system according to claim 7, wherein: The processor is configured to: generating a two-dimensional (2D) dose matrix using the plurality of dose distributions corresponding to the plurality of field sizes, the 2D dose matrix representing dose values ​​calculated for a plurality of locations and the plurality of field sizes; as well as The training data set is constructed including sampling of the 2D dose matrix at a sampling resolution according to the field size.

9. The radiation therapy planning system according to claim 1, wherein: The processor is further configured to: calculating a dose distribution using the generated beam model; and The beam model is determined to pass the validation check if the calculated dose distribution meets conditions related to beam characteristics measured from the radiotherapy apparatus.

10. The radiation therapy planning system according to claim 1, in, The processor is configured to generate a radiation therapy treatment plan based at least in part on the beam model, Wherein the storage circuit is configured to store the radiation therapy treatment plan accessible by the radiation therapy device.

11. A computer-implemented method for planning radiation therapy, the method comprising: constructing a training data set using simulated machine scan data measured by a radiation therapy device programmed with known values ​​of beam model parameters in a radiation simulation; modifying the training data set with additive noise of the simulated machine scan data; training a deep learning model using the modified training data set, the deep learning model being trained to establish a relationship between the modified training data set and known values ​​of the bundle model parameters; as well as A beam model for a radiation therapy plan is generated by applying a trained deep learning model to machine scan data collected from the radiation therapy device and indicating a configuration or operating state of the radiation therapy device.

12. The method according to claim 11, wherein: The additive noise includes the difference between the calculated dose distribution and the measured dose distribution.

13. The method according to claim 11, wherein: The trained deep learning model includes a denoising module, Wherein, training the deep learning model includes training the denoising module to remove or attenuate the additive noise from the modified training data set.

14. The method according to claim 11, wherein: The beam model includes one or more virtual particle sources, the one or more virtual particle sources including: Primary photon source; A source of scattered photons; or Sources of electronic pollution.

15. The method according to claim 11, wherein: The simulated machine scan data includes at least one dose characteristic, the at least one dose characteristic including: Percent depth dose PDD curve; Dose distribution; Dose-volume histogram; Overlap volume histogram; or Three-dimensional dose distribution.

16. The method of claim 11, further comprising interpolating or extrapolating the simulated machine scan data, in, The training data set includes interpolated or extrapolated simulated machine scan data.

17. The method according to claim 16, wherein: The simulated machine scan data includes a plurality of dose distributions corresponding to a plurality of field sizes, Wherein, interpolating or extrapolating the simulated machine scan data includes interpolating dose values ​​between at least two of the multiple dose distributions corresponding to different field sizes, or extrapolating dose values ​​beyond the at least two of the multiple dose distributions corresponding to different field sizes.

18. The method of claim 17, further comprising generating a two-dimensional (2D) dose matrix using the plurality of dose distributions corresponding to the plurality of field sizes, the 2D dose matrix representing dose values ​​calculated for a plurality of locations and the plurality of field sizes, in, The training data set includes a sampling of the 2D dose matrix at a sampling resolution according to the field size.

19. The method according to claim 11, further comprising: calculating a dose distribution using the generated beam model; as well as The beam model is determined to pass the validation check if the calculated dose distribution meets conditions related to beam characteristics measured from the radiotherapy apparatus.

20. The method of claim 11, further comprising generating a radiation therapy treatment plan for the radiation therapy device based at least in part on the beam model.

Images