Dose-directed radiotherapy plan generation using computer modeling techniques

Abstract

Methods and systems are provided herein for training and executing a computer model that uses an artificial intelligence approach (e.g., deep learning) to learn and predict multi-leaf collimator (MLC) openings and control the weights of radiotherapy treatment plans.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Application No. 17 / 839,206, filed on June 13, 2022, which is incorporated herein by reference in its entirety. Technical Field

[0003] The present application generally relates to the use of data analysis techniques to model and predict properties of radiation therapy treatments and to control radiation therapy machines. Background Art

[0004] Radiation therapy treatment planning (RTTP) is a complex process that involves specific guidelines, protocols, and instructions adopted by different medical professionals (such as clinicians, medical device manufacturers, etc.). Typically, identifying and applying guidelines for implementing the radiation therapy treatment is performed by a complex computer model that receives treatment goals from the treating physician and identifies appropriate properties of the RTTP. For example, the treating physician can identify the treatment modality (e.g., selecting between volumetric modulated arc therapy (VMAT) or intensity modulated radiation therapy (IMRT)). The treating physician can then input various goals and objectives to be achieved via the treatment, such as dose targets to be achieved for one or more structures of the patient. The software solution can then use various methods to calculate the properties of the patient's treatment, such as determining beam limiting device angles and radiation emission properties. In the case of IMRT, the beam delivery direction and number of beams are specific related variables that must be determined, while for VMAT, the software solution may need to select the number of arcs and their corresponding start and stop angles.

[0005] Therefore, for plan generation (and in particular for VMAT plans), some software solutions use an iterative trial-and-error process to optimize various properties of the RTTP. For example, after a treatment physician inputs a treatment goal, the software solution iteratively analyzes different possibilities (e.g., different iterations of different properties within a large search space) to identify which iteration produces the best (or acceptable) results. Specifically, a VMAT plan optimizer software solution can configure a set of linear accelerator machine instructions, such as a multi-leaf collimator (MLC) sequence and control point weights, to deliver a dose that meets the treatment goal that reflects the radiation oncologist's purpose. Therefore, the software solution may require a large amount of computing resources and may not produce results in a timely manner. Summary of the invention

[0006] For the reasons described above, there is a need for a system that can quickly and accurately analyze planning / treatment goals and patient information to provide RTTP properties. Using the methods and systems discussed herein, a computer model (e.g., an artificial intelligence (AI) or machine learning (ML) model) can utilize various AI / ML techniques, such as deep learning, to propose a set of VMAT machine instructions that include MLC sequences and control point weights, given the desired dose distribution and associated planning structure goals.

[0007] Using the methods and systems discussed herein, a processor can use deep learning to train a model to predict MLC openings for patient treatment. Specifically, a computer model can use data-driven statistical learning methods (e.g., deep learning) to establish correlations between image-level features of dose distribution and linear accelerator machine instructions, such as MLC order and control point weights. Using these learned correlations, a computer model (e.g., a software solution that generates RTTP) can quickly generate a set of machine instructions that allow a certain desired dose distribution to be delivered.

[0008] Currently, VMAT plans are typically generated in the clinic using a plan optimizer alone. However, this process is time consuming because multiple optimization iterations are typically required (partly due to reliance on two-dimensional (2D) plan targets based on DVHs), and each optimization iteration takes a long time to analyze due to the large search space. Using the methods and systems described herein, a computer model can reduce the time it takes traditional software solutions to generate RTTP. The described computer model can reduce time by improving current solutions in at least two different ways. First, the methods and systems described herein can use three-dimensional (3D) dose distributions as planning targets, rather than 2D DVH-based planning targets. Second, the methods and systems described herein utilize deep learning and / or other AI / ML techniques to quickly predict a set of linear accelerator machine instructions that can produce a desired dose distribution with minimal demand on an optimizer.

[0009] Using the methods and systems described herein, a server can train a model using various AI / ML techniques (such as deep learning) to generate a full linear accelerator machine instruction set to deliver a VMAT plan (including MLC openings and control point weights) using the desired 3D dose distribution, radiotherapy plan structure, and simulated medical images (e.g., computed tomography (CT) images) as input.

[0010] The model can predict the shape of the MLC opening and the corresponding weights for each control point. As used herein, "weights" can refer to the movement properties of the linear accelerator at a specific point (control point) during its rotation. For example, the weight can indicate the movement speed (or angular velocity) of the linear accelerator at the control point. In another example, the weight can indicate the angle of movement. In some embodiments, the weight can correspond to multiple attributes. For example, the weight can correspond to a movement property and a dose (e.g., indicating how the accelerator moves and the dose emitted at the same time). In essence, the control point weight can indicate how radiation is applied at a specific position / angle.

[0011] The MLC opening and control point weights may change during processing. For example, each position of the linear accelerator may correspond to a specific control point weight and MLC opening. In one non-limiting example, the MLC may have a first opening and a first control point weight at a first control point, and a completely different MLC opening and control point weight at a second control point. Using the methods and systems discussed herein, the server can predict the expected MLC opening and control point weights at any given time or position of the linear accelerator that will produce results that meet the planning goals.

[0012] In one embodiment, a method may include receiving, by a processor, a treatment target for a patient, the treatment target comprising at least a dose-volume of at least one structure of the patient to be treated via a radiotherapy machine; executing, by the processor, an artificial intelligence model to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set comprising training treatment targets and properties associated with previously performed radiotherapy treatments, the training treatment targets and properties comprising at least an actual or projected dose-volume histogram of the treated patient and a corresponding MLC opening position; and presenting, by the processor, a predicted MLC opening position of the radiotherapy machine and corresponding predicted movement properties of the accelerator.

[0013] The artificial intelligence model can be trained via a deep learning protocol to correlate actual or projected dose-volume histograms of treated patients and the corresponding MLC opening locations.

[0014] The predicted MLC opening positions may be a binary mask indicating the openings of the MLCs.

[0015] The predicted movement attribute may be a time associated with the movement of the accelerator.

[0016] The predicted movement attribute may be an angle associated with the movement of the accelerator.

[0017] The AI ​​model can also predict the order in which patients' MLC openings will occur.

[0018] The AI ​​model can utilize a loss function based on the MLC opening limit.

[0019] The method may also include generating, by the processor, machine readable instructions based on the predicted MLC openings and the corresponding predicted movement attributes.

[0020] The method may also include sending, by the processor, the machine-readable instructions to a radiation therapy machine.

[0021] In another embodiment, a computer system may include a server comprising a processor and a non-transitory computer-readable medium containing instructions, which, when executed by the processor, causes the processor to perform operations including: receiving a treatment goal for a patient, the treatment goal comprising at least a dose-volume of at least one structure of the patient to be treated by a radiotherapy machine; executing an artificial intelligence model to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set comprising training treatment goals and properties associated with previously performed radiotherapy treatments, the training treatment goals and properties comprising at least an actual or projected dose-volume histogram of the treated patient and a corresponding MLC opening position; and presenting a predicted MLC opening position of the radiotherapy machine and corresponding predicted movement properties of the accelerator.

[0022] The artificial intelligence model can be trained via a deep learning protocol to correlate actual or projected dose-volume histograms of treated patients and corresponding MLC opening locations.

[0023] The predicted MLC opening positions may be a binary mask indicating the openings of the MLCs.

[0024] The predicted movement attribute may be a time associated with the movement of the accelerator.

[0025] The predicted movement attribute may be an angle associated with the movement of the accelerator.

[0026] The AI ​​model can also predict the order in which patients' MLC openings will occur.

[0027] The AI ​​model can utilize a loss function based on the MLC opening limit.

[0028] The instructions may also cause the processor to generate machine readable instructions based on the predicted MLC opening and the corresponding predicted movement attributes.

[0029] The instructions may also cause the processor to send the machine-readable instructions to a radiation therapy machine.

[0030] In another embodiment, a system may include a server having one or more processors configured to receive a treatment goal for a patient, the treatment goal including at least a dose-volume of at least one structure of the patient to be treated via a radiotherapy machine; execute an artificial intelligence model to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set including training treatment goals and properties associated with previously performed radiotherapy treatments, the training treatment goals and properties including at least an actual or projected dose-volume histogram of the treated patient and the corresponding MLC opening position; and present a predicted MLC opening position of the radiotherapy machine and the corresponding predicted movement properties of the accelerator.

[0031] The artificial intelligence model can be trained via a deep learning protocol to correlate actual or projected dose-volume histograms of treated patients and the corresponding MLC opening locations. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Non-limiting embodiments of the present disclosure are described by way of example with reference to the accompanying drawings, which are schematic and not drawn to scale. Unless otherwise indicated to represent background technology, the accompanying drawings represent various aspects of the present disclosure.

[0033] Figure 1 Illustrated are components of an artificial intelligence (AI) plan generation system according to one embodiment.

[0034] Figure 2A A flow chart of a process performed in an AI plan generation system according to one embodiment is illustrated.

[0035] Figure 2B A flow chart of a process executed in an AI model in an AI plan generation system according to one embodiment is illustrated.

[0036] Figure 2C A flow chart of a process executed in an AI model in an AI plan generation system according to one embodiment is illustrated.

[0037] Figure 3 Illustrated is a visual representation of processed data for training an AI model in an AI plan generation system according to one embodiment.

[0038] Figure 4 Illustrated is a visual representation of processed data for training an AI model in an AI plan generation system according to one embodiment.

[0039] Figure 5 Illustrated is a visual representation of training an AI model in an AI plan generation system according to one embodiment.

[0040] Figure 6Illustrated is a non-limiting example of data presented in an AI plan generation system according to one embodiment.

[0041] Figure 7 Illustrated is a non-limiting example of data presented in an AI plan generation system according to one embodiment.

[0042] Figure 8 A flow chart of a process performed in an AI plan generation system according to one embodiment is illustrated. DETAILED DESCRIPTION

[0043] Reference will now be made to the illustrative embodiments depicted in the accompanying drawings, and specific language will be used to describe them herein. However, it should be understood that this is not intended to limit the scope of the claims or the disclosure. Changes and further modifications to the inventive features illustrated herein, as well as additional applications of the principles of the illustrated subject matter herein, that would occur to those skilled in the relevant art and those in possession of the disclosure, are to be considered within the scope of the subject matter disclosed herein. Other embodiments may be used and / or other changes may be made without departing from the spirit or scope of the disclosure. The illustrative embodiments described in the detailed specification are not meant to limit the subject matter presented.

[0044] Figure 1 Components of a system 100 of an artificial intelligence plan generation system according to one embodiment are illustrated. The system 100 may include an analysis server 110a, a system database 110b, an AI model 111, electronic data sources 120a-d (collectively referred to as electronic data sources 120), end user devices 140a-c (collectively referred to as end user devices 140), an administrator computing device 150, a medical device 160, and a medical device computer 162. Figure 1 The various components depicted in may belong to a radiation therapy treatment clinic where a patient may, in some cases, receive radiation therapy treatment via one or more radiation therapy machines (eg, medical device 160 ).

[0045] The system 100 is not limited to the components described herein, and may include additional or other components, not shown for the sake of brevity, that are considered to be within the scope of the embodiments described herein.

[0046] The components mentioned above may be connected to each other via a network 130. Examples of network 130 may include, but are not limited to, private or public local area networks (LANs), wireless local area networks (WLANs), metropolitan area networks (MANs), wide area networks (WANs), and the Internet. Network 130 may include wired and / or wireless communications according to one or more standards and / or via one or more transmission media. Communications via network 130 may be performed according to various communication protocols, such as transmission control protocol and Internet protocol (TCP / IP), user datagram protocol (UDP), and IEEE communication protocols. In one example, network 130 may include wireless communications according to a Bluetooth specification set or another standard or proprietary wireless communication protocol. In another example, network 130 may also include communications via a cellular network, including, for example, GSM (Global System for Mobile Communications), CDMA (Code Division Multiple Access), or EDGE (Enhanced Data for Global Evolution) networks.

[0047] The analysis server 110a may generate and display an electronic platform configured to receive patient information using various AI models 111 (including artificial intelligence and / or machine learning models) and output the execution results of the AI ​​models 111. The electronic platform may include a graphical user interface (GUI) displayed on each electronic data source 120, end-user device 140, medical device 160, and / or administrator computing device 150. An example of an electronic platform generated and hosted by the analysis server 110a may be a web-based application or website configured to be displayed on different electronic devices, such as mobile devices, tablet computers, personal computers, etc.

[0048] The information displayed by the electronic platform may include, for example, input elements to receive data associated with the patient to be treated (e.g., planned goals), and display the predicted results (e.g., predicted MLC opening images or numerical sequence data and / or control point weights) generated by the AI ​​model 111. The analysis server 110a may then display the results to the medical professional and / or directly correct one or more operational properties of the medical device 160. In some embodiments, the medical device 160 may be a diagnostic imaging device or a treatment delivery device.

[0049] The analysis server 110a can be any computing device including a processor and a non-transitory machine-readable storage device capable of performing the various tasks and processes described herein. The analysis server 110a can employ various processors, such as a central processing unit (CPU) and a graphics processing unit (GPU), etc. Non-limiting examples of such computing devices may include a workstation computer, a laptop computer, a server computer, etc. Although the system 100 includes a single analysis server 110a, the analysis server 110a may include any number of computing devices operating in a distributed computing environment, such as a cloud environment.

[0050] The electronic data source 120 may represent various electronic data sources that contain, retrieve, and / or access data associated with the medical device 160, such as operational information associated with previously performed radiation therapy treatments (e.g., electronic log files or electronic configuration files), data associated with previously monitored patients (e.g., RTTPs of previous patients and their corresponding treatment attributes and other machine instructions included in radiation therapy treatment files), or participants in a study, to train the AI ​​model 111 discussed herein. For example, the analysis server 110a may use a clinic computer 120a, a medical professional device 120b, a server 120c (associated with a physician and / or clinic), and a database 120d (associated with a physician and / or clinic) to retrieve / receive data associated with the medical device 160. The analysis server 110a may retrieve data from the end-user device 120, generate a training data set, and train the AI ​​model 111. The analysis server 110a may execute various algorithms to translate the raw data received / retrieved from the electronic data source 120 into a machine-readable object that may be stored and processed by other analysis processes described herein.

[0051] The end-user device 140 can be any computing device, including a processor and a non-transitory machine-readable storage medium capable of performing the various tasks and processes described herein. Non-limiting examples of the end-user device 140 can be a workstation computer, a laptop computer, a tablet computer, and a server computer. In operation, various users can use the end-user device 140 to access the GUI operably managed by the analysis server 110a. Specifically, the end-user device 140 can include a clinic computer 140a, a clinic server 140b, and a medical professional device 140c. Even if referred to as "end-user" devices in this article, these devices are not always operated by end users. For example, the clinic server 140b may not be used directly by the end user. However, the results stored on the clinic server 140b can be used to fill in various GUIs that the end user accesses via the medical professional device 140c.

[0052] Administrator computing device 150 may represent a computing device operated by a system administrator. Administrator computing device 150 may be configured to display radiation therapy treatment attributes generated by analysis server 110a (e.g., various analytical metrics determined during training of one or more machine learning models and / or systems); monitor various models 111 utilized by analysis server 110a, electronic data sources 120, and / or end-user devices 140; review feedback from end-user devices 140; and / or facilitate training or retraining (calibration) of AI models 111 maintained by analysis server 110a.

[0053] The medical device 160 may be a radiotherapy machine configured to implement radiotherapy treatment of a patient. The medical device 160 may include a linear accelerator and an MLC configured to control the emission of radiation to the patient. The medical device 160 may also communicate with a medical device computer 162, which is configured to display various GUIs discussed herein. For example, the analysis server 110a may display the results predicted by the AI ​​model 111 on a computing device described herein. In a non-limiting example, the GUI may display the projected MLC opening or control point weights predicted by the AI ​​model 111.

[0054] The AI ​​model 111 may be stored in the system database 110b. The AI ​​model 111 may be trained using data received / retrieved from the electronic data source 120 and may be executed using data received from the end-user device, medical device 160, and / or sensor 163. In some embodiments, the AI ​​model 111 may reside in a local or clinic-specific data repository. In various embodiments, the AI ​​model 111 uses one or more deep learning engines to generate MLC openings and control weights for patients.

[0055] It should be understood that any alternative and / or additional (one or more) machine learning models can be used to implement a similar learning engine. The deep learning engine can include a processing path trained during a training phase. Once trained, the deep learning engine can be executed (e.g., by analyzing server 110a) to generate predicted processing attributes.

[0056] As described herein, the analysis server 110a can store an AI model 111 (e.g., a neural network, a random forest, a support vector machine, a regression model, and / or a recurrent model, etc.) in an accessible data repository. The analysis server 110a can retrieve the AI ​​model 111 and train the AI ​​model 111 to predict the patient's treatment attributes, including MLC openings and control point weights.

[0057] Various machine learning techniques may involve "training" a machine learning model to predict treatment attributes, including supervised learning techniques, unsupervised learning techniques, or semi-supervised learning techniques, etc. In a non-limiting example, the predicted patient attributes may indicate MLC openings and / or control point weights. The AI ​​model 111 may therefore be used to predict real-time MLC openings and control point weights (e.g., the position and orientation of a linear accelerator).

[0058] In practice, the data used for the training data set can be generated by users through observation and experience to facilitate training the AI ​​model 111. For example, training data can be received and monitored during a previous radiotherapy treatment provided to a previous patient. In another example, the training data can be a data set including treatment targets (e.g., DVH targets), RTTP, projection dose distribution, MLC openings, and / or control weights associated with previously treated patients. The training data can be processed via any suitable data enhancement method (e.g., normalization, encoding, or any combination thereof) to produce a new data set with modified properties to improve the quality of the data. The methods and systems described herein are not limited to training AI models based on patients that have been previously treated. For example, a training data set can include data associated with any group of participants (not patients) who are willing to be monitored for the purpose of generating a training data set.

[0059] refer to Figure 2A , method 200 illustrates an operational workflow performed in an artificial intelligence plan generation system according to one embodiment. Method 200 may include steps 202-208. However, other embodiments may include additional or alternative steps, or may omit one or more steps entirely. Method 200 is described as being performed by a server, such as Figure 1 However, one or more steps of method 200 may be implemented by Figure 1 For example, one or more computing devices may locally execute Figure 2A Some or all of the steps described in .

[0060] Using method 200, an AI model can be trained based on a training data set to receive desired or predicted / projected dose distribution data, such as a dose target or projected 3D dose volume, and predict the corresponding MLC opening and movement properties of the linear accelerator. In implementation, using method 200, the AI ​​model can be executed in conjunction with a plan optimizer model to prepare the control weights and sequence of MLC openings. The results predicted by the AI ​​model (using method 200) can be displayed to clinicians for their approval and / or ingested by the radiotherapy machine itself.

[0061] At step 202, the analysis server may train an artificial intelligence model to predict images of a multi-leaf collimator (MLC) opening of a radiation therapy machine and corresponding movement properties of an accelerator. Before executing the AI ​​model (e.g., allowing a clinician to use the AI ​​model), the analysis server may first train the AI ​​model and ensure its accuracy. Training the AI ​​model may be done in different stages. For example, the analysis server may first prepare a training data set and train the AI ​​model, as described herein and Figure 2B and Figure 3-Figure 5 The analysis server can then implement the AI ​​model as described herein and Figure 2C Depicted.

[0062] Prepare training dataset

[0063] The analysis server can train the AI ​​model using a training data set that includes at least two sets of data associated with a group of previously treated patients (or participants in a clinical trial). First, the training data set may include RTTP data associated with each patient in the group of patients. For example, the training data may include various goals (e.g., dose goals) entered by a clinician. The training data may also include various data associated with how to generate RTTP. For example, the training data may include how a plan optimizer (or any other model or software solution) uses the plan goals to prepare RTTP for a patient. For example, the training data set may include a projected 3D dose volume. In some embodiments, an image projected by a plan optimizer may be obtained and included in the training data set.

[0064] Second, the training data set may include linear accelerator machine instructions, which indicate details associated with how to implement RTTP. Linear accelerator machine instructions may involve how to emit radiation and how to apply radiation (with properties generated by the plan optimizer for the patient) to the patient. Specifically, the linear accelerator machine instructions may include a log of instructions received by the radiotherapy machine, causing the linear accelerator to adjust and configure the MLC in certain ways (e.g., moving speed, moving angle, and / or MLC opening properties). Therefore, the training data set may include the MLC opening and its corresponding control weights. Specifically, the training data set may include a set of MLC opening data and its corresponding time, so that the AI ​​model can reconstruct the order of how the MLC is opened / configured during the patient treatment process. The training data set may also include control weights associated with the linear accelerator during the patient treatment process. While the patient is being treated, the linear accelerator machine instructions can be periodically obtained (e.g., via a radiotherapy machine log).

[0065] In one embodiment, the training data set may also include medical images (e.g., CT or 4DCT) depicting the patient's internal organs and RTTP implementation data. The medical images may be real or predicted by another model. For example, the training data set may include simulated CT images. Information from the simulated CT images may be used to form the basis for input to the AI ​​model. Specifically, the simulated CT images may allow the delineation of planned structures, where these structures may allow the generation of expected 3D dose distributions. The 3D dose distribution (more precisely, its 2D projection image) may be used as input to the AI ​​model.

[0066] The training dataset may also include additional data associated with the patients. For example, the AI ​​model may take into account demographic information and / or other biomarkers (e.g., age, weight, or body mass index) for each patient. Thus, the model may also take into account the patient's attributes when considering how to implement RTTP. During treatment (and its outcomes), some patients may experience physical changes (e.g., weight loss). As a result, their projection data may change slightly.

[0067] When reviewed as a whole, the training data set can include information that can indicate how each patient's RTTP is applied to the patient. Specifically, the training data set can indicate the properties of each patient's RTTP and the corresponding linear accelerator machine instructions (specifically, MLC openings and control weights). Each MLC opening and / or control weight can be analyzed based on its timestamp and the patient's corresponding RTTP properties. Using this data, the AI ​​model can establish correlations between linear accelerator machine instructions and image-level features of dose distribution.

[0068] The analysis server can then aggregate various data points associated with the group of patients and their treatments to generate an aggregated training data set. The analysis server can also perform various data cleaning protocols, such as deduplication and other analysis protocols, to ensure that the training data set can be ingested by the AI ​​model to produce results.

[0069] like Figure 2B As depicted, before training the artificial intelligence model, the analysis server can also prepare the data in the training data set. First, the analysis server can analyze the treatment data associated with each patient. Specifically, the analysis server can extract and analyze medical imaging and communications (DICOM) radiotherapy files for dose, structure and CT to construct a 3D dose tensor, a 3D structure tensor (e.g., one for each structure) and / or a 3D CT tensor for each treated patient. As used herein, DICOM may refer to standardized diagnostic imaging, which may include DICOM-RT objects (e.g., RT images, RT structure sets, RT plans, RT doses, RT beam processing records, RT close-range processing records, and RT processing summary records). Although various aspects of the present disclosure discuss DICOM as a standard, it should be understood that other standardized or structured data may also be used. As used herein, a tensor may refer to a mathematical element, such as a vector or component array, describing a function associated with a coordinate in a space corresponding to the original data points in the training data set.

[0070] The analysis server can retrieve a DICOM RT dose file, a DICOM RT structure file, a DICOM CT file, and a DICOM RT plan file (block 208), and then construct a 3D dose tensor, a 3D structure tensor, and a 3D CT tensor, respectively (block 210). After preparing the tensors, the analysis server can project the 3D tensor onto a 2D plane from the perspective of the beam. Each 2D projection can have a corresponding control point. For example, the analysis server can generate a 2D projection based on the 3D tensor at each control point. Each 2D projection can also have a corresponding control weight at the corresponding control point. The 3D tensor and the 2D projection can be concatenated to form input data for training an AI model (e.g., a deep learning model). For example, the analysis server can generate a concatenated projected 2D dose and structure tensor using the methods discussed herein (block 212). In a non-limiting example, as Figure 3 As depicted in example 300 depicted in , an analysis server may receive (eg, from a plan optimizer) a desired 3D dose plan structure.

[0071] The desired 3D dose plan structure may be a sequence of dose plan structures 302. The analysis server may first decompose the desired 3D dose plan structure 302 into frames 302a-n. The analysis server may then generate a stack of 2D projections 304 including frames 304a-n with time stamps. Each frame 304a-n corresponds to a plan of beam angles at a particular control point. In addition, as shown in FIG. Figure 4 As depicted (example 400), for each desired 3D dose plan structure 402 (similar to frames 302a-n), the analysis server can generate a projection (projection 406) for each control point. Each dose projection can also be generated / projected according to a corresponding percentage depth dose (PDD) curve. For example, projection 402 can be a weighted average based on the PDD for each sampling plane based on PDD 404. Projection 406 can also be a binary mask corresponding to the plan structure.

[0072] Return to reference Figure 2B , the analysis server may also analyze the data by extracting data associated with the MLC sequence and control point weights from the DICOM RT plan file (block 216). The analysis server may then construct a tensor for a digital representation of the MLC sequence from the extracted MLC sequence data (block 218). The analysis server may then generate a tensor for a graphical representation of the MLC sequence from the retrieved and extracted MLC sequence data (block 220). In addition, the analysis server may then construct a tensor for control point weights from the retrieved and extracted control point weights (block 222).

[0073] Training AI models

[0074] Using the training data set, the analysis server can train one or more AI models discussed herein. In various embodiments, the AI ​​model can use one or more deep learning engines to perform automatic segmentation of received images and / or correlate the data in the training data set so that they reveal patterns connecting how various prepared data correspond to linear accelerator machine instructions (e.g., MLC openings and control point weights). Specifically, AI model 224 can ingest data associated with blocks 212, 218, 220, and 222 to train itself.

[0075] The AI ​​model can first analyze the prepared training data set and determine the patterns between the tensors and the MLC openings and control weights. Using various machine learning techniques, the model can identify how each MLC opening and control weight corresponds to different RTTP attributes (filed, extracted and prepared in the training data set).

[0076] The AI ​​model 224 may include a neural network including a multi-layer convolutional neural network and may train itself using a deep learning method. One type of deep learning engine is a deep neural network (DNN). A DNN is a branch of a neural network and consists of a stack of layers, each of which performs specific operations such as convolution, pooling, loss calculation, etc. Each intermediate layer receives the output of the previous layer as its input. The starting layer is an input layer that is directly connected to or receives an input data structure that includes data items in one or more machine-readable objects and may have a number of neurons that are equal to the data items in one or more machine-readable objects provided as input. For example, a machine-readable object may be a data structure, such as a list or a vector, that includes multiple data fields that contain data in a training data set. Each neuron in the input layer may accept the contents of a data field as input.

[0077] The next set of layers can include any type of layer that may be present in a DNN, such as convolutional layers, fully connected layers, pooling layers, or activation layers. Some layers, such as convolutional neural network layers, can include one or more filters. Filters, often called kernels, have arbitrary sizes defined by the designer. Each neuron can only respond to a specific area of ​​the previous layer, which is called the receptive field. The output of each convolutional layer can be viewed as an activation map, which highlights the effect of applying a specific filter to the input. Convolutional layers can be followed by activation layers to apply nonlinearities to the output of each layer. The next layer can be a pooling layer, which helps reduce the dimensionality of the output of the convolutional layer. In various implementations, high-level abstractions are extracted by fully connected layers. The weights of the neural connections and kernels can be continuously optimized during the training phase.

[0078] Reference now Figure 5, example 500 depicts how to train an AI model using a multi-task learning framework based on a convolutional neural network. In example 500, the AI ​​model ingests dose and structure projections 502 and uses various encoding and decoding techniques to identify hidden patterns between the input data and predicted outputs 506, such as graphical and / or numerical MLC sequence predictions 510 and / or control point weights 508.

[0079] In some embodiments, the AI ​​model may also be configured to use various machine learning techniques to generate a graphical representation of the MLC, such as the graphical MLC sequence 504. Using the predicted tensors, the AI ​​model may also generate what the MLC openings should look like (e.g., a binary mask of the MLC openings).

[0080] Although multi-task supervised learning is discussed herein, other embodiments may include the use of other machine learning methods, such as reinforcement learning, adversarial learning, unsupervised learning, etc. The training of the AI ​​model may be performed in an unsupervised manner. In an unsupervised learning approach, the relationship between the RTTP attributes and the MLC opening and / or control weights may not always be known to the AI ​​model (as opposed to a supervised learning approach where data points are labeled as ground truth).

[0081] In some configurations, the analysis server can pre-train or partially train the AI ​​model. For example, the analysis server can train the AI ​​model based on a set of cohort patients or clinical data. Then, the analysis server can train (fine-tune) the AI ​​model using the specific data of a specific patient or a specific clinic. For example, when the AI ​​model is pre-trained, the analysis server can fine-tune the AI ​​model by feeding the information of the patient or clinic and customize it to a specific patient (or a group of patients). This allows the AI ​​model to be customized without the risk of overfitting. For example, because the number of possibilities is high when generating RTTP, each clinic or clinician may have their own preferences and may solve the same problem in different ways. Therefore, the AI ​​model can be customized for different users. Using data associated with a specific clinic (such as RTTP and processing data associated with patients treated in a specific clinic) enables the AI ​​model to learn various attributes common in the treatments administered to patients by the clinic. Therefore, the AI ​​model can fine-tune its learning (and therefore its predicted results) for a specific clinic.

[0082] During training, the analysis server can iteratively generate new prediction results (e.g., projections) based on the training data set (e.g., for each patient and its corresponding data). If the prediction result does not match the true result, the analysis server continues training unless and until the computer-generated recommendation meets one or more accuracy thresholds and is within an acceptable range. For example, the analysis server can segment the training data set into three groups (i.e., training, validation, and testing). The analysis server can train the AI ​​model based on the first group (training). The analysis server can then execute (at least partially) the trained AI model to predict the results of the second set of data (validation). The analysis server then verifies whether the prediction is correct. Using the method described above, the analysis server can evaluate whether the AI ​​model is trained correctly. The analysis server can use this method to continuously train and improve the AI ​​model. The analysis server can then use the remaining data points in the training data set (test) to measure the accuracy of the AI ​​model (e.g., area under the curve, precision, and recall).

[0083] How AI models are implemented

[0084] Reference Figure 2C , the flowchart depicts the execution and results of the AI ​​model. As depicted, the AI ​​model 224 (when properly trained) can generate a tensor for the digital representation of the MLC order 226, a tensor for the graphical representation of the MLC order 228, and / or a tensor for the control point weights 230 when implemented. The AI ​​model can use a regression-based loss function and / or a Dice coefficient loss function to evaluate its output for a defined loss. For example, the AI ​​model 224 can use a regression-based loss function to generate blocks 226 and 230. The AI ​​model can use a Dice coefficient-based loss function to predict block 228. A non-limiting example of a loss function can be the difference between a predicted MLC opening (a tensor associated with the MLC opening) and an actual MLC opening (e.g., within a training data set). Another loss function can be defined as a limit on the MLC opening. The MLC opening limit can correspond to a physical or software limit on how to implement the MLC opening.

[0085] The two tensors related to the MLC sequence (blocks 226 and 228) can be combined together to generate a digital MLC sequence (block 232). The analysis server can optionally generate an RTTP file (DICOM plan file 234) including the results predicted by the AI ​​model 224. The file can include machine-readable instructions that can be used to populate a GUI and / or directly instruct the radiation therapy machine to change its configuration (e.g., to open MLCs based on the data predicted by the AI ​​model 224).

[0086] Return to reference Figure 2AAt 204, the analysis server may receive a treatment goal for the patient, the treatment goal including at least a 3D dose volume (e.g., a dose-volume goal for at least one structure of the patient to be treated via a radiation therapy machine). The analysis server may receive the treatment goal from a clinician. Using various methods, the analysis server may generate or retrieve a desired dose plan for different structures of the patient. For example, the analysis server may instruct a plan optimizer to generate a 3D dose plan structure (e.g., a DVH) for the patient.

[0087] The input to the AI ​​model (e.g., 3D dose volume / distribution) can be generated in a variety of ways. It can be generated using an MLC position-free optimization process based on using a DVH target as input, or it can also be generated using another AI algorithm that predicts the 3D dose volume / distribution.

[0088] In step 206, the analysis server may execute an artificial intelligence model to predict the MLC opening position of the radiotherapy machine (e.g., a digital representation and / or image or binary mask of the MLC opening) and corresponding movement properties of the accelerator, wherein the artificial intelligence model is trained via a training data set comprising training treatment goals and properties associated with previously performed radiotherapy treatments comprising at least an actual or projected 3D dose volume of a treated patient and a corresponding MLC opening image.

[0089] The analysis server may execute an artificial intelligence model that has been trained using the methods discussed herein. The AI ​​model may use the methods discussed herein to generate a tensor for a digital representation of an MLC sequence, a tensor for a graphical representation of an MLC sequence, and / or a tensor for control point weights. The AI ​​model may first analyze the received 3D dose volume and project the MLC openings and control point weights accordingly.

[0090] At step 208, the analysis server may present the predicted MLC opening positions of the radiotherapy machine and the corresponding predicted movement properties of the accelerator. The analysis server may populate the GUI with the results predicted by the AI ​​model. Figure 6 , presents a non-limiting example of a GUI presented by an analysis server. GUI 600 presents a moving image of a predicted sequence of MLC openings during a patient treatment. Thus, GUI 600 includes different frames (602, 604, and 606). Each frame corresponds to a particular timeframe or time of the patient treatment. For example, frame 602 corresponds to a 0 second timestamp (at the start of the treatment), frame 604 corresponds to the 15th second of the treatment, and frame 606 corresponds to the 30th second of the treatment.

[0091] Each frame may include projection beam views 602a, 604a, and 606a. Each projection beam view depicts a projection of a 2D MLC opening from a beam view, as described herein, for example, in Figure 3-Figure 4 Each frame may also include graphical representations of MLC openings 602b, 604b, and 606b. These MLC openings are predicted by the AI ​​model and are represented by a binary mask associated with the openings themselves, which is generated from the MLC tensor (graphical and / or digital) generated by the AI ​​model. The frames may optionally include graphical elements 602c, 604c, and 606c depicting a graphical representation (3D) of a structure receiving radiation.

[0092] Reference now Figure 7 , describes the numerical representation of the control point weights.

[0093] The digital representation of the control point weights can be a list of numbers. A complete arc in VMAT can be defined by 178 equally spaced control points to describe the machine rotation movement that delivers the arc. Therefore, the weights can be represented by a list of 178 numbers that sum to 1.0: the first number is usually the weight of the first control point, the second number is the weight of the second control point, etc. For example, the digital representation of the control points can be as follows:

[0094] [0.0000,0.0027,0.0053,0.0054,0.0057,……,0.0051,0.0025]

[0095] exist Figure 7 , digital representation 702 depicts digital representations of 40 pairs of MLCs from a particular control point. For each pair, the first number is the position of the leaf edge of the left MLC, and the second number is the position of the leaf edge of the right MLC.

[0096] In addition, the binary mask 700 can depict a graphical representation of the same MLCs from the same control point (e.g., white = open, black = closed), as digitally represented via digital representation 702. For example, the first 10 pairs depicted have 0s for each MLC, meaning that in each row, the MLCs touch each other at position 0, and therefore these MLCs are considered "closed" (e.g., black in the binary mask). For the 11th pair (e.g., control point), the left MLC ends at position 1.25mm, and the right MLC ends at position 9.38mm, leaving a ~8mm gap in the middle, which is this portion in the binary mask 700.

[0097] In another example, the analysis server can use data predicted by the AI ​​model to correct one or more properties of the patient's radiotherapy treatment. For example, the analysis server can use control weights or predicted MLC data to correct the properties of the MLC (e.g., MLC opening), move the treatment table (treatment bed), pause the beam, or any combination of these examples. Specifically, in combination with one or more other software solutions, the analysis server can correct the opening of the MLC so that the radiation propagation is directed to the projected position of the PTV (e.g., using the MLC opening properties and / or control point weights projected using the AI ​​model). In this way, the analysis server provides a dynamic MLC correction method in which the MLC opening can be corrected in real time or near real time. In fact, the analysis server can match the gating of the beam with the treatment target.

[0098] In another example, the analysis server can send the data predicted by the AI ​​model to a downstream software solution. For example, the execution results of the AI ​​model can be sent to a dose calculation software solution, such as a plan optimizer. The plan optimizer can use the data predicted by the AI ​​model to further analyze the RTTP.

[0099] In a non-limiting example, such as Figure 8 In the example 800 depicted in , the analysis server may receive a treatment goal 802 from a treating physician. The analysis server may execute a plan optimizer 804 to generate various dose projections 806 for the patient. The analysis server may then execute an AI model 808 using the dose projections 806 to generate predicted MLC openings and control point weights 810. The analysis server may optionally send the predicted MLC openings and control point weights 810 back to the plan optimizer 804 so that the plan optimizer 804 may generate a revised RTTP for the patient based on the data predicted by the AI ​​model 808 (step 812). In addition, the analysis server may optionally instruct a radiation therapy machine 814 to dynamically revise one or more of its configurations (e.g., MLC openings) based on the results predicted by the AI ​​model 808. In addition, the analysis server may present the results of the execution of the AI ​​model 808 on a user computing device 816.

[0100] Various illustrative logic blocks, modules, circuits and algorithmic steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software or a combination of the two. In order to clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits and steps have been generally described above according to their functions. Whether this function is implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. Those skilled in the art can implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as causing departure from the scope of the present disclosure or claims.

[0101] Embodiments implemented in computer software can be implemented in software, firmware, middleware, microcode, hardware description language or any combination thereof. Code segments or machine executable instructions can represent any combination of processes, functions, subroutines, programs, routines, subroutines, modules, software packages, classes or instructions, data structures or program statements. Code segments can be coupled to another code segment or hardware circuit by passing and / or receiving information, data, independent variables, parameters or memory contents. Information, independent variables, parameters, data, etc. can be transmitted, forwarded or sent via any suitable means, including memory sharing, message passing, token passing, network transmission, etc.

[0102] The actual software code or dedicated control hardware used to implement these systems and methods does not limit the claimed features or the present disclosure. Therefore, the operation and behavior of the systems and methods are described without reference to specific software code, and it should be understood that software and control hardware can be designed to implement the systems and methods based on the description herein.

[0103] When implemented in software, a function can be stored as one or more instructions or codes on a non-transient computer-readable or processor-readable storage medium. The steps of the method or algorithm disclosed herein can be embodied in a processor executable software module, which can reside on a computer-readable or processor-readable storage medium. Non-transient computer-readable or processor-readable media include computer storage media and tangible storage media that facilitate the transfer of computer programs from one place to another. Non-transient processor-readable storage media can be any available medium that can be accessed by a computer. As an example and not a limitation, such non-transient processor-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage devices, disk storage devices or other magnetic storage devices, or any other tangible storage medium that can be used to store desired program codes in the form of instructions or data structures and can be accessed by a computer or processor. As used herein, disks and optical disks include compact disks (CDs), laser optical disks, optical disks, digital versatile disks (DVDs), floppy disks, and Blu-ray disks, wherein disks generally reproduce data magnetically, and optical disks reproduce data optically with lasers. The above combination should also be included in the scope of computer-readable media. Additionally, the operations of a method or algorithm may reside as one or any combination or set of codes and / or instructions on a non-transitory processor-readable medium and / or computer-readable medium, which may be incorporated into a computer program product.

[0104] The foregoing description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the embodiments described herein and variations thereof. Various modifications to these embodiments will be apparent to those skilled in the art, and the principles defined herein may be applied to other embodiments without departing from the spirit or scope of the subject matter disclosed herein. Therefore, the present disclosure is not intended to be limited to the embodiments shown herein, but to be consistent with the widest scope consistent with the following claims and the principles and novel features disclosed herein.

[0105] While various aspects and embodiments have been disclosed, other aspects and embodiments are also contemplated. The various aspects and embodiments disclosed are for purposes of illustration and are not intended to be limiting, with the true scope and spirit being indicated by the following claims.

Claims

1. A method comprising: receiving, by a processor, a treatment target for a patient, the treatment target comprising at least a dose-volume for at least one structure of the patient to be treated via a radiation therapy machine; executing an artificial intelligence model by the processor to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set, the training data set comprising training treatment objectives and attributes associated with previously performed radiotherapy treatments, the training treatment objectives and attributes comprising at least actual or projected dose-volumes of treated patients and corresponding MLC opening locations; as well as Predicted MLC opening positions of the radiotherapy machine and corresponding predicted movement properties of the accelerator are presented by the processor.

2. The method of claim 1 , wherein the artificial intelligence model is trained via a deep learning protocol to correlate the actual or projected dose-volume of the treated patient and the corresponding MLC opening locations.

3. The method according to claim 1 or 2, wherein the predicted MLC opening position is a binary mask indicating the opening of the MLC.

4. A method according to claim 1, 2 or 3, wherein the predicted movement attribute is a time associated with movement of the accelerator.

5. The method of any one of claims 1 to 4, wherein the predicted movement attribute is an angle associated with movement of the accelerator.

6. The method according to any one of claims 1 to 5, wherein the artificial intelligence model also predicts the sequence of MLC openings in the patient.

7. The method of any one of claims 1 to 6, wherein the artificial intelligence model utilizes a loss function based on the MLC opening limit.

8. The method according to any one of claims 1 to 7, further comprising: Machine readable instructions are generated by the processor based on the predicted MLC openings and corresponding predicted movement attributes.

9. The method according to claim 8, further comprising: The machine-readable instructions are sent by the processor to the radiation therapy machine.

10. A computer system comprising: A server comprising a processor and a non-transitory computer-readable medium containing instructions that, when executed by the processor, cause the processor to perform operations comprising: receiving a treatment goal for a patient, the treatment goal comprising at least a dose-volume for at least one structure of the patient to be treated via a radiation therapy machine; executing an artificial intelligence model to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set, the training data set comprising training treatment objectives and attributes associated with previously performed radiotherapy treatments, the training treatment objectives and attributes comprising at least actual or projected dose-volumes of treated patients and corresponding MLC opening locations; as well as The predicted MLC opening positions of the radiotherapy machine and the corresponding predicted motion properties of the accelerator are presented.

11. The computer system of claim 10, wherein the artificial intelligence model is trained via a deep learning protocol to correlate the actual or projected dose-volume and corresponding MLC opening locations of the treated patient.

12. The computer system of claim 10 or 11, wherein the predicted MLC opening position is a binary mask indicating the opening of the MLC.

13. The computer system of claim 10, 11 or 12, wherein the predicted movement attribute is a time associated with movement of the accelerator.

14. The computer system of any one of claims 10 to 13, wherein the predicted movement attribute is an angle associated with movement of the accelerator.

15. The computer system of any one of claims 10 to 14, wherein the artificial intelligence model further predicts a sequence of MLC openings in the patient.

16. The computer system of any one of claims 10 to 15, wherein the artificial intelligence model utilizes a loss function based on an MLC opening limit.

17. The computer system of any one of claims 10 to 16, wherein the instructions further cause the processor to generate machine-readable instructions based on the predicted MLC openings and corresponding predicted move attributes.

18. The computer system of claim 17, wherein the instructions further cause the processor to send the machine-readable instructions to the radiation therapy machine.

19. A system comprising a server having one or more processors, the one or more processors being configured to: receiving a treatment goal for a patient, the treatment goal comprising at least a dose-volume for at least one structure of the patient to be treated via a radiation therapy machine; executing an artificial intelligence model to predict properties of a multi-leaf collimator (MLC) opening of the radiotherapy machine and corresponding movement properties of an accelerator, wherein the artificial intelligence model is trained via a training data set, the training data set comprising training treatment objectives and attributes associated with previously performed radiotherapy treatments, the training treatment objectives and attributes comprising at least actual or projected dose-volumes of treated patients and corresponding MLC opening locations; as well as The predicted MLC opening positions of the radiotherapy machine and the corresponding predicted motion properties of the accelerator are presented.

20. The computer system of claim 19, wherein the artificial intelligence model is trained via a deep learning protocol to correlate the actual or projected dose-volume and corresponding MLC opening locations of the treated patient.

Images