External artificial intelligence model for radiotherapy plan optimization

Abstract

Embodiments of the present disclosure relate to an external artificial intelligence model for radiotherapy plan optimization. Methods and systems for evaluating cost values for different radiation therapy plans using an external AI model are disclosed herein. The method includes: receiving a radiation therapy plan target for a patient; executing a plan optimizer to generate one or more treatment attributes for the treatment plan conforming to the radiotherapy plan goal, the plan optimizer iteratively calculating the one or more attributes, where for each iteration, the plan optimizer modifies the one or more attributes of the treatment plan according to the cost value; executing the AI model to calculate a second cost value for the treatment plan, wherein the AI model is trained to: calculate the second cost value as a function of a likelihood of a health problem occurring after the patient has been treated via the treatment plan having one or more attributes; and outputting the treatment plan for the patient.

Description

Technical Field

[0001] The present application generally relates to generating, training, and operating artificial intelligence (AI) computer models for loss function prediction for radiation therapy treatment planning. Background Art

[0002] Radiation therapy (radiation-based therapy or radiotherapy) is a cancer treatment in which high-dose radiation that can kill cancer cells or shrink tumors is applied to the affected area. The target area (e.g., tumor) of the patient's body that is intended to receive radiation is called a planned target volume (PTV). The goal is to deliver enough radiation to kill cancer cells to the PTV (also referred to herein as a treatment plan or radiotherapy treatment). However, other organs or anatomical regions adjacent to or surrounding the PTV may be in the path of the radiation beam and may receive enough radiation to be damaged or injured. These organs or anatomical regions are referred to as organs at risk (OARs). Typically, a physician or radiologist identifies PTV and OARs before radiotherapy by using, for example, computed tomography (CT) images, magnetic resonance imaging (MRI) images, positron emission tomography (PET) images, images obtained via other imaging modalities, or combinations thereof.

[0003] Using the patient's medical images and the identified PTV and OARs, a team of medical professionals (e.g., physicians, radiologists, oncologists, radiographers, other medical personnel, or a combination thereof) determines the radiation parameters to be used during the radiation therapy treatment. These radiation parameters may include, for example, the type, angle, intensity, and / or shape of each radiation beam. In determining these parameters, the team attempts to achieve a radiation dose distribution to the patient that meets predefined criteria (also referred to herein as a planning goal or clinical goal).

[0004] In order to optimize the radiation parameters in a way that meets predefined criteria, software solution(s) or computer model(s) (often referred to as a "plan optimizer") are utilized that can run multiple simulations with various radiation parameters and can select a final set of radiation parameters to use based on the simulation results. These radiation parameters are often referred to as a radiation therapy treatment plan (RTTP).

[0005] However, this process may be very inefficient because the plan optimizer usually uses an iterative trial and error process to determine the various attributes of the RTTP. For example, after the treating physician inputs the plan goal, the software solution iteratively analyzes different possibilities (for example, different iterations of different attributes in a large search space) to identify one or more attributes of the RTTP. Therefore, the software solution may require a large amount of computing resources and may not be able to produce results in time. In addition, various long-term effects may not be obvious to the plan optimizer, and / or the plan optimizer may not be able to consider long-term effects when generating RTTP. Summary of the invention

[0006] For the reasons stated above, it is desirable to have an external computer model that analyzes the long-term impact of RTTP using a method and system that does not rely on a static / internal cost function of the plan optimizer computer model. It is desirable to have an improved AI modeling / training technique that does not require extensive reconfiguration of the plan optimizer computer model and that is computationally efficient, cost-effective, and produces results in a timely manner.

[0007] RTTP creation is typically based on the optimization of certain dosimetric parameters, such as "target coverage", "maximum dose in the spinal cord" and / or "mean dose in the parotid gland". The dosimetric parameters used and the desired target values ​​are typically based on previously performed outcome studies in which there was a correlation between these dosimetric parameters and certain clinical endpoints. The observed correlation between dose distribution and clinical outcomes is sometimes referred to as the "clinical model". Conventional plan optimizers can optimize RTTP for a given clinical model by defining the desired targets, but they inevitably use certain generic dosimetric parameters that are implemented in the software algorithms during product development and are sometimes hard-coded in the software algorithms (such as "dose-volume", "volume-dose" and / or "generalized equivalent uniform dose"). This static approach is undesirable because modifying the software code is time-consuming and computationally intensive.

[0008] As the pace of innovation in the field of radiotherapy planning accelerates, it is expected that more complex spatial prediction models will be created in the near future. For example, imaging technology is constantly improving to the point where it may be possible to create new types of spatially functional images. In addition, the current rapid development of machine learning techniques (such as neural networks) may allow for the creation of more accurate spatial prediction models. Therefore, implementing conventional static methods (e.g., requiring new dosimetry parameters to be implemented as part of the software code to support new prediction models) will become increasingly computationally expensive.

[0009] In one embodiment, the method may include: receiving, by a processor, a radiation therapy plan goal for a patient; executing, by the processor, a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy plan goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; executing, by the processor, an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated with the treatment plan having the one or more attributes; and outputting, by the processor, a treatment plan for the patient.

[0010] The method may also include sending, by the processor, the second cost value to the plan optimizer computer model, wherein the plan optimizer computer model modifies one or more properties of the treatment plan based on the second cost value.

[0011] The health problem may correspond to at least one of the following: dry mouth, decreased saliva secretion, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia.

[0012] Radiation therapy planning goals may correspond to dose volume goals.

[0013] Health problems can correspond to the development of secondary cancers.

[0014] The plan optimizer computer model may aggregate the second cost value generated by the artificial intelligence model with the cost value generated by the plan optimizer computer model.

[0015] AI models can be customized for use in clinics implementing treatment plans tailored to the patient.

[0016] In another embodiment, a non-transitory machine-readable storage medium has computer executable instructions stored thereon, which, when executed by one or more processors, can cause the one or more processors to perform operations including: receiving a radiation therapy plan goal for a patient; executing a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy plan goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; executing an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to: calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated via a treatment plan having one or more attributes; and outputting a treatment plan for the patient.

[0017] The instructions may further cause the one or more processors to: send the second cost value to the plan optimizer computer model, wherein the plan optimizer computer model modifies one or more properties of the treatment plan based on the second cost value.

[0018] The health problem may correspond to at least one of the following: dry mouth, decreased saliva secretion, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia.

[0019] Radiation therapy planning goals may correspond to dose volume goals.

[0020] Health problems can correspond to the development of secondary cancers.

[0021] The plan optimizer computer model may aggregate the second cost value generated by the artificial intelligence model with the cost value generated by the plan optimizer computer model.

[0022] AI models can be customized for use in clinics implementing treatment plans tailored to the patient.

[0023] In another embodiment, a system may include at least one processor configured to: receive a radiation therapy plan goal for a patient; execute a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy plan goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; execute an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to: calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated via a treatment plan having the one or more attributes; and output a treatment plan for the patient.

[0024] The at least one processor may be further configured to send the second cost value to the plan optimizer computer model, wherein the plan optimizer computer model modifies one or more properties of the treatment plan based on the second cost value.

[0025] The health problem may correspond to at least one of the following: dry mouth, decreased saliva secretion, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia.

[0026] Radiation therapy planning goals may correspond to dose volume goals.

[0027] Health problems can correspond to the development of secondary cancers.

[0028] The plan optimizer computer model may aggregate the second cost value generated by the artificial intelligence model with the cost value generated by the plan optimizer computer model.

[0029] AI models can be customized for use in clinics that implement treatment plans tailored to the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Non-limiting embodiments of the present disclosure are described by way of example with reference to the accompanying drawings, which are schematic and are not intended to be drawn to scale. Unless indicated as representing background art, these drawings represent aspects of the present disclosure.

[0031] Figure 1 Illustrated are components of an AI-enabled RTTP generation system according to one embodiment.

[0032] Figure 2 A flow diagram of a process performed in an AI-enabled RTTP generation system according to one embodiment is illustrated.

[0033] Figure 3 Embodiments of calculating cost values ​​according to various embodiments are illustrated.

[0034] Figure 4 Illustrated are different flow diagrams of different processes performed in an AI-enabled RTTP generation system according to one embodiment. DETAILED DESCRIPTION

[0035] Reference will now be made to the illustrative embodiments depicted in the accompanying drawings, and specific language will be used herein to describe them. 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 described herein and additional applications of the principles of the subject matter described herein (which are conceivable to those skilled in the relevant art and those who possess the disclosure) should be considered to be 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 description are not meant to limit the subject matter presented.

[0036] Figure 1 Components of an AI-enabled RTTP generation system (system 100) according to one embodiment are illustrated. System 100 may include an analysis server 110a, a system database 110b, a clinical model analysis computer model 111, end-user devices 120a-120d (collectively referred to as end-user devices 120), and an administrator computing device 150. Figure 1The various components depicted in the figure may belong to a radiation therapy clinic where patients may receive radiation therapy treatment via one or more radiation therapy machines (e.g., medical devices 140) in the clinic. The above components may be connected via a network 130. Examples of network 130 may include, but are not limited to, private or public LANs, WLANs, MANs, WANs, and the Internet. Network 130 may employ wired and / or wireless communications according to one or more standards and / or via one or more transmission media.

[0037] Communications on 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 the Bluetooth specification set or another standard or proprietary wireless communication protocol. In another example, network 130 may also include communications over a cellular network, such as a GSM (Global System for Mobile Communications), CDMA (Code Division Multiple Access), or EDGE (Enhanced Data for Global Evolution) network.

[0038] 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, which should be considered within the scope of the embodiments described herein.

[0039] The analysis server 110a can generate and display an electronic platform configured to use various computer models 111-112 to identify the properties of the RTTP used for patient treatment. The electronic platform can include a graphical user interface (GUI) displayed on the end-user device 120 and / or the administrator computing device 150. An example of an electronic platform generated and hosted by the analysis server 110a can be a web-based application or website that is configured to be displayed on different types of electronic devices, such as mobile devices, tablet computers, personal computers, etc.

[0040] In a non-limiting example, a physician operating the physician device 120b can access the platform, input patient attributes, characteristics and other data, and further instruct the analysis server 110a to generate RTTP (including beam angle). Additionally or alternatively, the analysis server 110a can optimize only one specific attribute of the RTTP (e.g., only optimize the beam angle).

[0041] The analysis server 110a can recommend RTTP (e.g., beam angles and other radiation parameters and / or treatment plan attributes) for proton radiation, photon radiation, and electron radiation. Specifically, the analysis server 110a can utilize the methods and systems described herein to execute a clinical model analysis computer 111 to identify cost values ​​that can ultimately be used to evaluate and re-evaluate the RTTP (e.g., by a plan optimizer model 112). In addition, the analysis server 110a can send the beam angles and other radiation parameters and / or treatment plan attributes to one or more other servers. Additionally or alternatively, the analysis server 110a (or another server) can adjust the configuration of one or more devices (e.g., a medical device 140) based on the optimized beam angle. For example, the RTTP can be sent directly to the medical device 140 and / or to a computer 142 that functionally controls the medical device 140.

[0042] The medical device 140 may be any medical device used to perform radiotherapy treatment on a patient (such as a CT scanning machine, a radiotherapy machine (e.g., a linear accelerator, a particle accelerator (including a circular accelerator), or a cobalt machine)). The medical device 140 may also include an imaging device capable of emitting radiation so that the medical device 140 can obtain images according to various methods to accurately image the internal anatomical structure of the patient. For example, the medical device 140 may include a rotating system (e.g., a static or rotating multi-view system). A non-limiting example of a multi-view system may include a stereoscopic system (e.g., two systems arranged orthogonally).

[0043] The analysis server 110a can host a website accessible by a user (e.g., an end user or a medical professional) operating any electronic device described herein, where the content presented via various web pages can be controlled based on the role or viewing permissions of each particular user. 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 workstation computers, notebook computers, server computers, 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).

[0044] The analysis server 110a can execute a software application configured to display an electronic platform (e.g., a hosted website) that can generate various web pages and provide various web pages to each end-user device 120. Different users can use the website to view and / or interact with the recommended (optimized) results. Different servers (such as the clinic server 120c) can also use the recommended results in downstream processing.

[0045] The analysis server 110a may be configured to require user authentication based on a set of user authorization credentials (e.g., username, password, biometric indicator, encryption certificate, etc.). The analysis server 110a may access a system database 110b configured to store user credentials, which the analysis server 110a may be configured to reference to determine whether the input set of credentials (which allegedly identify the user) matches an appropriate set of credentials for authenticating the user.

[0046] The analysis server 110a can generate and host web pages based on the role of a particular user within the system 100. In such an implementation, the role of the user can be defined by data fields and input fields in a user record stored in the system database 110b. The analysis server 110a can authenticate the user by executing an access directory protocol (e.g., LDAP) and can identify the user's role. The analysis server 110a can generate web page content customized according to the user's role (as defined by the user record in the system database 110b).

[0047] The analysis server 110a can receive various clinical objectives, patient data, and treatment data from the end-user device 120. For example, a physician can use the physician device 120b to access the platform provided by the analysis server 110a. The physician can use one or more input elements of the platform to enter various patient attributes and / or clinical objectives. The analysis server 110a can then perform the various methods discussed herein and display the RTTP on the platform.

[0048] The end-user device 120 may 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 120 may be a workstation computer, a notebook computer, a tablet computer, and a server computer. During operation, various users may use the end-user device 120 to access a GUI operated by the analysis server 110a. Specifically, the end-user device 120 may include a clinic computer 120a, a physician device 120b, a clinic server 120c, or a clinic database 120d.

[0049] To generate the RTTP, the analysis server 110a may then execute various models (eg, the clinical model analysis computer model 111 and / or the plan optimizer model 112) to analyze the retrieved / received data.

[0050] Administrator computing device 150 may represent a computing device operated by a system administrator. Administrator computing device 150 may be configured to: display beam angles, radiation parameters, and / or other radiation therapy treatment properties generated by analysis server 110a; monitor clinical model analysis computer models 111 utilized by analysis server 110a and / or end-user devices 120; review feedback; and / or facilitate training or retraining (calibration) of models 111-112 maintained by analysis server 110a.

[0051] The analysis server 110a can communicate (in real time or near real time) with the medical device 140 (or its computer 142) so that the server / computer hosting the medical device 140 can adjust the medical device 140 based on the RTTP (e.g., the beam angle, treatment properties, and / or radiation parameters determined by the analysis server 110a). For example, a radiation therapy machine can adjust the gantry, beam blocking device (e.g., multi-leaf collimator MLC), and couch based on an optimized beam angle, where the optimized beam angle is the angle at which the medical device 140 emits radiation in the direction of the PTV. The analysis server 110a can send instructions to the radiation therapy machine that indicate any number or type of radiation parameters and / or treatment properties to facilitate such adjustments.

[0052] The clinical model analysis computer model 111 and / or 112 can represent any set of algorithmic logic and / or artificial intelligence models (e.g., using various machine learning techniques). For example, the clinical model analysis computer model 111 can be used to determine various algorithms for the cost value of RTTP based on patient data and / or treatment attributes and how they may cause secondary and / or long-term effects (e.g., health problems). Using the clinical model analysis computer model 111, the analysis server 110a can execute the plan optimizer model 112. For example, the analysis server 110a can execute the model in conjunction with each other, and then modify one or more configuration parameters of the plan optimizer 112. Therefore, the efficiency discussed herein can be achieved without modifying the plan optimizer 112 itself. For example, the clinical model analysis computer model 111 can be executed independently and used in combination with an existing plan optimizer. Therefore, the method and system discussed herein can be used to transform the plan optimizer (or system infrastructure). This minimal interference to the existing infrastructure allows the improvement of the plan optimizer without modifying its source code.

[0053] In some embodiments, the analysis server 110a may Figure 1Patient data may be collected from various sources not shown in the figure. For example, the analysis server 110a may monitor and collect patient and treatment attributes associated with previously treated radiation therapy patients to include in the training data set. Additionally or alternatively, the analysis server 110a may collect / aggregate other patient data (e.g., data associated with how a collection of patients or clinical participants received their respective radiation therapy treatments). Using the collected data, the analysis server 110a may generate a training data set, which the analysis server 110a may utilize to train the clinical model analysis computer model 111. Thus, the clinical model analysis computer model 111 may be used to limit the search space used by the plan optimizer 112.

[0054] Figure 2 A flow chart of a process performed in an AI-enabled RTTP generation system according to one embodiment is illustrated. Method 200 includes steps 210-260. However, other embodiments may include additional or alternative execution steps, or may omit one or more steps entirely. Method 200 describes how an AI model may be trained and then executed so that it can predict the cost value of RTTP.

[0055] One or more steps of method 200 may be performed by Figure 1 The method 200 may be executed by any number of computing devices present in the distributed computing system described in . For example, a processor different from the processor executing the artificial intelligence model (or even a third-party processor) may train the same artificial intelligence model. In addition, the artificial intelligence model may be executed together with a plan optimizer executed for and / or by a third party. Therefore, even though the method 200 is described as being executed / implemented by an analysis server, each step of the method 200 may also be executed for a different entity / processor.

[0056] At step 210, the analysis server may receive radiation therapy planning goals for the patient. The analysis server may receive various attributes determined by a treating physician or clinician for the radiation therapy treatment of the patient. These attributes may be determined for the patient based on the patient's treatment and are generally referred to as planning goals. Non-limiting examples of planning goals may include dose constraints, in which the treating physician identifies a dose that needs to be applied to the tumor.

[0057] The analysis server can receive and / or retrieve (via the platform and / or via querying a database) patient data required to calculate a treatment plan for the patient. Therefore, as used herein, patient data can refer to any data associated with a patient (e.g., the patient's physical data, diagnostic data, and any data entered by a medical professional). In some embodiments, the analysis server can retrieve patient data using different patient identifiers. For example, the analysis server can query one or more databases to identify medical data associated with a patient. The analysis server can query data associated with the patient's anatomical structure, such as physical data (e.g., height, weight, and / or body mass index) and / or other health-related data (e.g., blood pressure or other data related to receiving radiation therapy treatment). The analysis server can also retrieve data associated with current and / or previous medical treatments received by the patient (e.g., data associated with a patient's previous surgery).

[0058] In some embodiments, the analysis server may also analyze the patient's medical data records to identify desired patient attributes. For example, the analysis server may query a database to identify the patient's BMI. However, because many medical records are not digitized, the analysis server may not be able to receive the patient's BMI value using simple query techniques. Therefore, the analysis server may retrieve the patient's electronic health data, and may execute one or more analysis protocols (e.g., natural language processing) to identify the patient's body mass index. In another example, if the analysis server does not receive tumor data (e.g., endpoint), the analysis server may execute various image recognition protocols and identify the tumor data.

[0059] The analysis server may also receive additional data from one or more medical professionals. For example, a treating oncologist may access a platform generated / hosted by the analysis server and may add, delete, or modify data associated with a particular patient, such as patient attributes, treatment attributes, tumor attributes, primary site of treatment, tumor stage, endpoint, whether the primary tumor has spread, etc. Since tumor staging and endpoint-level attributes are key pieces of information that affect patient treatment, this information is typically entered by the treating oncologist. In some embodiments, an AI model (e.g., a separate AI model trained to identify tumor information) may identify this information, and the treating oncologist may reject, approve, or modify the predicted results. In another example, the treating oncologist may specifically indicate whether the treatment should be unilateral or bilateral.

[0060] Another example of a planning goal and / or patient data / attribute may include a specific tumor location. More specifically, the data may indicate the location of the primary tumor relative to the centerline of the patient. This data may be entered by a treating oncologist, or may be analyzed using various image recognition or segmentation methods performed on medical images of the patient. In some embodiments, if the treating oncologist does not enter this information (or the analysis server does not otherwise receive this information), the AI ​​model may also be used to predict this information. Another patient attribute may indicate whether the tumor is adjacent to other non-diseased organs and how far away they are. For example, a tumor to be eradicated may be a few millimeters away from another organ. Since other organs must be avoided, this information may change the field geometry (generated by the optimizer).

[0061] At step 220, the analysis server may execute a plan optimizer computer model to generate one or more treatment attributes of a treatment plan that meets the radiation therapy planning goals, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on the cost value.

[0062] The analysis server can deploy a plan optimizer computer model to generate an RTTP for a patient based on patient data. The plan optimizer can iteratively calculate one or more properties of the RTTP. For example, for each iteration, the plan optimizer computer model can generate a candidate RTTP with various properties. Then, the plan optimizer computer model can use one or more loss functions to calculate the cost value for the generated candidate RTTP. The cost value can indicate the possibility that the candidate RTTP violates a set of rules, whether it is an internal rule or an external rule. For example, the cost value can indicate whether the candidate RTTP violates any of the plan goals (for example, received in step 210) of the plan goals. Then, the plan optimizer computer model can analyze the cost value. If necessary, for example, when the cost value meets a threshold, the plan optimizer computer model can modify the candidate RTTP, and re-execute its loss function to generate a new cost value. According to whether the new cost function increases or decreases, the plan optimizer computer model can modify the candidate RTTP again and recalculate the cost value. The plan optimizer computer model can continue the iterative method until it converges according to the RTTP (or final RTTP) with a cost value that meets the threshold.

[0063] At step 230, the analysis server may execute an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated via a treatment plan having one or more attributes.

[0064] The artificial intelligence model may include a neural network (NN) trained to predict certain clinical outcomes. Specifically, the artificial intelligence model may determine a cost value for the RTTP (in addition to the cost value generated by the plan optimizer computer model) that indicates the likelihood that a patient will develop a health problem (e.g., a side effect) from being treated using the attributes within the RTTP generated by the plan optimizer computer model. In some embodiments, the neural network architecture and weights may be loaded as a file by the plan optimizer computer model, so it may not need to be part of the software release or part of the plan optimizer computer model itself. Therefore, the methods and systems discussed herein allow an independent artificial intelligence model to be implemented in conjunction with the plan optimizer computer model without requiring reconfiguration of two separate models.

[0065] The method allows the AI ​​model to improve results as a retroactive solution that requires minimal technical knowledge. In addition, the methods and systems discussed herein allow the AI ​​model to be plan optimizer agnostic, wherein the AI ​​model is compatible with any plan optimizer computer model as long as it is able to ingest the results output / predicted by the plan optimizer computer model. For example, the AI ​​model can generate a file or other data format type that enables the evaluation of the plan optimizer computer model to be completed in a common software component.

[0066] The artificial intelligence model can predict the cost value of the advanced utility corresponding to certain predicted probabilities, such as the probability of one or more side effects of other medical problems. The cost value can correspond to the possibility that RTTP causes health-related problems, such as the long-term side effects or other health problems discussed herein. As used herein, the second cost value or the cost value calculated by the artificial intelligence model can refer to any data associated with the cost value used by the plan optimizer. For example, in some embodiments, the artificial intelligence model can calculate the actual cost value, which can be used directly by the plan optimizer computer model. Additionally or alternatively, the artificial intelligence model can generate one or more scores, which can then be used to calculate the cost value. For example, the value predicted by the artificial intelligence model can be converted to a cost value (e.g., using other algorithms). Therefore, the artificial intelligence model can be trained to predict one or more attributes, which can then be converted to a cost value used by the plan optimizer. Conventional functions can be used to convert / transform the value predicted by the artificial intelligence model into a cost value. In a non-limiting example, the cost value (second cost) predicted by the artificial intelligence model can be the possibility of a headache, which can then be used to calculate the cost function used by the plan optimizer.

[0067] The inputs to the AI ​​model may include a set of channels that describe different (independent) features of the final RTTP and / or candidate RTTPs generated by the plan optimizer computer model. Non-limiting examples of inputs to the AI ​​model may include: a dose distribution (the dose distribution of the current candidate RTTP, some other similar information (such as "2Gy equivalent dose" or "proton-LET"), and / or some dose rate corrected dose distribution that attempts to capture ultra-high dose rate (Flash) effects), a target or organ at risk mask, and / or a CT or other imaging modality.

[0068] In some embodiments, when the artificial intelligence model is used as part of the cost function for the plan optimizer computer model, the user may be asked to describe which channels are / should be used as inputs. In some embodiments, the channels may be automatically inferred by the plan optimizer computer model and / or the artificial intelligence model based on the type of optimizer data object.

[0069] In some embodiments, the plan optimizer may also send the functional image to the artificial intelligence model (as an input channel). In some embodiments, the optimizer computer model may not be aware of the specific imaging modality, and the interpretation of the imaging modality may rely solely on the specified artificial intelligence model. In addition, the input channel may need to be at a certain spatial resolution and size. Therefore, the model discussed herein may include a component that checks whether the dimensionality and resolution of the input data (such as dose) are set correctly so that the image can be ingested by the artificial intelligence model.

[0070] Using the methods and systems discussed herein allows the user-specified AI model to become part of the cost function and further allows support for new types of spatial prediction models without the need to release new optimizer algorithms. It also allows the user to use any other spatial information to support spatial clinical model evaluation.

[0071] Before the artificial intelligence model is executed to calculate the cost value, the analysis server can train the artificial intelligence using a training data set including training data, including previously treated patients and / or patients who have participated in one or more clinical trials. The analysis server can monitor and record health-related problems of treated patients (after their treatment). For example, patients can be evaluated for set schemes regarding side effects and / or known health-related problems. Non-limiting examples of these health-related problems can include: dry mouth; headache; hair loss; nausea; vomiting; extreme tiredness (fatigue); hearing loss or impairment; skin and / or scalp problems (changes), memory and / or speech disorders, skin pain, diarrhea, fertility problems, etc. When health-related problems are identified, a manual auditor (or sometimes a processor that uses preset rules) can quantify the health-related problems so that the quantified health-related problems and their corresponding categories can be included in the training data set. In addition, health-related problems can include the development of secondary cancers. The training data set can also include a timeline associated with each health-related problem.

[0072] The training data set may also include data associated with the patient and his or her specific RTTP. For example, for each patient, the training data set may include: patient data (e.g., all available data associated with the patient's anatomy, disease, diagnosis, and any other data ingested by the plan optimizer while the patient is being treated), any available data entered by the patient's treating physician, and / or the patient's RTTP and its attributes. That is, the training data set may include: information about each patient's anatomy and disease; how the patient was diagnosed; how the patient was treated; and how the patient experienced health-related issues and corresponding timelines.

[0073] Using the training data set, the analysis server can train an artificial intelligence model so that the artificial intelligence model can ingest data associated with new patients and the new patients' RTTP to predict the likelihood of health-related problems. Training can be performed using supervised, unsupervised, or semi-supervised approaches.

[0074] When training is complete, the artificial intelligence model can be used to generate new cost values ​​for the RTTP. Specifically, when the plan optimizer generates an RTTP or a candidate RTTP, the artificial intelligence model can ingest the RTTP, patient information, and other relevant data to predict the likelihood that the patient will develop one or more health-related problems if the patient is treated with the RTTP or candidate RTTP.

[0075] In some embodiments, the training of the artificial intelligence model can be specific to a clinic or a specific region. For example, certain clinics may have specific rules and tolerances that are unique to them. Therefore, the analysis server can customize the artificial intelligence model to predict cost values ​​that meet the clinic's specific requirements, tolerances, and rules. In addition, certain artificial intelligence models or sub-models can be customized for specific health-related issues. For example, one sub-model can be trained to predict the likelihood of secondary cancers, and another sub-model can be trained to predict the likelihood of developing dry mouth.

[0076] In some embodiments, the prediction of cost values ​​may be bifurcated between conventional / traditional calculations and the use of the artificial intelligence computer models discussed herein. The analysis server may then aggregate the cost values ​​generated using the two different methods. For example, to calculate the cost function, the artificial intelligence may use Figure 3 As shown, the cost function can be bifurcated into traditional cost calculation 310 and artificial intelligence model dependency 320 (e.g., C NN (M NN (D(F), P))), the traditional cost calculation 310 is evaluated entirely as part of the optimizer algorithm used by the planning optimizer computer model.

[0077] In calculating the cost value, the AI ​​model is assumed to depend on the degrees of freedom (F). This value can be optimized according to the dose matrix (D) to improve the RTTP. Therefore, as illustrated, the traditional cost calculation can emphasize how well the RTTP performs in terms of optimization and not violating the dose volume target. Non-limiting examples can include flux pixel values ​​and / or machine control points.

[0078] In the above formula, P can represent an organ mask or a functional image. The metrics provided by the AI ​​model can then be sent back to the high-level utility function The function contains parameters such as a target level for that metric or describing its importance relative to other plan quality metrics. Not all variables can be treated equally when generating RTTP. Therefore, some attributes or goals can have higher weights than others. For example, a dose target (and its corresponding negative impact of a possible violation) can be considered more important (have a higher weight) than a patient losing hair (as a side effect).

[0079] Return to reference Figure 2 At step 250, the analysis server may optionally send the cost value predicted by the artificial intelligence model back to the plan optimizer computer model so that the plan optimizer computer model can modify the RTTP or candidate RTTP accordingly.

[0080] At step 260, the analysis server may output a treatment plan for the patient. When the RTTP is finalized, the analysis server may output (e.g., display on a platform) the RTTP and its various attributes for a user (e.g., a clinician or a treating physician). In some embodiments, the analysis server may directly instruct the radiation therapy machine to change one or more configurations according to the finalized RTTP.

[0081] In a non-limiting example, the treating physician uses one or more input elements received via the electronic platform to input the patient's diagnosis and plan goals. Then, the analysis server can receive a request to generate an RTTP for the patient using the plan optimizer computer model. The analysis server can further retrieve patient information and execute the plan optimizer computer model to generate any other data required for the RTTP. Therefore, the plan optimizer computer model ingests aggregated data and generates candidate RTTPs that meet various cost functions inside the plan optimizer. RTTP and patient information are then ingested by the artificial intelligence model, where the artificial intelligence model predicts that the patient may (60% probability) develop secondary cancer within 6-8 months after treatment due to the RTTP treatment. The predicted cost value / likelihood is then ingested back by the plan optimizer computer model, where the RTTP is further modified and a second RTTP is generated. The second RTTP is then ingested by the artificial intelligence model, and the artificial intelligence model determines that the probability of secondary cancer is reduced to 5%, and 5% is then identified as being within a tolerable threshold. Therefore, the analysis server sends the second RTTP to the platform and presents the second RTTP to the clinician / therapeutic physician.

[0082] It should be understood that when the AI ​​predicted attribute (cost) is ingested back into the plan optimizer computer model, the cost value can be one or more factors considered by the plan optimizer. Therefore, the cost value can be among many other optimized attributes when generating the second RTTP. Therefore, the second RTTP depends on the manner and content (tradeoffs) of the solution implemented by the plan optimizer. For example, in some embodiments, the cost value predicted by the artificial intelligence model can be ignored (e.g., when the cost value of the artificial intelligence model indicates a high likelihood of a headache). In some embodiments, the cost value (e.g., when it indicates a high likelihood of a secondary cancer) can have a high impact on the second RTTP.

[0083] Reference now Figure 4, a non-limiting visual example of a workflow utilizing the methods and systems described herein is illustrated. In this non-limiting example 400, the analysis server provides the plan goals 410c, the patient anatomy 410a, the user input 410b, and other data that may be required to generate the RTTP (collectively referred to as the patient data 410) to the plan optimizer 430 to generate a recommended RTTP optimized for the patient and his treatment. The analysis server can also send the data 410 to the artificial intelligence computer model 420. Therefore, the artificial intelligence computer model 420 can send the cost function to the plan optimizer 430, where the two models can work together iteratively to generate a recommended treatment plan 440.

[0084] The analysis server may first collect patient data 410. The patient data may include patient anatomical data 410a (e.g., medical images, PTV, OARs) and user input 410b (clinical goals or rules received from a treating oncologist via a user interface, such as tumor data, PTV identification, etc.). Other non-limiting examples of clinical goals may include: dosimetry goodness functions, robustness metrics, biological effects of radiation, linear energy transfer metrics, etc. The patient data 410 may also include rules 410c for patient treatment (e.g., clinical / treatment goals or criteria identified by a medical professional, or any other special treatment requested by a medical professional).

[0085] In some configurations, the analysis server may access the patient's internal / external files and retrieve / extract the required patient data 410. The analysis server may then execute an artificial intelligence computer model 420 to identify a cost function for a candidate RTTP for the patient using the patient data 410. The results generated via the artificial intelligence computer model 420 may be ingested by a plan optimizer 430. The plan optimizer 430 may be a treatment planning and / or monitoring software solution. The plan optimizer 430 may analyze various factors associated with the patient and the patient's treatment to generate and optimize the RTTP for the patient (e.g., field geometry, treatment modality, and radiation parameters required to treat the patient).

[0086] One of the factors considered by the plan optimizer 430 may be a cost value calculated by the artificial intelligence computer model 420. The plan optimizer 430 may iteratively modify the patient's RTTP, wherein the plan optimizer 430 iteratively modifies different properties of the RTTP (e.g., field geometry). In some configurations, the plan optimizer 430 may send new treatment plan data back to the artificial intelligence model 420, whereby the dose calculation sector model 420 may recalculate / re-predict the cost value based on the modified treatment data generated by the plan optimizer (iteration 422). The plan optimizer 430 and the artificial intelligence model 420 may repeat iteration 422 until the patient's RTTP is optimized / finalized.

[0087] When the plan optimizer 430 completes the RTTP for the patient, the plan optimizer 430 can send a suggested treatment plan 440 to one or more electronic devices where a user (e.g., a clinician) can review the suggested plan. For example, the suggested treatment plan 440 (or any attributes predicted by the dose calculation sector model 420) can be displayed on a computer in the clinic where a radiation therapy technician or treating oncologist can review the data on the computer.

[0088] Various illustrative logic blocks, modules, circuits and algorithm 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 the interchangeability of hardware and software, various illustrative components, blocks, modules, circuits and steps have been summarized and described above in terms of their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person can implement the described functions in different ways for each specific application, but this implementation decision should not be interpreted as causing a deviation from the scope of the present disclosure or claims.

[0089] Embodiments implemented with computer software can be implemented with software, firmware, middleware, microcode, hardware description language or any combination thereof. Code segments or machine executable instructions can represent procedures, functions, subroutines, programs, routines, subroutines, modules, software packages, classes, or any combination of instructions, data structures or program statements. Code segments can be coupled to another code segment or hardware circuit by transmitting and / or receiving information, data, parameters, parameters or memory contents. Information, parameters, parameters, data, etc. can be transmitted, forwarded or transmitted via any appropriate means (including memory sharing, message passing, token passing, network transmission, etc.).

[0090] 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, when describing the operation and behavior of the systems and methods without reference to specific software code, it should be understood that software and control hardware can be designed to implement the systems and methods based on the description herein.

[0091] When implemented in software, the function can be stored as one or more instructions or codes on a non-transient computer-readable storage medium or a 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 storage medium or a processor-readable storage medium. Non-transient computer-readable storage media or processor-readable media include both computer storage media and tangible storage media, which help to transfer 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 limitation, such non-transient processor-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, or any other tangible storage media, which can be used to store the required program code in the form of instructions or data structures, and can be accessed by a computer or processor. The disks and optical disks used herein include compact disks (CDs), laser optical disks, optical disks, digital versatile disks (DVDs), floppy disks, and blue optical disks, wherein the disks usually reproduce data magnetically, and the optical disks reproduce data optically using lasers. The above combination should also be included in the scope of computer-readable media. Furthermore, the operations of a method or algorithm may reside as one or any combination of a 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.

[0092] The previous description of the disclosed embodiments is intended to enable those skilled in the art to make or use the embodiments described herein and variations thereof. Those skilled in the art will readily appreciate various modifications to these embodiments, 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 should be consistent with the widest scope consistent with the appended claims and the principles and novel features disclosed herein.

[0093] While various aspects and embodiments have been disclosed, other aspects and embodiments are also contemplated. The disclosed various aspects and embodiments are for purposes of illustration only 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, radiation therapy planning objectives for a patient; executing, by the processor, a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy planning goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; executing, by the processor, an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to: calculate the second cost value based on a likelihood that the patient will develop a health problem after being treated via the treatment plan having the one or more attributes; as well as The treatment plan for the patient is output by the processor.

2. The method according to claim 1, further comprising: The second cost value is sent, by the processor, to the plan optimizer computer model, wherein the plan optimizer computer model modifies the one or more attributes of the treatment plan based on the second cost value.

3. The method of claim 1, wherein the health problem corresponds to at least one of: dry mouth, decreased salivation, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia. The method of claim 1 , wherein the radiation therapy planning goal corresponds to a dose volume goal.

5. The method of claim 1, wherein the health problem corresponds to developing a secondary cancer.

6. The method of claim 1, wherein the plan optimizer computer model aggregates the second cost value generated by the artificial intelligence model and the cost value generated by the plan optimizer computer model.

7. The method of claim 1, wherein the artificial intelligence model is customized for a clinic implementing the treatment plan for the patient.

8. A non-transitory machine-readable storage medium having computer-executable instructions stored thereon, the instructions, when executed by one or more processors, causing the one or more processors to perform operations comprising: Receive radiation therapy plan goals for the patient; executing a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy planning goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; executing an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to: calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated via the treatment plan having the one or more attributes; as well as The treatment plan for the patient is output.

9. The non-transitory machine-readable storage medium of claim 8, wherein the instructions further cause the one or more processors to: send the second cost value to the plan optimizer computer model, wherein the plan optimizer computer model modifies the one or more attributes of the treatment plan based on the second cost value.

10. The non-transitory machine-readable storage medium of claim 8, wherein the health problem corresponds to at least one of: dry mouth, decreased salivation, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia.

11. The non-transitory machine-readable storage medium of claim 8, wherein the radiation therapy planning goal corresponds to a dose volume goal.

12. The non-transitory machine-readable storage medium of claim 8, wherein the health problem corresponds to developing a secondary cancer.

13. The non-transitory machine-readable storage medium of claim 8, wherein the plan optimizer computer model aggregates the second cost value generated by the artificial intelligence model and the cost value generated by the plan optimizer computer model.

14. The non-transitory machine-readable storage medium of claim 8, wherein the artificial intelligence model is customized for a clinic implementing the treatment plan for the patient.

15. A system comprising at least one processor, the at least one processor being configured to: Receive radiation therapy plan goals for the patient; executing a plan optimizer computer model to generate one or more treatment attributes for a treatment plan that meets the radiation therapy planning goal, the plan optimizer computer model iteratively calculating the one or more attributes, wherein for each iteration, the plan optimizer computer model modifies the one or more attributes of the treatment plan based on a cost value; executing an artificial intelligence model to calculate a second cost value for the treatment plan, wherein the artificial intelligence model is trained to: calculate the second cost value based on the likelihood that the patient will develop a health problem after being treated via the treatment plan having the one or more attributes; as well as The treatment plan for the patient is output.

16. The system of claim 15, wherein the at least one processor is further configured to: send the second cost value to the plan optimizer computer model, wherein the plan optimizer computer model modifies the one or more attributes of the treatment plan based on the second cost value.

17. The system of claim 15, wherein the health problem corresponds to at least one of: dry mouth, decreased saliva secretion, headache, hair loss, nausea, vomiting, fatigue, hair loss, skin irritation, memory loss, or aphasia.

18. The system of claim 15, wherein the radiation therapy planning goal corresponds to a dose volume goal.

19. The system of claim 15, wherein the health problem corresponds to developing a secondary cancer.

20. The system of claim 15, wherein the plan optimizer computer model aggregates the second cost value generated by the artificial intelligence model and the cost value generated by the plan optimizer computer model.

Images