Template generation and target structure tracking for radiation therapy

Abstract

The invention relates to template generation and target structure tracking for radiotherapy. Example methods and systems for template generation and target structure tracking are described. In an example, a computer system may acquire (a) planning image data associated with a target structure of a patient in need of radiation therapy or (b) transformed image data generated based on the planning image data. Based on the planned image data and / or the transformed image data, the computer system may generate first material property data representing a particular material property associated with the target structure. Based on the first material property data, the computer system may generate a template representing the particular material property. A template may be generated to match second material property data that also represents specific material properties used to track the target structure during the treatment phase.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] The subject matter of this application is related to US Patent Application No. 18 / 586,532, entitled "Image Data Processing and Target Structure Tracking for Radiation Therapy," the entire contents of which are incorporated herein by reference. Background Art

[0003] Radiation therapy is a widely used cancer treatment modality that uses high-energy radiation to reduce or eliminate cancerous tumors. In reality, the radiation applied cannot inherently distinguish between tumors and adjacent healthy structures, such as organs. Ideally, the goal is to deliver a lethal or therapeutic radiation dose to the tumor while maintaining an acceptable dose level in adjacent healthy structures. During the treatment time, the delivery of the planned radiation dose may be hindered by the presence of patient movement. In this case, motion management can be performed by tracking the position of the moving target structure and taking actions based on any deviations from the planned position to maintain a fairly high level of accuracy in radiation therapy. Conventionally, methods for target structure tracking involve tracking fiducial markers / transponders that have been implanted in the patient. However, the implantation of such markers / transponders may bring additional risks. In some cases, after implantation, some markers may migrate and become unreliable. Summary of the Invention

[0004] According to examples of the present disclosure, target structure tracking can be achieved using a template-based, marker-free approach. To improve tracking accuracy, target structure tracking can be achieved based on specific material properties associated with the target structure. In practice, examples of the present disclosure can be implemented to improve position verification, dose accuracy and consistency, and protection of healthy tissue during the treatment phase of radiotherapy. As will be further described below, examples of the present disclosure should be contrasted with conventional methods that rely on implanted fiducial markers and / or compare different material properties during target structure tracking.

[0005] As used herein, the term "material property" may generally refer to quantifiable characteristics, such as material density / thickness, effective atomic number, etc. (according to examples of the present disclosure, a target structure is trackable based on these characteristics). In practice, a material property may be a characteristic that affects the interaction of a material with ionizing radiation (e.g., X-rays, proton therapy) during radiation therapy. The term "target structure tracking" may generally refer to estimating position data associated with a target structure, such as to facilitate position monitoring and / or verification, target positioning, etc. during a treatment phase of radiation therapy. The term "target structure" may generally refer to any suitable structure that requires tracking, such as a tumor, an organ at risk (OAR), healthy tissue, a bone structure (e.g., a vertebra), etc.

[0006] (a) Template generation and target structure tracking

[0007] According to a first aspect, examples of the present disclosure provide methods and systems for template generation for radiotherapy. In an example, a first computer system may acquire (a) planning image data and / or (b) transformed image data associated with a target structure of a patient requiring radiotherapy. The planning image data may be acquired prior to a treatment phase of radiotherapy. The transformed image data may be generated based on the planning image data. Based on the planning image data and / or the transformed image data, the first computer system may generate first material property data representing specific material properties associated with the target structure. Based on the first material property data, the first computer system may generate a template representing the specific material properties. The template may be generated to match second material property data, which also represents specific material properties for tracking the target structure during the treatment phase.

[0008] According to a second aspect, examples of the present disclosure provide methods and systems for tracking a target structure based on a template representing specific material properties. In an example, a second computer system may acquire material property data representing specific material properties associated with a target structure of a patient requiring radiation therapy. The material property data may be generated based on treatment image data acquired during a treatment phase of radiation therapy. The second computer system may acquire a template that also represents specific material properties associated with the target structure. The template may be generated based on (a) planning image data acquired prior to the treatment phase or (b) transformed image data generated based on the planning image data. Based on the material property data and the template, the second computer system may perform template matching during the treatment phase to track the target structure based on the specific material properties.

[0009] (b) Image data processing and target structure tracking

[0010] According to a third aspect, examples of the present disclosure provide methods and systems for image data processing for radiation therapy. In an example, a second computer system may acquire treatment image data associated with a target structure of a patient requiring radiation therapy, wherein the treatment image data is acquired using an imaging system during a treatment phase of the radiation therapy. The second computer system may process the treatment image data using an artificial intelligence (AI) engine to generate material property data representing specific material properties associated with the target structure. In this manner, the material property data may be generated for matching with a template also representing the specific material properties, for use in tracking the target structure based on the specific material properties during the treatment phase.

[0011] According to a fourth aspect, an example of the present disclosure provides a method and system for tracking a target structure using the output of image data processing. In the example, a second computer system can acquire material property data representing specific material properties associated with a target structure of a patient requiring radiotherapy. The material property data can be generated using an AI engine based on treatment image data acquired during a treatment phase of radiotherapy. The second computer system can acquire a template that also represents specific material properties associated with the target structure. Here, the template can be generated based on (a) planning image data acquired before the treatment phase or (b) transformed image data generated based on the planning image data. Based on the material property data and the template, the second computer system can perform template matching during the treatment phase to track the target structure based on the specific material properties. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart illustrating an example overview process for target structure tracking for radiation therapy using a template-based, marker-free approach;

[0013] Figure 2 is a flow chart illustrating an example process for target structure tracking based on specific material properties;

[0014] Figure 3 is a schematic diagram illustrating a first example radiation therapy system for planning image data acquisition and processing during the planning phase of radiation therapy;

[0015] Figure 4 is a schematic diagram illustrating a second example radiation therapy system for treatment image data acquisition and processing during a treatment session of radiation therapy;

[0016] Figure 5 is a flow chart illustrating a detailed example process for target structure tracking based on material properties in the form of material density or material thickness;

[0017] Figure 6 This is used to generate the Figure 5 a graph of an example fitted model for first material property data in the form of example material density volume data;

[0018] Figure 7 This describes the use of an artificial intelligence (AI) engine to process image data to generate Figure 5 A flowchart of a detailed example process of generating second material property data in an example;

[0019] Figure 8 is a flow chart illustrating a detailed example process for target structure tracking based on material properties in the form of effective atomic number; and

[0020] Figure 9 This describes how image data is processed using an AI engine to generate Figure 8 Flowchart of a detailed example process of obtaining second material property data in an example. DETAILED DESCRIPTION

[0021] In the following detailed description, reference is made to the accompanying drawings which form a part thereof. In the accompanying drawings, similar reference numerals generally denote similar components unless the context dictates otherwise. The illustrative embodiments described in the detailed description, drawings, and claims are not intended to be limiting. Other embodiments may be utilized and other changes may be made without departing from the spirit or scope of the subject matter presented herein. It will be readily understood that the various aspects of the present disclosure as generally described herein and illustrated in the drawings may be arranged, replaced, combined, and designed in a variety of different configurations, all of which are explicitly contemplated herein. Although the terms "first" and "second" are used to describe various elements, these elements should not be limited by these terms. These terms are used to distinguish one element from another. For example, a first element may be referred to as a second element, and vice versa. Independent of the use of grammatical terms, individuals with male and female identities are included within the terms.

[0022] Overview

[0023] Figure 1 is a flow chart illustrating an example overview process 100 for target structure tracking for radiation therapy using a template-based markerless approach. The example process 100 may include one or more operations, functions, or actions illustrated by one or more blocks. Depending on the desired implementation, the various blocks may be combined into fewer blocks, divided into additional blocks, and / or removed. Figure 1 In an example of radiation therapy, radiation therapy may include (a) a pre-treatment planning phase (see 101) to generate a treatment plan for a patient requiring radiation therapy and (b) a treatment phase (see 102) to implement the treatment according to the treatment plan. The goal of the treatment plan is to deliver a high radiation dose to the target structure (e.g., a lung tumor) and a low radiation dose to adjacent organs at risk (OARs) and healthy tissue (e.g., central airways).

[0024] During the treatment phase 102, an important factor for effective treatment delivery is the location of the target structure within the planning target volume (PTV), to which a high radiation dose is delivered. In practice, the patient or the tumor may move outside the PTV due to subtle motions (e.g., respiratory motion, patient movement, etc.). One way to manage motion is to track the target structure using a template-based marker-free approach that does not rely on invasive metal markers / transponders implanted in the patient. Figure 1 In the example of , marker-less target structure tracking may include (a) template generation during the planning phase 101 and (b) template matching during the treatment phase 102. In this way, any deviation from the planned position may be detected during treatment delivery.

[0025] In more detail, Figure 1 At 110-130 in the planning phase 101, template generation can involve generating a set of K templates based on planning image data 110, such as planning computed tomography (CT) data acquired during the planning phase 101 using a first imaging modality (i.e., CT). For example, to enable tracking from all gantry angles, a set of K template images can be generated for K = 360 gantry angles spaced apart by L = 1 degree. Any additional data, such as segmentation data 130 related to the contoured surface (e.g., drawn by a physician or determined using software / AI engine for segmentation), can be used for template generation. For example, segmentation can be performed to generate volumetric image data 131 that identifies the contour, shape, size, and location of the patient's anatomy 132, target structure 133 (e.g., tumor), OAR 134, or any other structure of interest (e.g., soft tissue, bone). The volumetric image data 131 (also referred to as a digital or treatment volume) can be divided into a plurality of smaller volume pixels (voxels) 132, each representing a 3D element within the treatment volume. Note that the volumetric image data 131 may include multiple target structures and irregularly shaped voxels. Depending on the desired implementation, data acquired during treatment phase 102 may also be used for template generation. For example, using adaptive therapy, planning data from a planning CT scan may be propagated to a cone-based computed tomography (CBCT) scan on the day of treatment. In this way, templates may be generated based on the CBCT volume, each template being a container for adaptively encoding planning data from a planning image or a variant thereof.

[0026] exist Figure 1 At 140 in, treatment image data associated with a target structure 133 of the patient can be continuously acquired during irradiation, for example, when using a treatment delivery machine that includes an imaging system to facilitate target structure tracking. The imaging system can acquire treatment image data 140 in the form of kilovoltage (kV) projection images during treatment delivery using any suitable treatment machine. For example, kV projection images can be acquired during volumetric modulated arc therapy (VMAT), where the gantry of the treatment delivery machine continuously rotates around the patient. In another example, kV projection images can be acquired using a proton therapy machine that delivers treatment using protons rather than X-ray radiation. In practice, the treatment image data 140 can include two-dimensional (2D) projection image data, such as single-energy (SE) or dual-energy (DE) CBCT projection image data, or the like.

[0027] exist Figure 1 At 150 in the template selection process, template selection may include selecting a template associated with a gantry angle that is closest to the imaging angle of the treatment image data 140. At 160, any suitable image data processing for improving the quality of the treatment image data 140 may be performed prior to subsequent template matching. At 170, during template matching, any suitable method may be implemented to identify the best match between the selected template and the treatment image data 140, such as calculating a normalized cross-correlation of all possible template positions within a specified search region on the treatment image data 140 as a measure of similarity. The resulting match from template matching 170 may include the current 2D position data of the target structure 133.

[0028] exist Figure 1 At 180 in the figure, 3D position data associated with the target structure 133 can be estimated using any suitable method. In the case of a monoscopic acquisition, triangulation can be performed based on the current 2D position data and the 2D position data associated with the previous gantry angle. In the case of a stereoscopic acquisition using two kV imagers, the instantaneous 3D position can be estimated without using the 2D position data from the previous gantry angle. In this way, the 3D position of the target structure 133 can be monitored and verified while treatment delivery is performed during the treatment phase 102. At 190, any adjustments can be performed in response to detecting excessive position displacement (e.g., exceeding a threshold), such as interrupting treatment and / or recommending patient repositioning.

[0029] Target structure tracking based on material properties

[0030] Will use Figure 2 Further explanation Figure 1 Blocks 120, 160, and 170 in FIG. According to examples of the present disclosure, target structure tracking can be performed based on a template and material property data that both represent the same material property associated with the target structure 133. In particular, during the planning phase 101, a template representing specific material properties associated with the target structure 133 can be generated. During the treatment phase 102, the treatment image data 140 can be processed to generate material property data also representing the specific material properties. In this way, by performing template matching based on the template and material property data representing the same physical quantity, tracking accuracy and dose consistency can be improved.

[0031] The examples of the present disclosure should be contrasted with conventional methods involving template matching based on different physical quantities, which may result in less accurate results. For example, a conventional method may involve generating templates by positively projecting the masked target area from the planning CT volume (i.e., the planning image data 110 acquired using the first imaging mode). These templates represent different physical quantities measured in the 2DCBCT projections (i.e., the treatment image data 140 acquired using the second imaging mode). Another conventional method may involve using dual energy acquisition to create enhanced 2D projection images of soft tissue or bone. However, the enhanced tissue images represent different quantities compared to the templates generated from the CT volume.

[0032] Figure 2 2 is a flow chart illustrating an example process 200 for tracking a target structure based on specific material properties. The example process 200 may include one or more operations, functions, or actions illustrated by one or more blocks. Based on the desired implementation, the individual blocks may be combined into fewer blocks, separated into additional blocks, and / or removed. According to the example process 200, template generation (see blocks 210-250) may be performed during a planning phase 201. Image data processing (see blocks 260-280) and template matching (see block 290) may be performed during a treatment phase 202.

[0033] (a) Template generation

[0034] In more detail, Figure 2 At 210 in the planning phase 201, planning image data 110 and / or transformed image data 211 associated with the target structure 133 can be acquired. As used herein, the term "acquire" can generally refer to data received or retrieved from any suitable source, such as an imaging system, another module / component on the same computer system, another computer system, or a data storage location capable of storing data. The planning image data 110 can be acquired during the planning phase 201 using any suitable radiation therapy system. Figure 3 Examples thereof are described below. Transformed image data 211 can be generated based on planning image data 110, for example, physical property data in the form of one or more of the following: relative electron density (RED) volume data, effective atomic number data, contrast agent map data (e.g., iodine map), etc. As used herein, the term "transformed image data" can generally refer to image data generated based on planning image data by performing any suitable transformation, such as converting from source values ​​in planning image data to target values ​​in transformed image data, etc. Although various examples will be described below using RED volume data, it should be understood that transformed image data 211 can include any additional and / or alternative physical property data.

[0035] exist Figure 2At 220-230 in the figure, first material property data 230 representing a specific material property can be generated based on the planning image data 110 and / or the transformed image data 211. Any suitable "material property" can be used, such as base material density or thickness (see 221), effective atomic number (see 222), etc., based on which the target structure 133 can be tracked. At 240-250, based on the first material property data 230, a template 250 representing the specific material property can be generated for tracking the target structure. Here, a template can be generated to be matched with the second material property data (see 270-280), which template also represents the specific material property during subsequent template matching (see 290).

[0036] (b) Image data processing using AI engine

[0037] exist Figure 2 At 260 in the treatment phase 202, treatment image data 140 associated with a target structure 133 of a patient requiring radiation therapy can be acquired. At 270, the treatment image data 140 can be processed using an AI engine to generate material property data 280 representing specific material properties associated with the target structure 133. The material property data can be generated to match against a template that also represents the specific material properties for use in tracking the target structure based on the specific material properties during the treatment phase 202. Any suitable "material property" based on which the target structure 133 can be tracked can be used, such as base material density or thickness (see 271), effective atomic number (see 272), etc.

[0038] Depending on the desired implementation, the AI ​​engine can be trained to map input = treatment image data 140 and prior knowledge data 261 to output = second material property data 280. As will be further described below, prior knowledge data 261 (see also Figure 5 560 and Figure 8 860) can include (a) simulated projection image data generated by performing a multi-color simulation based on the transformed image data 211 and (b) projection material property data (e.g., first material property data 230) generated during the planning phase 201. Prior knowledge data 261 represents a source of prior information used to improve the results of the AI ​​engine. In practice, the transformed image data 211 can be generated based on the planning image data 110. Depending on the desired implementation, the transformed image data 211 can include one or more of the following physical property data: RED volume data, effective atomic number data, contrast agent map data (e.g., iodine map), etc.

[0039] (c) Target structure tracking

[0040] exist Figure 2At 290 in the figure, template matching can be performed to track the target structure 133 based on the template 250 and the second material property data 280, both of which represent the same specific material property. Depending on the desired implementation, prior knowledge data (see 261) generated based on the treatment image data 140 can also be acquired and processed using the AI ​​engine to generate the second material property data 280.

[0041] In the following, we will use Figures 3 to 9 Detailed examples related to template generation, image data processing, and target structure tracking based on template 250 are described. In particular, Figure 3 Describes an example of a first computer system for template generation and uses Figure 4 An example of a second computer system for image data processing and target structure tracking is described. Figures 5 to 7 Describe an example related to material properties = material density / thickness. Figures 8 and 9 Describe examples related to material properties = effective atomic number.

[0042] Examples of Imaging and Computer Systems

[0043] (a) Plan image data acquisition and processing

[0044] Figure 3 is a schematic diagram illustrating a first example radiation therapy system 300 for acquisition and processing of planning image data during the planning phase of radiation therapy. It should be understood that the first example system 300 may include, in addition to the Figure 3 Here, according to an example of the present disclosure, a first example system 300 may include an imaging system 310 for acquiring planning image data 110, a control system 360 for controlling the operation of the imaging system 310, and a first computer system 370 for processing the planning image data 110.

[0045] exist Figure 3In the example of FIG, the imaging system 310 may include a gantry 311 having an opening 312 and a patient support 313 for supporting a patient 320 requiring radiation therapy. The imaging system 310 may implement any suitable imaging modality for image data acquisition, such as CT, positron emission tomography (PET), single photon emission computed tomography (SPECT), magnetic resonance imaging (MRI), magnetic resonance tomography (MRT), any combination thereof, and the like. For example, when CT is used, the planning image data 110 (e.g., a planning CT scan) may include a series of 2D projection images or slices (e.g., CT slices) each representing a cross-sectional view of the patient's anatomy. For treatment planning, the planning image data may include 3D volumetric CT data, which (sometimes in combination with four-dimensional (4D) CT) is used to estimate the range of motion of the target structure. In practice, spectral CT data (e.g., dual-energy CT (DECT) and photon counting CT) may be acquired to provide access to various quantities during the planning phase.

[0046] exist Figure 3 In the example shown, the gantry 311 has a ring-based configuration. In an alternative example, the gantry can have a C-arm configuration. The imaging system 310 can also include a radiation source 330 (e.g., an X-ray source) to project an imaging beam 350 toward a detector 340 having a pixel detector disposed opposite the radiation source 330. A control system 360 can be electrically coupled to the gantry 311 to control the operation of the gantry 311 using control signals 361. The radiation source 330 can be configured to generate any suitable beam, such as a fan beam. During the imaging process, the gantry 311 can rotate about the opening 312 while the radiation source 330 generates the X-ray beam 350 and directs it along a projection line toward the patient 320 and the detector 340. The detector 340 can measure X-ray absorption and generate a voltage proportional to the intensity of the incident X-rays. The voltage can be read and digitized to generate the planning image data 110. In practice, the planning image data 110 can include image data acquired at different gantry angles.

[0047] According to an example of the present disclosure, the first computer system 370 can be coupled to the imaging system 310 to Figure 2 Blocks 210-250 in acquire and process the planning image data 110 for template generation. Figure 3 In the example of FIG. 3 , the first computer system 370 may include an interface 371 for interacting with the imaging system 310 to obtain the planning image data 110; a material property data generator 372 for generating the first material property data 230; and a template generator 373 for generating a template based on the first material property data 230. The first computer system 370 may include Figure 3In practice, the first computer system 370 may be implemented using a physical machine (bare metal machine) and / or a virtual machine deployed in a cloud-based environment (i.e., not located in the same physical location as the imaging system 310).

[0048] (b) Treatment image data acquisition and processing

[0049] Figure 4 is a schematic diagram illustrating a second example radiation therapy system 400 for acquisition and processing of treatment image data during a treatment session of radiation therapy. It should be understood that the example system 400 may include other components in addition to the system itself, depending on the desired implementation. Figure 4 Here, the example system 400 may include (a) a therapy delivery machine 410 for delivering therapy to a patient 320; (b) a control system 460 for controlling the operation of the machine 410; and (c) a second computer system 470 for processing therapy image data 110 according to examples of the present disclosure.

[0050] The treatment delivery machine 410 may include a gantry 411 that can rotate about an opening 412 and a patient support 413 (e.g., a treatment couch) for supporting the patient 320. Note that the gantry 411 may have a ring-based configuration (in Figure 4 ) or a C-arm configuration (not shown). The treatment delivery machine 410 may include a radiation source in the form of a linear accelerator (LINAC) 420 and an imager / detector in the form of a mega-electronvolt (MV) electron portal imaging device (EPID) 421. The LINAC 420 may be configured to generate a treatment beam 430 through a PTV associated with the patient 320 and direct the treatment beam toward the isocenter 414 as the gantry 411 rotates through the treatment arc during VMAT. In practice, the treatment beam 430 may be in the high energy range, such as 1 MV or greater. In practice, radiation therapy may be delivered as a fractionated treatment, in which the total radiation dose to be delivered to the tumor is divided into smaller "fractions". This is to allow healthy cells to recover from damage caused by the radiation between these fractions, while tumor cells, which are less efficient in recovery, may accumulate damage.

[0051] The treatment delivery machine 410 may also include an imaging system 440 to facilitate kV imaging during application of the MV treatment beam 430. Any suitable imaging modality or modalities, such as SE or DE CBCT, may be used. The imaging system 440 may include at least one kV imaging source 441 and at least one kV imager 442. In contrast to the LINAC 420, the kV imaging source 441 may be capable of generating imaging or diagnostic energies in the kV range. During treatment delivery, the control system 460 may configure the kV imaging source 441 to emit and direct the kV imaging beam 450 to the imager 442, thereby generating treatment image data 140 in the form of kV projection image data to facilitate markerless target structure tracking. Although described with reference to the MV LINAC 420 and the MV treatment beam 430, it should be understood that any additional or alternative treatment delivery technology may be used. For example, a proton therapy machine including a kV imaging system may be used instead.

[0052] According to an example of the present disclosure, a second computer system 470 can be coupled to the therapy delivery machine 410 to provide a therapeutically effective treatment according to the present disclosure. Figure 2 Blocks 260-290 in acquire and process the treatment image data 140. Figure 4 In the example of FIG, the second computer system 470 may include an interface 471 for interacting with the imaging system 440 to obtain the treatment image data 140; a priori knowledge data generator 472 for generating the priori knowledge data 261; an AI engine 473 for generating the second material property data 280; and a template matching engine 474 for performing template matching based on the second material property data 280. In practice, the AI ​​engine 473 and the template matching engine 474 may be implemented by the same computer system (e.g., in FIG). Figure 4 The term "computer system" may generally refer to one or more physical machines (bare metal machines) and / or virtual machines deployed in a cloud-based environment (i.e., not located in the same physical location as the treatment delivery machine 410). Figures 5 to 9 Describe some examples.

[0053] Template generation (base material density / thickness)

[0054] A first example of a material property is the base material density / thickness. In the following, we will use Figure 5 Described is an example template generation flow chart illustrating a detailed example process 500 for tracking a target structure based on a material property in the form of material density or material thickness. The example process 500 may include one or more operations, functions, or actions illustrated by one or more blocks. Depending on the desired implementation, the various blocks may be combined into fewer blocks, divided into additional blocks, and / or removed.

[0055] exist Figure 5 In the example of FIG. 5 , during the planning phase 501, the first computer system 370 may be configured to perform the following operations according to blocks 510 - 550 (i.e., in accordance with Figure 2 During treatment phase 502, the second computer system 470 may be configured to perform template generation based on the first material property data according to blocks 560-590 (i.e., related to Figure 2 260-290 related) perform image data processing and template matching. Can be deployed Figure 3 The first computer system 370 in the embodiment implements blocks 510-550 and may be deployed Figure 4 The second computer system 470 in the embodiment implements blocks 560-590. Figure 5 In the case of the example in , target structure tracking can be performed in an improved manner based on material properties in the form of base material density / thickness.

[0056] (a) Transformation of image data

[0057] exist Figure 5 510 of them, based on the use Figure 3 , computer system 370 may generate transformed image data, physical property volume data, such as in the form of RED volume data, etc. Based on the planning image data 110 (e.g., planning CT data) acquired by imaging system 310 as shown, computer system 370 may generate transformed image data, physical property volume data, such as in the form of RED volume data, etc. In practice, RED volume data 510 is typically generated to facilitate dose calculations during the planning phase 501 and, therefore, may also be used for template generation. In practice, the term "electron density" may generally refer to the number of electrons per unit volume in a material. RED volume data 510 may be a 3D representation of the electron density distribution within the patient's anatomy, typically expressed as a normalized value relative to water. The electron density of a given material affects how radiation interacts with the material, thereby affecting the amount of energy deposited and the type of interaction that may occur.

[0058] Physical property volume data can be obtained from spectral CT (such as DECT scans and photon counting CT), The RED volume data 510 may be acquired using CT reconstruction (a registered trademark of Siemens Healthineers GmbH), multi-color CBCT reconstruction, etc. For example, the generation of RED volume data 510 may involve a transformation or conversion from CT Hounsfield Unit (HU) values ​​to RED values, for example, using a suitable calibration curve. Unlike HU, RED is a well-defined material property and is not dependent on the kilovolt peak (kVp) setting in the CT acquisition. Depending on the desired implementation, HU-RED calibration may not be necessary, such as when using DirectDensity, etc. In addition, RED volume data 510 may be acquired in a multi-color CBCT reconstruction to reflect the current anatomy on the day of treatment in adaptive radiation therapy (particularly for soft tissue tracking).

[0059] (b) First material property data

[0060] exist Figure 5 At 520 in FIG. 5 , based on the RED volume data 510, the computer system 370 may generate first material property data, the first material property data including (a) material density volume data 521 (denoted as p m ) and (b) projected material thickness data 524 (expressed as ). For example, block 521 may involve performing material decomposition in volumetric space (i.e., 3D) to map the RED volume data 510 into base material density volume data 521, which includes one or more of the following: water density volume data (see 522), bone density volume data (see 523), etc. A specific base material may be selected based on the type of target anatomy requiring radiation therapy, such as bone material for vertebral tracking, water material for lung tumor tracking, etc. Projected material thickness data 524 (2D) may be generated by performing a forward projection based on the material density volume data 521 (3D) as follows: Where A represents the front projector and α represents a specific gantry angle.

[0061] Depending on the desired implementation, the basis for the volumetric material decomposition may include a physical model that reproduces the X-ray attenuation properties of a given material (e.g., human tissues such as muscle and fat, or common metal implants) as a combination of selected base materials such as water, bone, titanium alloy, etc. In an example, input RED volume data 510 may be transformed into base material density volume data 521 using a polychromatic attenuation model. Given the attenuation curves of the base materials (e.g., water and cortical bone), the model may optimize a scaling factor for the base material mass densities such that the resulting linear combination of the base material attenuation curves approximates the attenuation curve of the selected material (e.g., muscle tissue).

[0062] For example, to consider Figure 3 The spectral properties associated with the imaging system 310 in the image may be used to weight the differences between the attenuation curves during the optimization process by system sensitivity data reflecting the source spectrum and detector response. Next, a base material scaling factor may be calculated for different human tissues and typical metallic implant materials. In practice, the patient 320 may have a metallic implant (e.g., for a hip replacement), in which case the implant material may be appropriately modeled in the multicolor simulation. A piecewise linear model may then be fit to map the RED values ​​of the tissue and implant material to the base material density. Thus, by construction, a model may be used to account for the spectral properties of the imaged tissue as well as the specific characteristics of the imaging system 310 used to acquire the 2D projections (i.e., the planning image data 110).

[0063] exist Figure 6 Some examples are shown in Figure 6 is a chart illustrating an example fitting model 600 for generating first material property data in the form of base material volume data 530. In this example, the x-axis (see 610) represents inputs = RED values, and the y-axis (see 620) represents outputs = base material density values, such as water density, bone density, and titanium density. Each labeled point represents a different tissue type or metallic implant material sorted by its RED value, e.g., "+" represents water density, "x" represents bone density, and "Δ" represents metal (i.e., titanium) density. The dashed lines (see 630-650) represent different fitting models that can be used by the first computer system 370 to perform material decomposition. A first fitting model 630 is used to map RED values ​​to water density values. A second fitting model 640 is used to map to bone density values. A third fitting model 650 is used to map to metal (i.e., titanium) density values.

[0064] (c) Template generation

[0065] exist Figure 5 At 540 in the block 540, based on the base material density volume data 521 in the volume space, the first computer system 370 can perform template extraction to generate a template 550 representing the material property = base material thickness. The template generation can depend on the target contour and the selected material property = material thickness. For example, block 540 can involve masking the projected material thickness 524 using the projected target contour associated with the target structure 133 along with a margin (e.g., for movement or shape change of the target structure 133).

[0066] In an example, the template 550 can be a soft tissue template image (see 551) generated based on water density volume data (see 521) and associated projected water equivalent thickness data (not shown). In another example, the template 550 can be a bone tissue template image (see 552) generated based on bone density volume data (see 532) and associated tissue equivalent thickness (not shown). The resulting template 551 / 452 can also be referred to as a projected template thickness. In practice, a set of K templates can be generated for all gantry angles, for example, K = 360 templates for a full rotation of the gantry 311 of the imaging system 310. For target structure tracking during the treatment phase 502, each template 550 can be generated to be matched with second material property data 580 that also represents the same material properties.

[0067] (d) Template matching

[0068] exist Figure 5At 590 in FIG. 5 , during the treatment phase 502, target structure tracking can be achieved by performing template matching based on a template 550 and second material property data 580 representing the same material properties. In an example, the template 550 can be a soft tissue template image (see 551 ), which can be matched with the second material property data 580 representing water equivalent thickness for soft tissue tracking (see 581 ). In another example, the template 550 can be a bone tissue template image (see 552 ), which can be matched with the second material property data 580 representing bone equivalent thickness for bone anatomy tracking (see 582 ).

[0069] Depending on the desired implementation, template matching 590 can involve calculating a normalized cross-correlation within a specific search region around the isocenter, which results in a match score value between 0 and 1. Calculating the match score at different pixel offsets produces a match score surface defined over the search region. A possible match can be represented by the highest peak in the match score surface. However, in practice, this may be an incorrect match (especially for noisy images with little contrast). For each match, a peak-to-sidelobe ratio (PSR) can also be calculated, where the PSR is the peak value divided by the standard deviation of the sidelobes. A minimum threshold for the match score value and the PSR can be used to reject possible false matches. The results of template matching 590 can include 2D position data associated with the relevant target structure (e.g., lung tumor, bone anatomy).

[0070] Image data processing (base material density / thickness)

[0071] Will use Figure 7 Further explanation Figure 5 In blocks 560-580, the Figure 7 This describes how image data is processed using an AI engine to generate Figure 5 Flowchart of a detailed example process 700 for determining the second material property data in an example of FIG. Example process 700 may include one or more operations, functions, or actions illustrated by one or more blocks. Depending on the desired implementation, various blocks may be combined into fewer blocks, divided into additional blocks, and / or removed. Figure 4 The second computer system 470 in can be deployed to implement Figure 7 Examples in .

[0072] In fact, according to the example of the present disclosure, image data processing (see Figure 7 and Figure 9) can be implemented to enhance the treatment image data 140 (e.g., 2D projection images), for example, to increase the visibility / contrast of the target structure 133, as represented using specific material properties. This, in turn, can improve the ability to detect the target structure 133 in the 2D projection images during the treatment phase 502. By enhancing the quality of the treatment image data 140 by extracting the second material property data 580, examples of the present disclosure can be implemented to reduce the likelihood that tracking accuracy will be negatively impacted by the low contrast of the target structure 133 (e.g., soft tissue occluded by bony anatomy).

[0073] As used herein, the term "AI engine" may refer to any suitable hardware and / or software component of a computer system that is capable of executing an algorithm according to any suitable AI model. Depending on the desired implementation, the "AI engine" may be a machine learning engine based on a machine learning model, a deep learning engine based on a deep learning model, etc. In general, deep learning is a subset of machine learning in which multi-layer neural networks can be used for feature extraction as well as pattern analysis and / or classification. The term "depth" in deep learning generally refers to the number of layers in the neural network. For example, a deep learning model can have dozens or even hundreds of layers compared to a shallow learning model. This allows the deep learning model to extract more complex and subtle features from the input image data, resulting in better tracking accuracy and performance in radiation therapy.

[0074] Depending on the desired implementation, any suitable AI model may be used to implement the examples of the present disclosure, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), deep belief networks, generative adversarial networks (GANs), autoencoders, variational autoencoders, long short-term memory architectures for tracking purposes, or any combination thereof. In practice, neural networks are typically formed using a network of processing elements (referred to as "neurons," "nodes," etc.) interconnected via connections (referred to as "synapses," "weight data," etc.). The processing layers of a convolutional neural network may be convolutional layers, pooling layers, non-pooling layers, rectified linear unit (ReLU) layers, fully connected layers, loss layers, activation layers, dropout layers, transposed convolution layers, cascade layers, or any combination thereof. For example, a CNN may be implemented using any suitable architecture, such as UNet, LeNet, AlexNet, ResNet, VNet, DenseNet, OctNet, etc.

[0075] (a) Prior knowledge data

[0076] exist Figure 5 560 and Figure 7 At 710-720 in FIG. 1 , the second computer system 470 may generate a plurality of image data based on the planning image data 110 and / or the RED volume data 510 (denoted as p ). e) generates prior knowledge data 560. In an example, the prior knowledge data 560 may include (a) simulated projection image data 710 (I sim ) and (b) prior material thickness data 720(p m，prior ) = Projected material thickness image 524. In practice, the RED volume data 510 (ρ e ) to perform multicolor simulation (see Figure 7 705) to generate I sim Thus, additional information from the multicolor simulation 705 can be used to guide the Figure 7 Model predictions made by the AI ​​engine 740 / 750 in (to be discussed below).

[0077] Simulation projection image data (I sim ) 710 may be a polychromatic attenuation model based on RED volume data (ρ e )510 Simulation intensity image generated For example:

[0078]

[0079] In formula (1), μ(ρ e , E j ) represents the function used to convert ρ e Transformed into a polychromatic attenuation model with coefficient μ; represents the energy-resolved air norm; ∝ represents a specific gantry angle; and k represents the energy spectrum that is low (k=L) or high (k=H) or is discarded in the case of a single energy acquisition.

[0080] (b) AI Engine

[0081] exist Figure 5 At 570 in, the second computer system 470 may use an AI engine, such as Figure 7 The deep learning engine 740 / 750 in

[15] processes the input data (treatment image data 140) and the prior knowledge data 560 to generate predicted output data (second material property data 580). Here, the deep learning engine 740 / 750 can be designed to be "physics-informed" to map the X-ray intensity data in the treatment image data 140 to the second material property data 580 representing the thickness of the underlying material traversed by the beam. This way, during template matching, the second material property data 580 can be directly compared to the template 550 to improve tracking accuracy, as both represent the same material properties. This should be contrasted with conventional methods that compare different physical quantities and / or rely on weighted logarithmic subtraction, which is typically limited to dual-energy images.

[0082] exist Figure 7In the example shown in FIG, the deep learning engine 740 / 750 uses a multi-color forward simulation 705 of attenuation projections and a base material thickness for a specific gantry angle, given pre-treatment RED volume data 510. In this way, the deep learning engine 740 / 750 can leverage the explicit physics model along with prior knowledge data 560 from the planning phase 501 to predict the base material thickness for the entry projection. Prior knowledge data 560 can be used as an additional input channel to constrain the predicted base material thickness given the measured projections.

[0083] Generally, the problem to be solved by the deep learning engine 740 / 750 can be expressed as a mapping defined as follows:

[0084] in,

[0085] Here, the deep learning engine 740 / 750 can be trained to learn a model (f θ ), given the prior knowledge data In the case of I a (ie, treatment image data 140) is mapped to p m (i.e., the second material property data 580). In fact, I a This may include acquired SE or DE CBCT projection image data representing intensity or log-normalized attenuation. sim Indicates that (See 730) Encoded simulated projection image data under the same or similar imaging parameters (see 710), such as gantry angle α, energy resolution air norm etc. p m，prior Represents prior projection material property data, such as projection material thickness data 524 (e.g. bone / water thickness).

[0086] The deep learning engine 740 / 750 can be implemented using any suitable deep neural network architecture for deep regression. Generally, the deep learning engine 740 / 750 may include a plurality of deep neural network architectures denoted as L1 to L2. N A hierarchy of multiple (N) processing layers. Each layer is denoted as w L1 to w LNThe example architecture may include (a) an encoder 741 / 751 to learn a representative latent space and (b) a decoder 742 / 752 to recover the expected output. Depending on the desired implementation, the encoder 741 / 751 may be a neural network layer for extracting features from the input data. The decoder 742 / 752 may be a neural network layer for processing the features extracted from the encoder 741 / 742 to generate output data. Figure 7 Two examples are shown in .

[0087] (a) Input as channels: In the example, the first deep learning engine 740 may be deployed with multiple channels as input. In this case, the imaging parameters can be expanded to a 2D array (e.g. by repetition) and with (I a ;I sim , p m，prior ) stacked to form a path to the encoder 741 Output data (p m ) may be a representation of a 2D projection (I a ) associated with a multi-channel image of predicted material properties.

[0088] (b) Channel plus latent vector: Alternatively, a second deep learning engine 750 with multiple channels plus latent vectors can be deployed. Here, instead of a ;I sim , p m，prior ) stacked to indicate expansion The 2D array can be stacked with the latent vector at a much lower size at the bottleneck of the encoder 751. Note that a shallow set of fully connected layers can be applied to the original to map it to the dimensions of the encoder’s latent space. In practice, the “latent space” may be a representation of the compressed data generated by the encoder 751 and stored in a latent vector.

[0089] Any suitable encoder architecture can be implemented, such as the CNN-based encoder used in UNet, the transformer in Visual Transformer (ViT), or any variant thereof. The ViT model represents the input image as a series of image patches similar to a series of word embeddings used for text-based processing. Any suitable decoder architecture can be used, as long as the final output matches the desired size. Figure 7The dotted lines 743 / 753 in the figure may represent (optional) skip connections between the encoder 741 / 751 and the decoder 742 / 752. Typically, the skip connection 743 / 753 is a direct connection between the encoder 741 / 751 and the decoder 742 / 752 to allow the decoder 742 / 752 to access data from the encoder 741 / 751. The skip connection 743 / 753 is also called a residual connection because it allows the deep learning engine 740 / 750 to learn a residual mapping between input data and output data.

[0090] Using examples of the present disclosure, the deep learning engine 740 / 750 can be designed to be physically informed and trained to learn the model f θ Performing material decomposition in the projected space, the learning model is based on prior knowledge of the data The output of the deep learning engine 740 / 750 is a material property = base material thickness Second material property data 580 such as bone thickness, water thickness, etc. The second material property data 580 may represent an enhanced version of the treatment image data 140 where visibility / contrast of the target structure 133 is improved.

[0091] Template generation: effective atomic number

[0092] A second example of a material property is the effective atomic number (Z eff ). In the following, we will use Figure 8 Describes sample template generation, Figure 8 8 is a flow chart illustrating a detailed example of target structure tracking based on material properties in the form of effective atomic number. Example process 800 may include one or more operations, functions, or actions illustrated by one or more blocks. Depending on the desired implementation, various blocks may be combined into fewer blocks, divided into additional blocks, and / or eliminated.

[0093] In practice, different tissue types typically have varying electron densities and interact differently with radiation. An example use case could be tracking tissues with high effective atomic numbers (Z eff ) associated with a target structure, such as a tumor filled with iodine contrast agent, a metal structure, etc. In this example, during the planning phase 801, the Figure 2 During treatment phase 802, template generation and first material property data generation may be performed according to blocks 860-890 (related to blocks 860-890). Figure 2 260-290 related in) perform image processing and template matching for generating second material property data. Figure 3 The first computer system 370 in the embodiment implements blocks 810-850 and can be deployed Figure 4 The second computer system 470 in implements blocks 860-890.

[0094] (a) Converting image data and first material property data

[0095] exist Figure 8 At 810-830 in FIG. 1 , the first computer system 370 may acquire or generate RED volume data (denoted as p e ) and first material property data, which includes (a) effective atomic number volume data (expressed as Z eff )821 and (b) projected effective atomic number data (expressed as ) 822. Any suitable method may be used to map the planning CT data to (ρ e , Z eff ). Similar to Figure 5 In the example of FIG, the RED volume data 810 can be a 3D representation of the electron density distribution within the patient's anatomy, expressed as a normalized value relative to water. Figure 5 Various explanations related to the RED volume data 510 in FIG. 5 are also applicable here and will not be repeated for the sake of brevity.

[0096] Projected effective atomic number data 822 (ie, 2D) can be generated by performing a forward projection based on the effective atomic number volume data 821 (ie, 3D). This can include forward projecting the RED volume data (ρ) using a forward projection operator (A) e )810 and effective atomic number volume data (Z eff ) 830 to obtain the p seen by a ray passing through the volume towards a particular pixel at the detector for a particular gantry angle (α) e and Z eff The line integral of :

[0097]

[0098] According to the publication by Alvarez RE and Macovski A. entitled “Energy-selective reconstructions in X-ray computerized tomography” (see Phys Med Biol. 1976 Sep; 21(5): 733-54) and incorporated herein by reference, the energy-dependent attenuation μ(E) can be approximated using the following formula:

[0099]

[0100] In formula (3), ρ e Indicates the RED value, fKN (E) represents the material-specific Klein-Nishina function, and c1, c2, and n are model parameters obtained by fitting the model to the data. By knowing the spectral properties of the CBCT imaging system, equation (3) can be used to obtain ρ from the planning image data 110 (e.g., planning CT). e and Z eff Volume data are mapped to the energy acquisition I for low and high L and I H The measured intensity of the treatment image data 140 (e.g., CBCT projections):

[0101]

[0102] In formula (4), k represents the low energy index (i.e., k = L, in I L Medium) or high energy index (i.e. k = H, in I H In addition, denotes the energy-resolved air norm, A is the forward projection operator, α is the gantry angle, and μ(E) is calculated according to the approximation in equation (3).

[0103] (b) Template generation

[0104] exist Figure 8 At 840 in the first computer system 370, the template extraction can be performed to generate a template 550 representing the material property = effective atomic number associated with the target structure 133 to be tracked. For example, at 841, in order to increase the visibility of targets with high atomic numbers, the original target structure Z can be replaced by water within the target structure volume. eff To generate projected background effective atomic number data:

[0105]

[0106] exist Figure 8 At 842, the first computer system 370 may calculate the projected effective atomic number data based on the projected effective atomic number data. and projected background effective atomic number data A template representing the material property = effective atomic number for enhancing the target structure is generated as follows:

[0107]

[0108] (c) Template matching

[0109] exist Figure 8 At 890 in the treatment phase 802, the template 850 may be based on a material property representing the same material property = effective atomic number. and second material property data 880 (pm ) performs template matching to achieve target structure tracking. Similar to Figure 5 In the example of , template matching 890 may involve computing normalized cross-correlations, etc. Figure 5 The various explanations related to block 590 in are also applicable here and will not be repeated for the sake of brevity.

[0110] Image data processing (effective atomic number)

[0111] Will use Figure 9 Further explanation Figure 8 In blocks 860-880, the Figure 9 This describes how image data is processed using an AI engine to generate Figure 8 900. The example process 900 may include one or more operations, functions, or actions illustrated by one or more blocks. Depending on the desired implementation, various blocks may be combined into fewer blocks, divided into additional blocks, and / or removed. Figure 4 The second computer system 470 in can be deployed to implement Figure 9 Examples in .

[0112] (a) Prior knowledge data

[0113] exist Figure 8 860 and Figure 9 At 910-920 in FIG. 1 , the second computer system 470 may process the image data 110 and / or the RED volume data 810 (ρ e ) generates prior knowledge data 860. In an example, the prior knowledge data 560 may include (a) simulated projection image data 810 (I sim ) and (b) are the projection effective atomic number data 822 in formula (2) The a priori projection effective atomic number data 820 (p m，prior ). Similar to Figure 7 In the example, I sim The RED volume data 810 (ρ e ) and effective atomic volume data (Z eff ) 821 performs a multicolor simulation 905, for example, using the example multicolor attenuation model in equation (4). sim It can also be called simulated intensity image data.

[0114] (b) AI Engine

[0115] exist Figure 8 At 870 in, the second computer system 470 may use an AI engine, such as Figure 9The deep learning engine 940 / 950 in the image processing module processes the input data (treatment image data 140) and the prior knowledge data 860 to generate predicted output data (second material property data 880). Here, the deep learning engine 940 / 950 can be designed to be "physics-informed" to map the X-ray intensity data in the treatment image data 140 to the second material property data 880 representing the effective atomic number. During template matching, the second material property data 880 can be directly compared with the template 850 to improve tracking accuracy because both represent the same material properties.

[0116] Similar to Figure 7 In the example in

[15] , the deep learning engine 940 / 950 can be trained to learn a model (f θ ), so that I a Mapping to p m Here, I a Represents the acquired SE or DE CBCT projection image data. sim Indicates that (See 930) Encoded simulated projection image data under the same or similar imaging parameters (see 910), such as gantry angle α, energy resolution air norm etc. p m，prior Prior material property data 920 = projected effective atomic number data in formula (4)

[0117] in,

[0118] Similar to Figure 7 In the example shown in FIG, the deep learning engine 940 / 950 may include (a) an encoder 941 / 951 to learn a representative latent space and (b) a decoder 942 / 952 to recover the expected output. Any suitable encoder / decoder architecture may be implemented. In addition, the skip connection 943 / 953 connecting the encoder 941 / 951 and the decoder 942 / 952 is optional. Figure 9 Two example implementations are shown in FIG.

[0119] (a) Input as channels: In the example, the first deep learning engine 940 can be deployed with multiple channels as input. In this case, the imaging parameters can be expanded to a 2D array (e.g. by repetition) and with (I a ;I sim， p m，prior ) stacked to form a path to encoder 941 Output data (pm ) may be a representation of I collected during treatment phase 802 a A multi-channel image of the associated projected effective atomic number data 880 .

[0120] (b) Channel plus latent vector: Alternatively, a second deep learning engine 950 with multiple channels plus latent vectors can be deployed. Compared with the first deep learning engine 940, the extended The 2D array of can be stacked with the latent vector at a much lower size at the bottleneck of encoder 951.

[0121] The output of the deep learning engine 940 / 950 is the second material property data 880 (p m ). The second material property data 880 may represent an enhanced form of the treatment image data 140, wherein the visibility / contrast of the target structure 133 is improved to improve tracking accuracy and treatment outcomes for the patient 320. Figure 8 During template matching at block 890 in , the second material property data 880 can be directly compared with the template 850 since both represent the same material property = effective atomic number. Again, this should be contrasted with conventional methods of comparing different physical quantities / properties.

[0122] For the collected low energy projection I L and High Energy Projection I H , is expressed as The second material property data 880 can be obtained using the model (f θ ). Note that the transformation from the measured intensity I to the attenuation μ is given by the logarithmic normalization step: And for a specific acquisition, the corresponding air norm I0 is usually known. By subtracting the projected background effective atomic number Target enhanced image data may be generated and subsequently matched with the template in a template-based matching process at block 890. Make a comparison.

[0123] AI engine training

[0124] The deep learning engine 740 / 750 / 940 / 950 may be trained prior to the treatment delivery phase 502 / 802 using any suitable method, such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, combinations thereof, etc. In the case of supervised learning, the deep learning engine 740 / 750 / 940 / 950 may be trained using labeled training data comprising labeled examples in order to learn to Mapping to the desired output data = pm The model (f θ ). In contrast, when using unsupervised learning, the deep learning engine 740 / 750 / 940 / 950 can be trained using unlabeled training data to reveal hidden patterns and structures within the data.

[0125] In the case of using semi-supervised learning, a combination of labeled training data and (possibly a larger set of) unlabeled training data can be used to train the deep learning engine 740 / 750 / 940 / 950. In this case, the deep learning engine 740 / 750 / 940 / 950 can use the labeled training data to learn to model (f θ ), and then the model can be refined using unlabeled training data (f θ ). For semi-supervised methods, the measured intensities in the collected projections can be compared with the Figure 7 Projected material thickness data 580 (or Figure 9 The intensity of the projected RED and the projected effective atomic number data 880 are compared. For example, in the case of inferred water / bone thickness, the intensity estimate can be calculated as follows:

[0126]

[0127] In formula (7), m is the material index (bone / water), is the estimated material thickness in the CBCT projection, μ m (E j ) represents the material-specific attenuation curve, and k represents the low / high energy spectrum or is discarded in the case of single energy acquisition. Possible loss functions can be measured (ie, treatment image data 140) and the estimated The mean square error (MSE) between them.

[0128] Computer system

[0129] The above examples may be implemented by hardware (including hardware logic circuits), software, or firmware, or a combination thereof. The above examples may be implemented by any suitable computing device, computer system, or the like. The computer system may include a processor, a memory unit, and a physical NIC, etc., which may communicate with each other via a communication bus. The computer system may include a non-transitory computer-readable medium having instructions or program codes stored thereon, which, when executed by a processor, cause the processor to perform the processes described herein with reference to the accompanying drawings.

[0130] The techniques described above can be implemented in dedicated hard-wired circuitry, in software and / or firmware in combination with programmable circuitry, or in a combination thereof. The dedicated hard-wired circuitry can be in the form of, for example, one or more application-specific integrated circuits (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), etc. The term "processor" should be interpreted broadly to include a processing unit, an ASIC, a logic unit, or a programmable gate array, etc.

[0131] The foregoing detailed description illustrates various embodiments of devices and / or processes through the use of block diagrams, flow charts, and / or examples. Where such block diagrams, flow charts, and / or examples include one or more functions and / or operations, those skilled in the art will appreciate that each function and / or operation within such block diagrams, flow charts, or examples may be implemented individually and / or collectively by a variety of hardware, software, firmware, or any combination thereof.

[0132] Those skilled in the art will recognize that some aspects of the embodiments disclosed herein may be equivalently implemented in whole or in part in an integrated circuit as one or more computer programs running on one or more computers (e.g., one or more programs running on one or more computer systems), one or more programs running on one or more processors (e.g., one or more programs running on one or more microprocessors), firmware, or virtually any combination thereof, and that designing circuits and / or writing code for the software and / or firmware will be well within the skill of those skilled in the art in light of this disclosure.

[0133] The software implementing the techniques described herein can be stored on a non-transitory computer-readable storage medium and can be executed by one or more general-purpose or special-purpose programmable microprocessors. As used herein, the term "computer-readable storage medium" includes any mechanism that provides (i.e., stores and / or transmits) information in a form accessible to a machine (e.g., a computer, a network device, a personal digital assistant (PDA), a mobile device, a manufacturing tool, any device having a group of one or more processors, etc.). Computer-readable storage media can include recordable / non-recordable media (e.g., read-only memory (ROM), random access memory (RAM), magnetic or optical storage media, flash memory devices, etc.).

[0134] The accompanying drawings are merely illustrative examples, wherein the units or processes shown in the accompanying drawings are not necessarily necessary for implementing the present disclosure. Those skilled in the art will appreciate that the units in the devices in the examples may be arranged in the devices in the examples as described, or may alternatively be located in one or more devices different from the devices in the examples. The units in the described examples may be combined into a module or further divided into multiple subunits.

Claims

1. A method for performing template generation for tracking a target structure on a computer system, wherein: The method comprises: acquiring (a) planning image data associated with a target structure of a patient requiring radiation therapy and acquired prior to a treatment phase of the radiation therapy, or (b) transformed image data generated based on the planning image data; generating first material property data representing a specific material property associated with the target structure based on the planning image data or the transformed image data or both; and Based on the first material property data, a template representing the specific material property is generated, wherein the template is matchable with second material property data also representing the specific material property for tracking the target structure during the treatment phase.

2. The method according to claim 1, wherein Generating the first material property data includes: The first material property data is generated, the first material property data representing the specific material property in the form of a material density or a material thickness associated with the target structure.

3. The method according to claim 2, wherein: Generating the first material property data includes: generating the first material property data including material density volume data based on the transformed image data; and The first material property data including projected material thickness data associated with the target structure is generated by performing a forward projection based on the material density volume data.

4. The method according to claim 3, wherein: Generating the template includes: The template representing a material thickness associated with the target structure is generated based on the projected material thickness data.

5. The method according to claim 1, wherein Generating the first material property data includes: The first material property data is generated, the first material property data representing the specific material property in the form of an effective atomic number associated with the target structure.

6. The method according to claim 5, wherein: Generating the first material property data includes: generating the first material property data including effective atomic number volume data based on the planning image data; and The first material property data including projected effective atomic number data associated with the target structure is generated by performing a forward projection based on the effective atomic number volume data.

7. The method according to claim 6, wherein: Generating the template includes: The template representing the effective atomic number is generated based on the projected effective atomic number data.

8. A method for performing target structure tracking for radiotherapy on a computer system, wherein: The method comprises: acquiring material property data representing specific material properties associated with a target structure of a patient requiring radiation therapy, wherein the material property data is generated based on treatment image data acquired during a treatment session of the radiation therapy; and acquiring a template that also represents the specific material properties associated with the target structure, wherein the template is generated based on (a) planning image data acquired prior to the treatment phase or (b) transformed image data generated based on the planning image data; and Based on the material property data and the template, template matching is performed during the treatment phase to track the target structure based on the specific material properties.

9. The method according to claim 8, wherein Performing the template matching includes: The template matching is performed based on the material property data and the template, both of which represent the specific material property in the form of a material thickness associated with the target structure.

10. The method according to claim 9, wherein: Acquiring the template includes: The template representing material thickness is acquired based on projected material thickness data associated with the target structure, wherein (a) the projected material thickness data is generated based on material density volume data, and (b) the material density volume data is generated based on the transformed image data.

11. The method according to claim 8, wherein Performing the template matching includes: The template matching is performed based on the material property data and the template, both of which represent the specific material property in the form of an effective atomic number associated with the target structure.

12. The method according to claim 11, wherein Acquiring the template includes: The template representing the effective atomic number is acquired based on projected effective atomic number data associated with the target structure, wherein (a) the projected effective atomic number data is generated based on effective atomic number volume data and the transformed image data, and (b) the effective atomic number volume data is generated based on the planning image data.

13. The method according to claim 8, wherein Acquiring the material property data includes: The material property data is generated by processing the treatment image data using an artificial intelligence (AI) engine.

14. The method according to claim 8, wherein Performing the template matching includes: The template matching is performed to match the material property data with the template to determine two-dimensional (2D) position data associated with the target structure based on the specific material property.

15. A computer system comprising: processor; as well as A non-transitory computer-readable medium having stored thereon instructions that, when executed by the processor, cause the processor to: acquiring (a) planning image data associated with a target structure of a patient requiring radiation therapy and acquired prior to a treatment phase of the radiation therapy, or (b) transformed image data generated based on the planning image data; generating first material property data representing a specific material property associated with the target structure based on the planning image data or the transformed image data or both; and Based on the first material property data, a template representing the specific material property is generated, wherein the template is matchable with second material property data also representing the specific material property for tracking the target structure during a treatment phase.

16. The computer system according to claim 15, wherein: The instructions for generating the first material property data cause the processor to: The first material property data is generated, the first material property data representing the specific material property in the form of a material density or a material thickness associated with the target structure.

17. The computer system of claim 16, wherein: The instructions for generating the first material property data cause the processor to: generating the first material property data including material density volume data based on the transformed image data; as well as The first material property data including projected material thickness data associated with the target structure is generated by performing a forward projection based on the material density volume data.

18. The computer system of claim 17, wherein: The instructions for generating the template cause the processor to: The template representing a material thickness associated with the target structure is generated based on the projected material thickness data.

19. The computer system according to claim 15, wherein: The instructions for generating the first material property data cause the processor to: The first material property data is generated, the first material property data representing the specific material property in the form of an effective atomic number associated with the target structure.

20. The computer system of claim 19, wherein: The instructions for generating the first material property data cause the processor to: generating the first material property data including effective atomic number volume data based on the planning image data; as well as The first material property data including projected effective atomic number data associated with the target structure is generated by performing a forward projection based on the effective atomic number volume data.

21. The computer system of claim 20, wherein: The instructions for generating the template cause the processor to: The template representing the effective atomic number is generated based on the projected effective atomic number data.

Images