Iterative image reconstruction with upsampling

Abstract

The present disclosure relates to up-sampled iterative image reconstruction. In one example, a computer system can acquire projection image data associated with a target structure of a patient. The computer system can perform an iterative image reconstruction for a first number of iterations based on the projection image data to generate first volumetric image data. The first volumetric image data can be associated with a first resolution level. The computer system can up-sample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level that is higher than the first resolution level. The computer system can perform an iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data associated with the second resolution level.

Description

Technical Field

[0001] This application generally relates to the field of image processing, and more specifically, to iterative image reconstruction utilizing upsampling. Background Technology

[0002] Radiation therapy is a widely used cancer treatment that uses high-energy radiation to shrink or eliminate cancerous tumors. In practice, the applied radiation itself cannot distinguish between the tumor and nearby healthy structures, such as organs and healthy tissues. Ideally, the goal is to deliver a lethal or curative dose of radiation to the tumor while maintaining acceptable dose levels in healthy structures. Image reconstruction can be performed to generate volumetric image data based on projection image data associated with the patient. Based on volumetric image data, clinicians and planning tools can more accurately target the tumor while avoiding unnecessary radiation exposure to healthy structures. Therefore, improving the quality of image reconstruction to enhance the efficacy of radiation therapy and ultimately improve patient outcomes is desirable. Summary of the Invention

[0003] Examples of this disclosure can be implemented to improve the computational efficiency of image reconstruction for radiotherapy, resulting in faster clinical workflows and improved patient outcomes. According to a first aspect of this disclosure, methods and systems for iterative image reconstruction utilizing upsampling are described. In one example, a computer system can acquire projected image data associated with a target structure of a patient. The computer system can perform iterative image reconstruction for a first number of iterations based on the projected image data to generate first volumetric image data. The first volumetric image data can be associated with a first resolution level. The computer system can upsample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, which is higher than the first resolution level. The computer system can perform iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data. The output volumetric image data can be associated with a second resolution level. Various embodiments will be used. Figure 1 , Figures 2B-11 To explain.

[0004] According to a second aspect of this disclosure, methods and systems for reconstructing multi-resolution images using upsampled images are described. In one example, a computer system may acquire first volumetric image data generated based on projected image data associated with a target structure of a patient. The computer system may upsample the first volumetric image data associated with a first resolution level to generate second volumetric image data associated with a second resolution level, which is higher than the first resolution level. The computer system may acquire mask data that identifies regions of interest (ROIs) associated with the target structure. The computer system may generate multi-resolution volumetric image data based on the second volumetric image data and the mask image data. The multi-resolution volumetric image data may include at least a first voxel associated with a second resolution level outside the ROI and a second voxel associated with a third resolution level within the ROI. Various embodiments will use... Figures 12-15 To explain.

[0005] Examples of this disclosure may also include a computer system comprising a processor and a non-transient computer-readable medium storing instructions that, when executed by the processor, cause the processor to perform multiple aspects of the methods described above. Another aspect may include a non-transient computer-readable storage medium comprising a set of instructions that, when executed by a processor or computer, cause the processor or computer to perform multiple aspects of the methods described above. Yet another aspect may include a computer program comprising instructions that, when executed by a processor or computer, cause the processor or computer to perform multiple aspects of the methods described above. A further aspect may include an imaging system comprising an imaging source and a detector (also referred to as an imager) for acquiring projected image data; and a computer system for performing multiple aspects of the methods described above. Attached Figure Description

[0006] Figure 1 This is a flowchart illustrating an example process by which a computer system performs iterative image reconstruction using upsampling for radiotherapy;

[0007] Figure 2A This is a schematic diagram illustrating traditional iterative image reconstruction;

[0008] Figure 2B This is a schematic diagram illustrating a first example of image reconstruction using an iterative image reconstruction process with an upsampling phase;

[0009] Figure 3 This is a schematic diagram illustrating a second example of iterative image reconstruction using multiple upsampling stages;

[0010] Figure 4This is a schematic diagram of an example radiotherapy system including a computer system that performs image reconstruction during the pre-treatment phase of radiotherapy;

[0011] Figure 5 This is a schematic diagram of an example radiotherapy system that includes a computer system that performs image reconstruction during the treatment phase of radiotherapy.

[0012] Figure 6 This is a flowchart illustrating a detailed example process of a computer system performing iterative image reconstruction using upsampling for radiotherapy;

[0013] Figure 7 It is the computer system in Figure 6 The flowchart illustrates the example process for implementing the resolution refinement stage;

[0014] Figure 8 It is the computer system in Figure 6 The flowchart illustrates the example process for implementing the upsampling phase.

[0015] Figure 9 This is a flowchart illustrating an example process by which a computer system performs iterative image reconstruction using Algebraic Reconstruction Technique (ART) and / or Penalized Likelihood (PL) techniques.

[0016] Figure 10A This is a schematic diagram illustrating a first example user interface (UI) view of a user interacting with first volumetric image data;

[0017] Figure 10B This is a schematic diagram illustrating a second example UI view of user interaction with output volumetric image data;

[0018] Figure 11 This is a schematic diagram illustrating multiple UI views used to display multiple sets of volumetric image data;

[0019] Figure 12 This is a flowchart illustrating an example process by which a computer system performs reconstruction of multi-resolution images using upsampled images for radiotherapy;

[0020] Figure 13 This is a schematic diagram illustrating a first example of multi-resolution image reconstruction using an upsampling stage;

[0021] Figure 14 This is a schematic diagram illustrating a second example of multi-resolution image reconstruction utilizing multiple upsampling stages;

[0022] Figure 15 This is a flowchart illustrating a detailed example process of a computer system performing multi-resolution image reconstruction using upsampled images for radiotherapy. Detailed Implementation

[0023] In the following detailed description, reference is made to the accompanying drawings, which form a part of the description. In the drawings, unless the context otherwise requires, similar symbols are generally used to indicate similar components. The illustrative embodiments described in the detailed description, drawings, and claims are not intended to be limiting. Other embodiments may be utilized, and other modifications may be made, without departing from the spirit or scope of the subject matter presented herein. It will be readily understood that the aspects of this disclosure, as generally described herein and illustrated in the drawings, can be arranged, substituted, combined, and designed in a variety of different configurations, all of which are expressly included 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 elements from one another. For example, a first element may be referred to as a second element, and vice versa.

[0024] Overview

[0025] Imaging modalities such as cone-beam computed tomography (CBCT) are widely used in clinical practice for diagnosing various diseases, as tools during surgical procedures, and as localization tools before radiotherapy. To facilitate visualization of a patient's internal anatomy, image reconstruction can be performed to generate three-dimensional (3D) volumetric image data of the patient's internal anatomy based on two-dimensional (2D) projected image data (also known as projection) acquired using the imaging system.

[0026] As used herein, the term "image reconstruction" can generally refer to the process of generating volumetric image data based on projected image data. The term "projected image data" (which is used interchangeably with "2D projected data," "2D projected image," and "projection") can generally refer to data representing the characteristics of illumination radiation rays passing through an object. The term "volumetric image data" (also referred to as "reconstruction" or "reconstructed image") can generally refer to 3D reconstructed data generated based on projected image data.

[0027] Current CBCT imaging systems can be configured to provide high-resolution axial slices, such as 512x512 pixels, with a resolution of approximately 0.5 mm / pixel or mm pixels for head scans and approximately 1 mm / pixel or mm pixels for in-plane body scans. However, these systems employ relatively coarse longitudinal (out-of-plane) sampling intervals, such as 2 mm / pixel or mm pixels for image-guided radiotherapy (IGRT) and 3 mm / pixel or mm pixels for CBCTp for treatment planning. Generally, coarse longitudinal sampling is used to reduce the number of slices and the computational time and workload required for analyzing and depicting visible anatomical structures. However, using coarse longitudinal sampling can lead to undesirable side effects.

[0028] First, using conventional, mismatched projectors for CBCT image reconstruction can introduce sampling artifacts, such as aliasing and shadowing caused by the Gibbs phenomenon when recovering high-frequency components. Second, while axial views may retain usable detail, sagittal and / or coronal views can be prone to severe sampling artifacts, thus impairing image quality in these planes. Third, coarse sampling methods may limit the feasibility of further functional improvements, such as arbitrary planar slicing.

[0029] In principle, existing image reconstruction algorithms can use current-generation flat-panel imagers to generate images with higher longitudinal resolution. Using cubic voxels (i.e., a grid with the same resolution in every direction) offers several potential advantages, but may require improvements for feasibility. In practice, the time required for image reconstruction depends on the voxel grid resolution. Higher resolution grids can significantly increase the computational load and time required for image reconstruction. Therefore, it is desirable to achieve higher resolution image reconstruction in a more efficient manner.

[0030] The first aspect: Reconstructing images using upsampled iterative images

[0031] According to a first aspect of this disclosure, a method and computer system for iterative image reconstruction utilizing upsampling are described. Using examples from this disclosure, iterative image reconstruction can be implemented using multiple stages, including K upsampling stages and corresponding... K +1 resolution level associated K +1 resolution refinement stage. At least one upsampling stage can be implemented (i.e., K ≥1). Will use Figure 1 and Figure 2B To explain K The case is =1 (i.e., one upsampling stage). This will be used... Figure 3 To explain K=2 (i.e., two upsampling stages). Based on the desired implementation, the highest resolution level ( k +1) can be configured to be lower than or equal to the original resolution level associated with the imaging system (denoted as RESn).

[0032] More in detail, Figure 1 This is a flowchart of an example process (see 100) of a computer system performing iterative image reconstruction using upsampled images for radiotherapy. Example process 100 may include one or more operations, functions, or actions illustrated by one or more boxes such as 110 to 170. Depending on the desired implementation, the individual boxes may be combined into fewer boxes, divided into more boxes, and / or deleted. Any suitable computer system, such as... Figures 4-5 The computer system 470 (explained below) can be configured to perform various examples of this disclosure.

[0033] exist Figure 1 110 locations, Figures 4-5 The computer system 470 can acquire projection image data associated with the target structures of a patient requiring radiotherapy. Projection image data 110 (denoted as P_RES1) can be associated with a first resolution level (denoted as RES1). Projection image data 110 may include a set of multiple (M) projection images, denoted as... ,in In practice, projected image data 110 (P_RES1) can represent first downsampled projected image data generated by downsampling the original resolution projected image data 105 (P_RESn). The original resolution level (RESn) can represent the resolution level associated with the imaging or image acquisition system.

[0034] As used herein, the term “acquisition” can generally refer to receiving or retrieving data from any suitable source, such as an imaging system, a module / component of the same or different computer system, or a database storing the data. The term “target structure” can generally refer to any suitable structure of interest, such as a tumor, an organ at risk (OAR), healthy tissue, skeletal structures (e.g., vertebrae), etc. Note that the projected image data 110 is not limited to acquisition from a single detector; it can come from multiple sources, angles, energy levels, or even imaging modes.

[0035] exist Figure 1At point 120, based on the projected image data 110, the computer system 470 can perform iterative image reconstruction for a first number of iterations (denoted as N1). Box 120 can be executed to generate first volumetric image data associated with RES1 (denoted as V1_OUT 130). As used herein, the term "iterative image reconstruction" or "iterative reconstruction" can generally refer to the computational process of generating volumetric image data from projected image data through multiple iterations. Unlike traditional reconstruction methods that apply direct mathematical formulas, iterative reconstruction can be performed iteratively to refine the volumetric image data, gradually improving its accuracy. Figures 6-7 This section describes an example algorithm for iterative image reconstruction.

[0036] exist Figure 1 At position 140, computer system 470 can upsample V1_OUT 130 associated with resolution RES1 to generate second volumetric image data (denoted as V2_IN 150) associated with a higher second resolution (denoted as RES2), i.e., RES2 > RES1. In this specification, the term "upsampling" can generally refer to converting lower-resolution volumetric image data into higher-resolution volumetric image data, thereby improving its resolution level. (The last sentence appears to be incomplete and possibly refers to a different context.) Figure 6 and Figure 8 As further described, any suitable method for upsampling, such as bilinear upsampling, cubic upsampling, artificial intelligence (AI) engines for upsampling, etc., can be implemented at box 125. In general, more sophisticated methods can be implemented to further reduce the number of iterations required at higher resolutions to recover fine / high-frequency details.

[0037] exist Figure 1 At position 160, based on V2_IN 150, computer system 470 can perform iterative image reconstruction for a second iteration (denoted as N2). The result is output volumetric image data (denoted as V2_OUT 170) associated with the second resolution level (RES2).

[0038] Compare

[0039] The examples of this disclosure can be implemented to improve the computational efficiency of image reconstruction, resulting in faster clinical workflows and more timely clinical decisions. The examples of this disclosure should be compared with traditional iterative reconstruction methods that involve processing entire high-resolution datasets. The following will use… Figures 2A-2B To explain the comparison. Here, Figure 2A This is a schematic diagram illustrating a traditional iterative image reconstruction method; while Figure 2B This is a schematic diagram illustrating a first example of image reconstruction using upsampling iterative methods.

[0040] exist Figure 2A In this process, iterative image reconstruction (see 220) can be performed based on an initial volume (see 210) to generate high-resolution volumetric image data (see 230). Since the reconstruction time depends on the voxel grid resolution, therefore... Figure 2A High-resolution iterative reconstruction in image processing can be inherently slow and computationally expensive. This is because iterative image reconstruction typically relies on executing a pair of projectors (i.e., a forward projector and a backward projector) during the iteration process. This pair of projectors works together to progressively update the volumetric image data until convergence is achieved or stopping conditions / criteria (such as reaching convergence or the maximum number of iterations) are met. See also Figure 7 Example projectors at positions 710 and 760 (described below).

[0041] In practice, forward projectors simulate X-rays passing through a volume to generate simulated projection data, while backward projectors use this projection data to update information within the volume domain. The implementation of forward and backward projection operations typically constitutes the majority of the computational time required for iterative image reconstruction. The runtime of these projectors is directly proportional to the total number of voxels and the corresponding number of detector pixels. The number of floating-point operations and memory accesses for each of these projectors is likely approximately equal to the number of voxels in the image domain multiplied by the number of pixels in the projection domain. Therefore, computational demands increase with higher resolutions. For large image volumes with fine resolution, computational complexity can rise rapidly.

[0042] In contrast, the first aspect of this disclosure provides an iterative image reconstruction method with multi-resolution in the temporal domain, meaning that the resolution of the volumetric image data evolves over time and changes at different stages of reconstruction. Examples of this disclosure can be implemented to reduce computation time by leveraging the observation that early iterations in the iterative reconstruction process first recover low-frequency features, while later iterations contribute to higher-frequency features. Furthermore, examples of this disclosure can be implemented to reduce clinical workflow time, for example by generating a user interface (UI) view to provide a user (e.g., a clinician) with a lower resolution V1_OUT 130 as a preview, followed by a higher resolution V2_OUT 170.

[0043] exist Figure 2BIn the example shown, lower-resolution data is processed during a first time period (see 280), and then higher-resolution data is processed during a second time period (see 290). During the first resolution refinement phase (see 250), iterative reconstruction can be performed based on the initial volumetric image data (V1_IN) 240 and the first downsampled projected image data (P_RES1) 110 to generate V1_OUT 130 associated with RES1. During the upsampling phase (see 260), V1_OUT 130 can be upsampled to V2_IN 150 associated with RES2. During the final resolution refinement phase (see 270), iterative reconstruction can be performed to refine V2_IN 150 to output volumetric image data (V2_OUT) 170. This can be based on... Figure 1 The original resolution P_RESn 105 in the image is downsampled to generate a second downsampled projected image data (denoted as P_RES2) to perform the final stage 270. This is used below. Figure 6 Further explanation of the downsampling of P_RESn 105.

[0044] In practice, any suitable RES1 and RES2 can be used. For example, in Figure 2B In this model, RES1 can be 256x256x193, while RES2 is 512x512x386. In this case, V2_IN 150 has 2x2x2 = 8 more voxels compared to V1_OUT 130. This multi-stage image reconstruction method can lead to faster convergence, shorter processing time, and lower computational resource requirements, while still achieving high-resolution results. This can ultimately improve patient outcomes and increase the number of patients treatable on a single radiotherapy system.

[0045] Multiple (K>1) upsampling stages

[0046] Figure 3 It shows K =The case of 2 upsampling stages, this Figure 3 This is a schematic diagram illustrating a second example (see 300) of iterative image reconstruction using multiple upsampling stages. Figure 3 Examples include K+1 = 3 resolution levels: RES1 ("first resolution level") for k=1, a lower resolution RES2 ("second resolution level") for k=2, and a higher resolution RES3 ("third resolution level") for k=3. This allows for multi-stage iterative reconstruction, first recovering the low-frequency features associated with RES1, then the mid-frequency features associated with RES2, and finally the finer details associated with RES3. This should be compared to traditional methods involving iterative reconstruction using only one resolution level, as explained using 2A.

[0047] exist Figure 3 During the first stage 301 shown, resolution refinement (see 315) can be performed for N1 iterations based on the initial volumetric image data (V1_IN) 310 and the first downsampled projected image data (P_RES1) 110 to generate V1_OUT 320 associated with RES1. During the second stage 302, upsampling (see 325) can be performed to transform V1_OUT 320 into V2_IN 330 associated with the intermediate resolution RES2. During the third stage 303, resolution refinement (see 335) can be performed for N2 iterations based on V2_IN 330 and the second downsampled projected image data (P_RES2) associated with RES2 to generate V2_OUT 340. Figure 3 Operations 315, 325, and 335 in the context of... Figure 2B The corresponding operations in 250-270 are related.

[0048] Furthermore, during stage 4 (304), upsampling (see 345) can be performed to transform V2_OUT 340 into V3_IN 350 associated with the higher resolution RES3. Additionally, during stage 5 (305), resolution refinement can be performed for N3 iterations based on V3_IN 350 and the third downsampled projected image data (P_RES3) associated with RES3 (see 355). The following will use... Figure 6 This explains how the original resolution P_RESn is downsampled to generate downsampled projected image data (P_RES1, P_RES2, P_RES3). The final / output volumetric image data can be denoted as V( k +1)_OUT, in this case for K +1=3 is labeled as V3_OUT 360. Therefore, for K =2, a total of 2 can be executed. K +1 = 5 stages (including) Figure 3 In K =2 upsampling stages and K(+1 = 3 resolution refinement stages) to generate the output.

[0049] In one example, RES3 = 512x512x256 voxel grid, which has eight times more voxels than RES2 = 256x256x128 voxel grid. Furthermore, RES2 has eight times more voxels than RES1 = 128x128x64 grid. Using hivx = the total number of high-resolution voxels, N1 = 10 iterations can be performed during the first stage 301 to process a total relative voxel count of 10 / 8 / 8 hivx. Subsequently, N2 = 5 iterations can be performed during the third stage 303 to process 5 / 8 hivx. Finally, N3 = 5 iterations can be performed during the fifth stage 305 to process 5 hivx. Compared to performing only 10 iterations in RES3, Figure 3 The multi-stage image reconstruction process in the middle uses two upsampling stages ( K With a time complexity of 2, this can result in a theoretical time saving of 1 - (10 / 8 / 8 + 5 / 8 + 5) / 10 = 42.2%. Even when only one upsampling stage is implemented ( K When =1), a theoretical time saving of 1-(10 / 8+5) / 10=37.5% can also be achieved.

[0050] Using the examples of this disclosure, iterative reconstruction processes can be accelerated, thereby reducing the time required to acquire high-quality images, which is important in time-sensitive applications such as medical diagnostics. Performing iterative reconstruction at lower resolution levels also reduces computational requirements, making the process more efficient and feasible on standard hardware. The accelerated iterative reconstruction process also improves the efficiency of clinical workflows, enabling more timely decision-making.

[0051] In this disclosure, it should be understood that V k _OUT and V( k +1)_IN can represent different resolution levels, but... Figures 1-15 The same physical volume is represented in the two. For example, in Figure 2B and Figure 3 In this context, the external dimensions of V1_OUT 130 / 320 can be the same as the external dimensions of V2_IN 150 / 330. Figure 3In this example, the external dimensions of V2_OUT 340 can be the same as the external dimensions of V3_IN 350, and so on. Examples of this disclosure can be implemented using any suitable space-filled representation in the spatial domain, such as cubic voxels, cuboids, triangular prisms, hexagonal prisms, truncated octahedrons, rhombic dodecahedrons, etc. Upsampling according to examples of this disclosure can be performed at (multiple) arbitrary dimensions and / or directions. For example, cuboid voxels can be upsampled to any other mesh in each direction with any suitable step size. The upsampling process can vary for each upsampling level, allowing for different resolutions and levels of detail to meet the needs of different applications. The following uses... Figures 5-11 Provide a detailed example.

[0052] Example radiotherapy system

[0053] Based on the examples of this disclosure, any suitable computer system can be configured to implement the first and second aspects of this disclosure. Two examples will be discussed below. In the first example, the computer system (see...) Figure 4 The 470 in the image can be configured to perform image reconstruction using upsampling during the pre-treatment phase of radiotherapy, based on the first and / or second aspects, for purposes such as diagnosis and treatment planning. High-quality reconstructed images are important for segmentation, which identifies and delineates the target tumor and surrounding healthy tissue. Based on the segmentation, an effective treatment plan can be developed to deliver radiation doses to the tumor while protecting healthy tissue.

[0054] In the second example, the computer system (see...) Figure 5 The 470 in the diagram can be configured to perform image reconstruction using upsampling based on the first and / or second aspects during the treatment phase of radiotherapy. In practice, real-time or near-real-time volumetric image data (i.e., reconstructed images) allows clinicians to monitor and adjust treatment delivery based on any detected changes in the patient's anatomy, tumor location, and size. The output volumetric image data can also be used for patient localization and target structure tracking during treatment, thereby improving the accuracy and effectiveness of radiation delivery.

[0055] (a) Pre-treatment phase

[0056] Figure 4 This is a schematic diagram illustrating an example radiotherapy system 400, which includes a computer system 470 for performing image reconstruction during the pre-treatment phase of radiotherapy. Depending on the desired implementation, system 400 may include, in addition to... Figure 4Additional and / or alternative components not shown. In this example, the radiotherapy system 400 may include an imaging system 410 for acquiring raw resolution projected image data 105; a control system 460 for controlling the operation of the imaging system 410; and a computer system 470 for performing image reconstruction according to an example of this disclosure. A display device 480 may be communicatively coupled to the computer system 470 to display a user interface (UI) view associated with the volumetric image data generated by the computer system 470. The imaging system 410 may include a rack 411 having an opening 412 and a patient support 413 for supporting a patient 420 requiring radiotherapy.

[0057] Imaging system 410 can implement any suitable imaging modality for image data acquisition, such as computed tomography (CT), positron emission tomography (PET), single-photon emission computed tomography (SPECT), magnetic resonance imaging (MRI), magnetic resonance imaging (MRT), or any combination thereof. For example, when using CT, projection image data 105 (e.g., planning a CT scan) may include a series of 4D projection images or slices (e.g., CT slices), each slice representing a cross-sectional view of a patient's anatomy. For treatment planning, projection image data 105 may include 4D volumetric CT data, which (sometimes combined with 5D CT) is used to estimate the range of motion of (multiple) target structures. For example, alternatively or additionally, spectral CT data (e.g., dual-energy CT (DECT) and photon-counting CT) may be acquired to provide a variety of numbers of accesses during the planning phase.

[0058] exist Figure 4 In one example, gantry 411 has a ring-based configuration. In an alternative example, the gantry may have a C-arm configuration. Imaging system 410 may include an imaging or radiation source 430 (e.g., an X-ray source) for projecting an imaging beam 450 toward detector 431, which has a pixel detector opposite the radiation source 430. Control system 460 may be electrically coupled to gantry 411 to control the operation of gantry 411 using multiple control signals 461. Radiation source 430 may be configured to generate any suitable beam, such as a fan beam. During the imaging process, gantry 411 may rotate about opening 412 while radiation source 430 generates multiple X-ray beams 450 and guides them along projection lines toward patient 420 and detector 431. Detector 431 may measure X-ray absorption and generate a voltage proportional to the intensity of the incident X-rays. This voltage may be read and digitized to generate projected image data 105. Projected image data 105 may include image data acquired at different gantry angles.

[0059] According to an example of this disclosure, computer system 470 can acquire raw resolution projected image data (P_RESn) 105 from imaging system 410 and perform image reconstruction to generate output volumetric image data for display on display device 280. Figure 4 In this example, computer system 470 may include interface 471 for interacting with imaging system 410 to acquire projected image data 105; and processing core 472 for performing downsampling and K +1 resolution refinement stage; upsampling module / unit 473 for performing K upsampling stages; and UI module 474 for generating and displaying UI view 481 on display device 480. UI module 474 can also be configured to receive input data from user 490 (e.g., a clinician). Computer system 470 may include... Figure 4 Any alternative and / or additional components not shown in the diagram.

[0060] (b) Treatment phase

[0061] Figure 5 This is a schematic diagram illustrating an example radiotherapy system 500, which includes a computer system 470 that performs image reconstruction during the treatment phase of radiotherapy. Depending on the desired implementation, the radiotherapy system 500 may include, in addition to... Figure 5 Additional and / or alternative components not shown. In this example, the radiotherapy system 500 may include a treatment delivery machine 510 for delivering treatment to a patient 420; a control system 550 for controlling the operation of the machine 510; and a computer system 470 for performing image reconstruction according to an example of this disclosure.

[0062] The treatment delivery machine 510 may include a gantry 511 rotatable about an opening 512 and a patient support 513 (e.g., a treatment bed) for supporting the patient 420. Note that the gantry 511 may have a ring-based configuration (e.g., Figure 5(as shown in the diagram) or a C-arm configuration (not shown). The treatment delivery machine 510 may include a radiation source in the form of a linear accelerator (LINAC) 520 and an imager / detector in the form of a megaelectronvolt (MV) electron field imaging device (EPID) 521. The LINAC 520 may be configured to generate a treatment beam 530 and guide it toward an isocenter 514 through the PTV associated with the patient 420 during VMAT as the gantry 511 rotates through the treatment arc. In practice, the treatment beam 530 may be in a higher energy range, such as 1 MV or greater. Radiation therapy may be delivered as fractionated therapy, wherein the total radiation dose to be delivered to the tumor is divided into smaller “micro-dose”. This is to allow healthy cells to recover from radiation-induced damage between fractions, while tumor cells, which have poorer recovery capabilities, may accumulate damage.

[0063] The treatment delivery machine 510 may also include an onboard imaging system 540 to facilitate kilovolt (kV) imaging during the application of the MV treatment beam 530. Any suitable imaging modality, such as SE or DE CBCT, can be used. The imaging system 540 may include at least one kV imaging source 541 and at least one kV imager 542. Compared to the LINAC 520, the kV imaging source 541 may be able to generate imaging or diagnostic energy in the kV range. During treatment delivery, the control system 560 may configure the kV imaging source 541 to emit a kV imaging beam 543 and direct that kV imaging beam toward the imager 542, thereby generating projected image data 105 in the form of kV projected image data. Although described with reference to the MV LINAC 520 and the MV treatment beam 530, it should be understood that any additional or alternative treatment delivery techniques may be used. For example, a proton therapy machine including a kV imaging system may be used alternatively.

[0064] Figure 5 The computer system 470 can be communicatively coupled to the imaging system 510 to acquire raw resolution projected image data (P_RESn) 105 from the vehicle-mounted imaging system 540 via interface 471, and perform image reconstruction according to the examples of this disclosure. The computer system 470 may also include a processing core 472 for performing downsampling and... K +1 resolution refinement stage; upsampling module / unit 473 for performing K upsampling stages; and UI module 474 for generating and displaying UI view 481 on display device 480. UI module 474 can also be configured to receive input data from user 490 (e.g., clinician). Figures 4-5 The computer system 470 can be implemented using physical machines and / or virtual machines deployed in a cloud-based environment (i.e., not in the same physical location as the imaging systems 410 / 540).

[0065] Example resolution refinement stage

[0066] Figure 6 This is a flowchart of a detailed example process 600 of computer system 470 performing iterative image reconstruction using upsampling. Example process 600 may include one or more operations, functions, or actions illustrated by one or more boxes such as 605 to 695. Depending on the desired implementation, the boxes may be combined into fewer boxes, divided into more boxes, and / or deleted. Figure 6 The examples in [the document] can be used Figures 4-5 The computer system 470 in the middle is used to execute.

[0067] Figure 6 The examples in the text can be applied to any suitable K To achieve this, ≥1. For the k-th resolution refinement stage (see 640), and RES k =The input associated with the k-th resolution level=V k _IN (see 630) can be refined to be similar to RES through multiple iterations. k Associated output = V k _OUT (see 650). For the k-th upsampling stage (see 660), compared with the lower resolution RES k Related V k _OUT can be converted to a higher resolution RES ( k +1) associated upsampled V( k +1)_IN.

[0068] exist Figure 6 610 locations, based on the use Figure 4 or Figure 5 The raw resolution projected image data P_RESn acquired by the imaging system 410 / 540 (see [link]). Figure 6 605 and Figure 1 (105 in the middle), computer system 470 can perform downsampling to generate targeted k =1, …, K +1 downsampled projection image data (denoted as P_RES) k (See 620). For resolution level RES k This can be achieved by using any suitable downsampling factor (e.g. = Use any suitable base a (e.g., 2, 3, or 4, etc.) downsample P_RESn 605 to generate P_RES k 620. Here, P_RES kThis represents a downsampled (binned) version of the original resolution P_RESn 605. Any suitable downsampling method can be implemented, such as value binning, which is a relatively simple and fast approach. Based on P_RESn 605, a series of downsampling or binning operations can be performed to generate P_RES. k 620, for example from the highest resolution P_RES( K +1) to the lower resolution P_RES1. Note that RES1 ≤ RES k ≤RES ( k +1)≤RES( K +1)≤RESn, meaning the original resolution level is the highest possible resolution level. In P_RES( When ) = P_RESn, P_RESn can be used as input data in the iterative reconstruction phase.

[0069] exist Figure 6 At position 640, computer system 470 can achieve the resolution refinement stage. k To generate with RES k Related output data = V k _OUT (see 650). At 641, iterative reconstruction can be performed for Nk iterations based on the following input data: (a) downsampled projected image data (P_RES) k (a) 620 and (b) are labeled as V k The initial volumetric image data for _IN 630. For k=1, V1_IN 630 can be zero or a prior low-resolution volume generated using any suitable method. At 642, determine whether stopping conditions, such as gradient data (see...), are met. Figure 7 Whether 770-780 in the code has disappeared, or whether the maximum number of iterations that the user can configure has been reached, etc.

[0070] Any suitable iterative reconstruction algorithm can be implemented at box 640 to improve the quality of the reconstructed image through repeated thinning over multiple iterations. During iterative reconstruction, any suitable fast convergence optimization technique, such as subset or momentum, can be used. Furthermore, different regularization methods can be employed to promote convergence stability and noise suppression in the output volume. An example implementation of box 640 will be used below. Figure 7 The following explanation is provided. Additional implementation details can be found in U.S. Patent No. 11,173,324 entitled "Iterative reconstruction in image-guided radiation therapy," which is incorporated herein by reference in its entirety.

[0071] Figure 7 It is computer system 470 in Figure 6 The example implementation of the resolution refinement stage is shown in the flowchart (see 700). First, for example, based on the initial volume estimate (e.g., Figure 3 The output data can be initialized or retrieved using V1_IN 310 or output data from a previous upsampling stage. k The estimate for _OUT 650. Then, in Figure 7 At position 710 in the middle, V k _OUT 650 can withstand forward projection, whereby V k _OUT 650 is projected onto the same or similar detector geometry used during the actual acquisition of the projected image data 110, generating 2D simulated projection data 720.

[0072] exist Figure 7 At point 740, the simulated projection data 720 can be compared with the measured projection data 730 to determine the difference data representing any error between them (see 750). The measured projection data 730 can be... Figure 6 The original resolution projection data (P_RESn) or downsampled projection data (P_RES) in the data. k 620. Subsequently, at 760, the difference data 750 can be back-projected to generate iterative gradient data (3D) 770. Back-projection 760 is the inverse or approximately inverse process of forward projection 710.

[0073] At point 780, determine if the stopping condition is met, such as whether the gradient data 770 has vanished (i.e., convergence has been achieved) or whether the maximum number of iterations has been reached. If so, the iteration process ends. Otherwise, at point 790, V can be updated based on the gradient data 770. k _OUT 650 is used to reduce error, and the forward and backward projections continue iteratively. Depending on the desired implementation, V can be adjusted... k Regularization (see 735) is performed on _OUT 650 and / or gradient data 770 to stabilize (weighted) data fidelity and promote convergence of gradient data 770. Note that regularization can be applied to _OUT 650 and / or gradient data 770. Figure 7 Executed in different parts of the example.

[0074] Example upsampling phase

[0075] See you again Figure 6 At 660, computer system 470 can implement upsampling stage k based on upsampling factor (f2) to compare (a) with the lower resolution RES. k Related input data = V k _OUT 650 is converted to (b) with a higher resolution RES(k +1) Associated output data = V( k +1)_IN. Here, V( k +1)_IN 670 is used for subsequent resolution refinement stages. k +1 is the input data. Any suitable input data can be used. f2 For example, 2, 3, or 4. This will be used. Figure 8 Example implementation of Explanation Box 660 Figure 8 It is implemented by computer system 470. Figure 6 The flowchart shows an example process for the upsampling phase in the example.

[0076] exist Figure 8 At positions 801-803, box 660 may include input data V along one or more of the following axes. k Upsampling is performed at _OUT 650: (a) Z-axis = vertical axis representing the depth of the volumetric image data; (b) X-axis = horizontal axis representing the width of the volumetric image data; (c) Y-axis = sagittal axis representing the height of the volumetric image data. Upsampling along any combination of these spatial dimensions helps capture finer details and edges within the volumetric image data, thus more clearly distinguishing target structures from their surrounding healthy tissue. For example, upsampling along the Z-axis can be performed to improve resolution in the longitudinal direction, thereby improving the longitudinal (out-of-plane) resolution level. Upsampling along the X-axis improves resolution in the horizontal direction, thus providing more detail and sharpness across the width of the volume. Upsampling along the Y-axis improves resolution in the vertical direction, thus allowing for more precise visualization of vertically extending structures and details.

[0077] Depending on the desired implementation, at least one of the following upsampling methods can be performed: bilinear upsampling (see...) Figure 8 810 in the middle), triple or double triple upsampling (see 810 ...) Figure 8 (820 in the middle) and upsampling using an AI engine (see 820) Figure 8 (referring to 830 in the original text). In practice, bilinear upsampling may involve performing linear interpolation based on a weighted average of the four nearest neighbor pixels to compute a new pixel value. Cubic or bicubic upsampling may involve performing cubic interpolation based on even more nearest neighbor pixels, such as 16 adjacent pixels (i.e., four in each direction), to compute a new pixel value. Compared to bilinear upsampling, cubic upsampling provides fewer artifacts and smoother gradients / transitions, resulting in higher quality images. Any other and / or alternative upsampling techniques may be used.

[0078] exist Figure 8 In the example, AI Engine 830 may include multiple (X) processing layers (denoted as... arrive A hierarchical structure, such as an input layer, an output layer, and multiple (i.e., two or more) "hidden" layers between the input and output layers. Processing layer ( arrive ) and corresponding weight data ( arrive This is related to the AI ​​engine 830's learning of weight data during training. arrive ), to achieve this by using a lower resolution RES k Associated input V k _OUT 650 is converted to a higher resolution RES ( k +1) Associated output V( k +1)_IN 650 to perform upsampling.

[0079] As used herein, the term "AI engine" can refer to any suitable hardware and / or software component of a computer system capable of executing algorithms based on any suitable AI model(s). An "AI engine" can be a machine learning engine based on machine learning models(s), a deep learning engine based on deep learning models(s), and so on. Generally, deep learning is a subset of machine learning in which multi-layered neural networks can be used for feature extraction as well as pattern analysis and / or classification.

[0080] Any suitable AI model(s) can be used to implement the AI ​​Engine 830, such as convolutional neural networks, recurrent neural networks, deep belief networks, generative adversarial networks (GANs), (multiple) autoencoders, (multiple) variational autoencoders, long short-term memory architectures for tracking purposes, generative AI models, transformer networks, or any combination thereof. In practice, neural networks are typically formed using networks of processing units (called "neurons," "nodes," etc.) interconnected by connections (called "synapses," "weight data," etc.). The processing layers of a convolutional neural network can be convolutional layers, pooling layers, unpooling layers, rectified linear unit (ReLU) layers, fully connected layers, loss layers, activation layers, dropout layers, transposed convolutional layers, cascaded layers, attention layers, or any combination thereof. For example, convolutional neural networks can be implemented using any suitable architecture(s), such as UNet, LeNet, AlexNet, ResNet, VNet, DenseNet, OctNet, etc.

[0081] The AI ​​Engine 830 can be trained using any suitable method, such as supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, etc. For example, using supervised learning, the AI ​​Engine 830 can be trained on a dataset of labeled examples to learn the relationship between (a) input data and (b) output data. Any suitable training data can be used, such as synthetic data, real patient data, or a combination of both. The AI ​​Engine 830 can be trained using patient-specific training data or training data with a wide range of variations specific to possible patients. For example, patient-specific training strategies can address variations between patients and between tumors (e.g., tumor size, shape, location, movement).

[0082] Alternatively, using unsupervised learning, the AI ​​Engine 830 can be trained on a dataset of unlabeled examples to learn patterns and relationships in the data without any prior knowledge about the output labels. In semi-supervised learning, both labeled and unlabeled data can be used. Semi-supervised learning is very useful when there is a large amount of unlabeled data but labeling all of it is too costly or difficult. In reinforcement learning, the AI ​​Engine 830 can learn to perform image reconstruction through trial and error, where it is rewarded for actions that lead to the desired outcome and penalized for actions that lead to undesirable outcomes.

[0083] After upsampling, Figure 6 At position 680, the computer system 470 can achieve the resolution refinement stage. k +1, to generate a higher resolution RES ( k +1) Associated output data V( k +1)_OUT 690. Can be based on (a) and RES( k +1) associated V( k +1)_IN 670 (which is the output of the upsampling stage 660) and (b) downsampling projected image data P_RES( k +1) 675 to execute box 680. At 681, it can be based on the input data V( k +1)_IN 670 and P_RES( k +1) 675 for N( k +1) iterations are performed to perform iterative reconstruction. k = K And RES( k When +1)=RESn, box 1590 can be performed based on P_RESn (i.e., no downsampling is needed when the highest resolution level is the original resolution level). At 682, it is determined whether the stopping condition is met. Box 680 can be used Figure 7The example in the diagram above has already been explained and will not be repeated here for the sake of brevity.

[0084] exist Figure 6 At point 695, computer system 470 can determine... k Is it equal to K That is, all K Each upsampling stage and K Has the +1 resolution refinement stage been completed? If not, then the computer system 470 can be configured... k = k +1 and execute box 660 again. Otherwise, respond to confirmation. k = K , Figure 6 The example process in the example stopped at 600.

[0085] Iterative image reconstruction methods

[0086] Based on the expected outcome, execution can be performed. Figure 6 The multi-stage iterative reconstruction process in [the image] supports multiple image reconstruction techniques. The following will use [this method]. Figure 9 Let's discuss some examples. Figure 9 This is a flowchart of an example process by which a computer system performs iterative reconstruction using Algebraic Reconstruction Technique (ART) and / or Penalized Likelihood (PL) techniques. Figure 9 The example process 900 may include one or more operations, functions, or actions illustrated by one or more boxes such as those in Figures 910 to 960. Depending on the desired implementation, the boxes may be combined into fewer boxes, divided into more boxes, and / or deleted. Figure 9 The examples in [the document] can be used Figures 4-5 The computer system 470 in the middle is implemented, for example, by using (multiple) processing cores 472, upsampling modules 473, etc.

[0087] (a) Preprocessing

[0088] exist Figure 9 At point 910, based on the projected image data 110 (e.g., raw resolution projected data), the computer system 470 can perform preprocessing before image reconstruction. Any suitable preprocessing operation(s) can be performed at point 910, such as defect correction, scattering correction, nonlinear correction, beam hardening correction, or any combination thereof. Preprocessing can be performed on a projection-based basis while data acquisition is still running during scanning.

[0089] exist Figure 9 At point 911, hardware defect correction can involve identifying and correcting the detector system (e.g., Figure 4 Detector 431 or Figure 5Any defects in detector (542) (or any of the multiple defects). For example, defects (e.g., dead pixels or regions with inconsistent response) may cause artifacts in the projected image data 110 and the resulting volumetric image data. Figure 9 At position 912, scattering correction can be performed to mitigate the effects of scattered radiation from the projected image data 110, thereby enhancing its sharpness and contrast. Scattered radiation that occurs when X-rays deviate from their original path can cause image blurring and reduced contrast.

[0090] exist Figure 9 At position 913, nonlinear correction can be performed to compensate for the nonlinear response of the detector system. This is because the detector may not respond linearly to different radiation intensities, leading to distortion in the projected image data 110. Correcting these nonlinearities in the projected image data 112 ensures a more accurate representation of the true distribution of radiation. Figure 9 At position 914, beam hardening correction can be performed to compensate for the effects of beam hardening, where, for multicolor X-ray sources, low-energy X-rays are absorbed more than high-energy X-rays as X-rays pass through an object. By applying beam hardening correction, the projected image data 110 can be adjusted to account for different energy absorption, thereby improving image quality. In practice, attenuation does not necessarily decrease with increasing energy; for example, when considering the k-edge, the attenuation coefficient may suddenly increase with a slight increase in energy. Furthermore, note that not all sources used for transmission imaging need to be multicolor.

[0091] (b) Iterative reconstruction

[0092] exist Figure 9 At position 930, computer system 470 can perform iterative reconstruction according to a first method of the form ART. In practice, ART can typically refer to an iterative reconstruction technique that solves a system of linear equations derived from the projected data and iteratively updates the values. For example, the volumetric image data estimate can be updated voxel-by-voxel in the reconstructed space using contributions from each projection. Figure 6 The example in the text can perform K upsampling stages based on normalized projected image data 920 and K +1 resolution refinement stage (see 931-932) to generate output ART volume 933.

[0093] Additional or alternative land, in Figure 9 At position 950, the computer system 470 can perform iterative reconstruction according to a second method of the form PL. In practice, PL is a method that maximizes the likelihood function using an added penalty term to achieve desired properties (e.g., smoothness) in the reconstructed image. In practice, PL can also be referred to as model-based iterative reconstruction (MBIR) or statistical iterative reconstruction (SIR). Figure 6 The example in the text can perform K upsampling stages based on the denormalized projected image data 940. K +1 resolution refinement stage (see 951-952) to generate output PL volume 953. ART volume image data 933 can be used to initialize this process at box 950.

[0094] Note that normalized projected image data 920 can be used at box 930 because ART typically operates in the attenuation value space, i.e., the same space as the ART volumetric image data 933. This simplifies the reconstruction process but may not fully utilize the statistical properties of the data. In contrast, denormalized projected image data 940 can be used at box 950 because PL typically operates in the intensity space. In this case, denormalization can be performed, for example, using a Poisson distribution model, to more accurately model the statistical properties of the acquisition system (e.g., ...). Figures 4-5 ).

[0095] (c) Post-processing

[0096] exist Figure 9 At box 960, computer system 470 can perform post-processing based on ART volumetric image data 933 and / or PL volumetric image data 953 to generate final volumetric image data 970. Any suitable post-processing operation(s) can be performed at box 960, such as cropping, Huntsfield unit (HU) mapping, ring suppression, noise reduction, contrast enhancement, etc. Cropping (see box 961) can be performed to remove any unnecessary or irrelevant portions of the image, thereby focusing on preferred regions of interest and reducing the amount of data stored and processed.

[0097] HU mapping (see 962) can be performed to transform the raw reconstructed data into normalized Huntsfield units, which are used to quantify the radiation density of tissue. In practice, CBCT imaging systems used for radiotherapy can also produce electron density (ED) values, not just HU values. Ring suppression (see 963) can be performed to reduce or eliminate ring artifacts that may occur due to defects in the detector system or inconsistencies in the data acquisition process. Noise reduction (see 964) can be used to reduce image noise within the reconstructed volume. Contrast enhancement (see 965) allows visualization of different absorption structures in a single view, even in cases of misfit windows / levels. One or more of these post-processing steps can be implemented to enhance usability and accuracy, resulting in better clinical outcomes.

[0098] Example UI view

[0099] As used herein, the term “UI” or “UI view” can generally refer to a set of UI elements that can be generated and displayed on a display device. The term “UI element” can generally refer to graphic (i.e., visual) and / or textual elements that can be displayed on a display device, such as shapes (e.g., circles, rectangles, ovals, polygons, lines, etc.), windows, modals, panels or panes, buttons, checkboxes, menus, dropdowns, editable grids, sections, sidebars, sliders, text boxes, text blocks, toggle switches (on / off buttons), or any combination thereof. UI views can be displayed side-by-side or nested within each other to create more complex layouts. The term “interaction” can generally refer to a series of actions by which a user interacts with a UI view, such as viewing, clicking, swiping, changing the orientation and / or size associated with the content of a UI view, annotating, etc. The term “display device” (see, for example, [link to relevant documentation]) can also refer to a set of UI elements that can be generated and displayed on a display device. Figures 4-5 The 480 in this context can typically refer to any suitable hardware component used to present visual information to the user, such as a display, touchscreen, etc.

[0100] (a) First UI view

[0101] Figure 10A This is a schematic diagram illustrating a first example UI view 1000 of a user 490 interacting with first volumetric image data. In this example, a computer system 470 (e.g., using a UI module 474) can generate the first UI view 1000 to include UI elements (see 1010) for the user 490 to configure various settings associated with iterative image reconstruction utilizing upsampled data. At 1011, a target resolution can be configured, for example, RES2 = 518x518x386. At 1012, the number of upsampling stages(s) can be configured, for example... K =1. Will be used Figure 2B In K To explain the case where =2 Figure 10A .

[0102] exist Figure 10A At positions 1020-1030, computer system 470 can generate first UI view 1000 to include... K +1 Multiple status indicators associated with each resolution refinement stage. The first status indicator 1020 can be generated and dynamically updated to indicate... Figure 2B The progress (e.g., 100%) of the first resolution refinement stage 250 in generating V1_OUT 130 can be recorded. A second status indicator 1030 can be generated and dynamically updated to indicate the progress of this first resolution refinement stage 250. Figure 2BThe second resolution refinement stage 270 of the generated V2_OUT 170 shows the reconstruction progress (e.g., 62%). Any suitable status indicator can be generated and displayed, such as progress bars, charts, graphs, gauges, tabs with built-in progress bars, and any other type of visual feedback.

[0103] Once a low-resolution V1_OUT 130 is generated, the computer system 470 (e.g., using UI module 474) can generate and display a first UI view 1000 on a display device 480 for the user 490 to interact with V1_OUT 130. Figure 10A At 1040, the first UI view 1000 may show any suitable orientation or perspective(s) of V1_OUT 130, such as a sagittal view. Other example views include cross-sectional views, front views, 3D views, etc. Furthermore, at 1041, the first UI view 1000 may specify at least one of the following: (a) RES1 = 256x256x193; and (b) the total number of iterations performed, i.e., N1 = 10 iterations associated with RES1. The lower resolution V1_OUT 130 may be presented to the user as a preview before generating the higher resolution V2_OUT 170.

[0104] (b) Second UI view

[0105] Figure 10B This is a schematic diagram illustrating a second example UI view 1001 where user 490 interacts with output volumetric image data. Once a higher resolution V2_OUT 170 is generated, computer system 470 (e.g., using UI module 474) can generate and display the second UI view 1001 on display device 480 for user 490 to interact with V2_OUT 170. Figure 10B At 1050, the second UI view 1001 may show any suitable orientation(s) or view(s) of V2_OUT 170, such as a sagittal view. Furthermore, at 1051, the second UI view 1001 may specify at least one of the following: (a) RES2 = 518x518x386; and (b) the total number of iterations performed, i.e., N1 = 10 iterations associated with RES1 = 256x256x193 plus N2 = 5 iterations associated with RES2.

[0106] (c) Real-time switching

[0107] Figures 10A-10B The examples in can be extended to K Cases involving ≥2 upsampling stages. Figure 11 An example is shown in the image. Figure 11This is a schematic diagram illustrating multiple UI views used to display multiple sets of volumetric image data. (Reference) Figure 3 Example explanation Figure 11 Once a lower resolution V1_OUT 320 is generated, the computer system 470 can generate and display a first UI view 1110 on a display device 480 for the user 490 to interact with V1_OUT 320. The first UI view 1110 generated at time t1 can identify the number of iterations performed, i.e., N1 iterations associated with RES1.

[0108] Subsequently, once the intermediate resolution V2_OUT 340 is generated, the first UI view 1110 can be switched at time t2 (see 1115) to the second UI view 1120. Here, the computer system 470 can generate and display the second UI view 1120 for the user 490 to interact with V2_OUT 340. The second UI view 1120 can identify the number of iterations performed, namely the N1 iterations associated with RES1 plus the N2 iterations associated with RES2.

[0109] Finally, once the higher resolution V3_OUT 360 is generated, the second UI view 1120 can be switched at time t3 (see 1125) to the third UI view 1130. Here, the computer system 470 can generate and display the third UI view 1130 for the user 490 to interact with V3_OUT 360. The third UI view 1130 generated at time t3 can identify the number of iterations performed, namely N1 iterations associated with RES1 plus N2 iterations associated with RES2 plus N3 iterations associated with RES3.

[0110] Using the examples of this disclosure, higher resolution volumes (V0) can be generated. k While presenting +1)_OUT), a lower resolution volume (V) is displayed to the user at 490. k _OUT). User 490 can utilize the available lower resolution V. k _OUT and higher resolution V( k The time between +1)_OUT can be used to perform other tasks, such as scrolling to a preferred position in the reconstructed volume, adjusting the window / level, performing coarse adjustments (e.g., coarse matching tasks), etc. Coarse adjustments typically do not require high-quality volume data.

[0111] The second aspect: Reconstruction using upsampled multi-resolution images

[0112] According to the second aspect, a method and computer system for reconstructing multi-resolution images using upsampling are described. According to examples of this disclosure, at least one upsampling stage (i.e., K≥1) to achieve multi-resolution image reconstruction, reducing computational time associated with image reconstruction while taking into account truncation artifacts. In practice, truncation artifacts can occur when a small-diameter X-ray beam occupies an area smaller than the entire cross-section of the patient. These artifacts may be caused by the limited beam width, leading to incomplete coverage and partial data capture. Reducing such artifacts is desirable because they often impair the accuracy and quality of image reconstruction, especially in regions of interest that represent more important areas for diagnostic and treatment planning.

[0113] If will use Figures 12-15 As described, the examples in this disclosure provide a multi-resolution image reconstruction method that is multi-resolution in both the temporal and spatial domains. In the temporal domain, the resolution level of the estimated volumetric image data evolves over time and changes at different stages of the reconstruction process. This method saves computational time by performing some (if not most) computations at lower resolution levels and higher resolution computations to refine finer features. The final output (e.g.) Figures 12-13 V2_OUT 1270 or Figure 14 The V2_OUT in the output volumetric image data is multi-resolution in the spatial domain, meaning it includes voxels at different resolution levels. This allows a region of interest in the output volumetric image data to be generated at a higher resolution level where more detail is needed, while another less important region can be generated at a lower resolution level. Utilizing upsampling to perform multi-resolution image reconstruction balances the need for high-quality imaging with practical constraints on processing time and computational power.

[0114] As used herein, the term "multi-resolution image reconstruction" can generally refer to an image reconstruction process used to generate multi-resolution volumetric image data. The term "multi-resolution volumetric image data" can generally refer to volumetric image data associated with multiple resolution levels. The term "multi-resolution iterative image reconstruction" can generally refer to an image reconstruction process performed iteratively to generate multi-resolution volumetric image data. The following will use... Figure 12 Explain the example of the second aspect. Figure 12 This is a flowchart of an example process (see 1200) performed by computer system 470 for radiotherapy, utilizing upsampled multi-resolution image reconstruction. Example process 1200 may include one or more operations, functions, or actions illustrated by one or more boxes such as 1210 to 1270. Depending on the desired implementation, the individual boxes may be combined into fewer boxes, divided into more boxes, and / or deleted. Any suitable computer system can be configured to perform various examples of this disclosure, such as... Figures 4-5 Computer system 470.

[0115] exist Figure 12At point 1210, computer system 470 can acquire first volumetric image data (V1_OUT) associated with a first resolution level (RES1). Depending on the desired implementation, block 1210 may include computer system 470, which, based on projected image data associated with a target structure of patient 420 requiring radiotherapy (see [link to documentation]),... Figure 1 V1_OUT1210 is generated from 110 in the middle.

[0116] exist Figure 12 At position 1220, the computer system 470 can upsample V1_OUT 1210 associated with RES1 to generate second volumetric image data (V2_IN) 1230 associated with the second higher resolution (RES2), i.e., RES2 > RES1. Box 1220 can be used... Figure 8 Various examples, such as bilinear upsampling, cubic or bicubic upsampling, and AI engines for upsampling, are used to implement this.

[0117] exist Figure 12 At position 1240, computer system 470 can acquire mask image data 1250 (denoted as MASK), which identifies at least one region of interest 1251 associated with the target structure of patient 420. As used herein, the term "region of interest" or "ROI" can generally refer to a specific region requiring higher resolution image reconstruction. Depending on the desired implementation, each region of interest can be defined in projection space and / or volume space. The term "mask image data" can refer to any suitable data used to distinguish the region of interest from one or more other regions that do not require higher resolution image reconstruction. Figure 13 As described, box 1240 may include computer system 470, which generates mask 1250, for example, based on V1_OUT 1210, V2_IN 1230, projected image data 110, additional data (e.g., segmentation data), or any combination thereof.

[0118] For example, mask 1250 can identify one or more regions of interest representing areas important for medical diagnosis and treatment planning. Each region of interest can be of any suitable size and shape, such as a 2D or 3D shape. Region of interest 1251 can be a region with highly high spatial frequency (HF) content, such as at the boundary between an organ and bone. In practice, region of interest 1251 can include tumor margins (i.e., the edges or boundaries of the tumor) to accurately distinguish the tumor from its surrounding healthy tissue; small areas where the tumor may have spread into adjacent tissues; vascular structures for understanding the tumor's blood supply; enlarged or affected lymph nodes near the tumor, etc.

[0119] exist Figure 12At position 1260, computer system 470 can generate multi-resolution volumetric image data (denoted as V2_OUT 1270) based on V2_IN 1230 and mask 1250. Here, the term "multi-resolution volumetric image data" can generally refer to volumetric image data with multiple resolution levels. Figure 12 In the example, V2_OUT 1270 may include a first voxel associated with RES2 outside the region of interest (see 1271). V2_OUT 1270 may also include a second voxel associated with a higher third resolution level (RES3) within the region of interest (see 1272), where RES3>RES2>RES1.

[0120] Example upsampling phase

[0121] Depending on the desired implementation, computer system 470 can be configured to perform one or more upsampling stages (i.e., K ≥1). Will use Figure 13 explain K In the case where =1 (i.e., one upsampling stage), Figure 13 This is a schematic diagram illustrating a first example (see 1300) of multi-resolution image reconstruction using an upsampling stage. It will use... Figure 14 explain K In the case of =2 (i.e., two upsampling stages), this Figure 14 This is a second example of multi-resolution image reconstruction utilizing multiple upsampling stages (see 1400). Using Figure 6 Box 660 and Figure 8 The implementation details described in boxes 810-830 also apply here.

[0122] (a) An upsampling stage ( K =1)

[0123] exist Figure 13In the example, iterative reconstruction (see 1320) can be performed based on the initial volumetric image data (V1_IN) 1310 to generate V1_OUT 1210 associated with RES1. During the upsampling phase (see 1330), V1_OUT 1210 can be upsampled to V2_IN 1230 associated with RES2. During the resolution refinement phase (see 1340), multi-resolution iterative reconstruction can be performed based on V2_IN 1230 and mask 1250 to generate multi-resolution volumetric image data (V2_OUT) 1270. Mask 1250 can be generated based on V1_OUT 1210 and / or V2_IN 1230 to identify the region of interest 1251. This method allows the generation of V2_OUT 1270, which includes a first voxel 1271 associated with RES2 outside the region of interest 1251 and a second voxel 1272 associated with a higher resolution RES3 within the region of interest 1251. As desired, the iterative reconstruction process at box 1320 can be performed based on the first downsampled projected image data (P_RES1) associated with RES1. Similarly, the iterative reconstruction process at box 1340 can be performed based on the second downsampled projected image data (P_RES2) associated with RES2. (The last sentence appears to be incomplete and possibly refers to a different implementation.) Figure 6 As described in box 610, the original resolution P_RESn from imaging system 410 / 540 can be adjusted by using any suitable downsampling factor (f1) (see [link to image 410 / 540]). Figure 6 The sampled image data 620, which includes P_RES1 and P_RES2, is generated by downsampling (605) in the sampled image data.

[0124] (b) Multiple upsampling stages ( K =2)

[0125] exist Figure 14In the example, N1 iterations of iterative reconstruction (see 1401) can be performed based on the initial volume image data (V1_IN) 1410 and P_RES1 to generate V1_OUT 1420 associated with RES1. During the first upsampling stage (see 1402), V1_OUT 1420 associated with the lower resolution RES1 can be upsampled to V2_IN 1430 associated with the intermediate resolution RES2, where RES2 > RES1. During the resolution refinement stage (see 1403), N2 iterations of iterative reconstruction can be performed based on V2_IN 1430 and P_RES2 to refine V2_IN 1430 to V2_OUT 1440. During the second upsampling stage (see 1404), V2_OUT 1440 associated with the intermediate resolution RES2 can be upsampled to generate V3_IN 1450 associated with the higher resolution RES3.

[0126] During the final resolution refinement stage (see 1405), iterative reconstruction can be performed for N3 iterations based on V3_IN 1450, P_RES3 associated with RES3, and the mask 1250 to generate the multi-resolution volume image data (V3_OUT) 1460. Depending on the desired implementation, the mask 1250 can identify multiple regions of interest, such as the first region of interest 1251 and the second region of interest 1252. The mask 1250 can be generated based on the volume image data 1420 / 1440 / 1450, the projection image data 110, additional data (e.g., segmentation data), or any combination thereof.

[0127] To provide more details and finer granularity within the regions of interest 1251 - 1252, V3_OUT 1460 includes voxels reconstructed at different resolution levels within the same spatial domain. The first voxels 1461 located outside the regions of interest 1251 - 1252 are associated with RES3, while the second voxels 1462 - 1463 located within the regions of interest 1251 - 1252 are associated with RES4 > RES3. Note that the first voxel 1462 can have a different resolution level compared to the second voxel 1463. It should be understood that V k _OUT and V( k +1)_IN can represent different resolution levels, but in Figures 12-15 represent the same physical space of the same volume. For example, in Figures 13-14 the outer dimensions of V1_OUT 1210 / 1420 may be the same as the outer dimensions of V2_IN 1230 / 1430. In Figure 14 the outer dimensions of V2_OUT 1440 may be the same as the outer dimensions of V3_IN 1450, and so on.

[0128] Example Detailed Process

[0129] The following will use Figure 15 Describe the example implementation. Figure 15 This is a flowchart of a detailed example process 1500 of a computer system performing multi-resolution image reconstruction using upsampling. Example process 1500 may include one or more operations, functions, or actions illustrated by one or more boxes such as 1505-1596. Depending on the desired implementation, the boxes may be combined into fewer boxes, divided into more boxes, and / or deleted. Figure 15 The examples in [the document] can be used Figures 4-5 The computer system 470 in the middle is used to execute.

[0130] (a) First volumetric image data

[0131] Figure 15 1505-1570 in the middle corresponds to Figure 6 For the sake of brevity, the descriptions of boxes 605-670 in the text will not be repeated below. For Figure 15 V1_OUT 1550 can be generated by performing an iterative reconstruction of N1 iterations. Alternatively, V1_OUT 1210 can be generated by performing one of the following: Filtered Back Projection (FBP) algorithm, Feldkamp-David-Kress (FDK) algorithm and Defrise-Clark algorithm, iterative reconstruction with metal artifact suppression, four-dimensional (4D) image reconstruction, image reconstruction using an AI engine, etc.

[0132] In practice, FBP can involve applying filters to the projected image data 620 before back-projecting it onto the image plane. The FDK and Defrise-Clack algorithms extend the FBP algorithm to account for the geometry of a cone X-ray beam. Iterative reconstruction with metal artifact suppression can involve reducing artifacts caused by metal implants within the patient's body. 4D reconstruction can be implemented to extend the concept of 3D image reconstruction by incorporating a fourth dimension (e.g., time), allowing the reconstruction of dynamic processes, such as time-varying respiratory or cardiac motion. This is particularly valuable in radiotherapy, where understanding the movement of the tumor relative to surrounding tissues helps improve treatment planning and delivery. Alternatively, an AI engine can be trained to generate V1_OUT 1550 based on any suitable projected image data.

[0133] The description of the FDK algorithm can be found in the following publications: "Practical cone-beam algorithm" by Feldkamp, ​​LA, Davis, LC, Kress, JW, et al., published in J. Opt. Soc. Am. 1(6) (1984). The description of the Defrise-Clack algorithm can be found in the following publications: "Cone-beam reconstruction by the use of Radon transform intermediate functions" by R. Clack, M. Defrise, et al., published in J. Opt. Soc. Am 11(2) (February 1994); and "Direct Reconstruction of Cone-Beam Data Acquired with a Vertex Path Containing a Circle" by Noo. M. Defrise, R. Clack, et al., published in IEEE Transactions on Image Processing 7(6) (June 1998). These publications are incorporated herein by reference.

[0134] Descriptions of metal artifact suppression can be found in US patent applications US20230095240 and US20230100798, which are incorporated herein by reference. Descriptions of 4D reconstruction can be found in Star-Lack, J. et al., “A modified McKinnon-Bates (MKB) algorithm for improved 4D cone-beam computedtomography (CBCT) of the lung”, published in Medical Physics (45(8), 3783-3799, 2018); and Yoon, S. et al., “A motionestimation and compensation algorithm for 4D CBCT of the abdomen”, published at SPIE, 15th International Conference on Full 3D Image Reconstruction in Radiation and Nuclear Medicine (Vol. 11072, pp. 59-63, May 2019). Further implementation details regarding iterative reconstruction can be found in the following publication: “Acuros CTS: A fast, linear Boltzmann transportequation solver for computed tomography scatter – Part II: System modeling, scatter correction, and optimization”, by Wang, A. et al., in Medical Physics (45(5), 1914–1925, 2018). These publications are incorporated herein by reference.

[0135] (b) Upsampling

[0136] exist Figure 15 At position 1560, computer system 470 can use any suitable upsampling factor ( f2 ), for lower resolution RES k Related input data V k Upsampling is performed on _OUT 1550 to generate a higher resolution RES ( k +1) Associated output data V( k +1)_IN 1570. Box 1560 can be used. Figure 8The examples explained in the text will be used to implement this, and for the sake of brevity, the details will not be repeated here. For example, box 1560 could involve computer system 470 upsampling the input data along the Z-axis (vertical axis), X-axis (horizontal axis), Y-axis (vertical axis), or any combination thereof. Upsampling along the Z-axis improves the vertical resolution, thereby improving the vertical (out-of-plane) resolution level. Bilinear upsampling can be implemented (see...). Figure 8 810 in the middle), triple / double triple upsampling (see 810 ...) Figure 8 820 in the middle), AI-based upsampling engine (see 820), Figure 8 (830 in the middle) or any combination thereof.

[0137] (c) Mask image data generation

[0138] exist Figure 15 At position 1580, computer system 470 can generate mask image data (mask 1250 / 1585) that identifies regions of interest (ROIs) associated with the patient's target structures. In one example, mask 1250 may be in the form of a binary image used to label each ROI 1251. In this case, a pixel or voxel within a ROI 1251 is assigned a flag value of 1 (i.e., true), and all other pixels or voxels are assigned a value of 0 (i.e., false). In another example, mask 1250 may be a labeled image that uses labels to distinguish multiple ROIs within the image, such as "ROI1" for a first ROI, "ROI2" for a second ROI, and so on. Mask 1250 can be used during image reconstruction to achieve higher resolution within the ROIs 1251. Mask 1250 may also include data for identifying the shape and / or contour associated with the ROIs 1251.

[0139] Depending on the desired implementation, the mask 1250 can be generated based on volumetric image data 1550 / 1570 in the 3D volumetric domain and / or projected image data 1505 / 1520 in the 2D projection domain. Additionally or alternatively, the computer system 470 can acquire and process any suitable external data to generate the mask 1250. The external data may include any suitable segmentation data relating to the contour surface(s) and / or edges(s) associated with the target structure. For example, segmentation data can be used to trace the perimeter of a tumor to approximate its boundaries. The external data may also include user input data that can guide the identification of the region of interest 1251 to generate the mask 1250. For example, a UI view can be generated and displayed on a display device 480 for the user 490 to select the region of interest(s)(s) within any suitable image data.

[0140] The mask 1250 can be generated using any suitable method, such as based on a range of values ​​associated with the region of interest 1251, a relative value difference associated with the region of interest 1251, a heuristic method for identifying the region of interest 1251, or any combination thereof. In the first example, the computer system 470 identifies the region of interest 1251 based on a range of values, such as by analyzing V... k _OUT 1550 Pixel or voxel intensity values. If the region of interest 1251 is associated with different intensity values, a threshold can be set to include values ​​within a specific range, thereby creating a mask 1250 in which the region of interest 1251 is marked or labeled.

[0141] In the second example, the computer system 470 can identify the region of interest 1251 based on relative value differences, which can identify edges or boundaries where intensity values ​​change. This helps isolate the region of interest 1251, which stands out from the background based on its contrast with adjacent regions. In the third example, the computer system 470 can determine the region of interest 1251 by applying predefined rules and / or algorithms, such as by probing specific patterns, shapes, or sizes within the volumetric image data 1550 / 1570 and / or projected image data 1505 / 1520, to achieve a heuristic approach. Depending on the desired implementation, the mask 1250 can be updatable or modifiable during subsequent multi-resolution iterative image reconstruction processes. For example, if K upsampling stages are configured, and the voxel values ​​within a parent voxel remain the same in the next iteration (i.e., the standard deviation between the child voxel and its parent voxel is low), the parent voxel can be marked as static in a subsequent thinning step, thereby updating the mask 1250.

[0142] (d) Multi-resolution volumetric image data

[0143] exist Figure 15 At point 1590, the response to determination k < K Computer system 470 can perform iterative reconstruction to generate output data = V( k +1)_OUT 1595. Iterative reconstruction can be based on... Figure 7 The example in [reference] can be used for execution. At 1591, it can be based on the input data V( k +1)_IN 1570 for N k The next iteration performs iterative reconstruction. At 1592, it determines whether stopping conditions have been met, such as whether the gradient data has disappeared or whether the user-configurable maximum number of iterations has been reached. See also 1596.

[0144] Response to determination k = K (Instructions for all)K (The upsampling phase has been completed), computer system 470 can be based on mask 1250 and V (…). k +1)_IN 1570 to perform multi-resolution iterative reconstruction to generate output data = multi-resolution volumetric image data V( k +1)_OUT 1595. Based on the desired implementation, it can be based on the downsampled projected image data P_RES( k +1) 1575 to execute box 1590. In RES( K When +1)=RESn, box 1590 can be performed based on P_RESn (i.e., no downsampling is required).

[0145] for Figure 13 In the example K =1, outputting multi-resolution volumetric image data as V2_OUT 1270. For Figure 14 In the example K =2, output is V3_OUT 1460. Use Figure 7 In the example shown, computer system 470 may apply mask 1250 during forward projection (see 710) and / or backward projection (see 760). Here, mask 1250 can be applied to identify region of interest 1251 and selectively enhance the resolution level of pixels or voxels associated with region of interest 1251. For example, a higher resolution level within region of interest 1251 can be achieved by initializing a complete field of view (FOV) for both voxels and pixels at a lower resolution level, then applying a finer resolution level in the projection domain, and iteratively solving an iterative reconstruction algorithm to refine the resolution level in the image domain. Note that mask 1250 is updatable during multi-resolution iterative image reconstruction.

[0146] When mask 1250 is Figure 14When identifying multiple regions of interest 1251-1252 in the example, the improved resolution level in regions of interest 1251-1252 can be implemented serially or in parallel. In serial execution, switching from lower to higher resolution in the sinusoidal domain and the image domain can be performed serially for each region of interest 1251-1252. In parallel execution, the resolution level for all regions of interest 1251-1252 can be refined simultaneously. Additional implementation details related to multiresolution iterative reconstruction can be found in U.S. Patent No. US10,517,543, entitled "Multiresolution iterative reconstruction for region of interest imaging in X-ray cone-beam computed tomography," which is incorporated herein by reference.

[0147] UI view

[0148] According to examples of this disclosure, computer system 470 can generate and display (multiple) UI views for user 490 to interact with various output volumetric image datasets. Figure 13 In the example, a first UI view can be generated and displayed on display device 480 for user 490 to interact with V1_OUT 1210 associated with RES1. Once multi-resolution iterative reconstruction based on mask 1250 is completed, a second UI view can be generated and displayed on display device 480 for user 490 to interact with multi-resolution volumetric image data = V2_OUT 1270, which includes a first voxel 1271 associated with RES2 and a second voxel 1272 associated with RES3.

[0149] Similarly, in Figure 14 In the example, a first UI view can be generated and displayed on display device 480 for user 490 to interact with V1_OUT 1420 associated with RES1. Subsequently, a second UI view can be generated and displayed on display device 480 for user 490 to interact with V2_OUT 1440 associated with RES2. Once multi-resolution iterative reconstruction based on mask 1250 is completed, a third UI view can be generated and displayed on display device 480 for user 490 to interact with multi-resolution volumetric image data V3_OUT 1460, which includes non-ROI voxels 1461 associated with RES3 and ROI-related voxels 1462-1463 associated with RES4.

[0150] Computer System

[0151] The examples described above can be implemented using hardware (including hardware logic circuitry), software or firmware, or a combination thereof. These examples can be implemented using any suitable computing device, computer system, etc. A computer system may include processor(s), memory(s), and physical NIC(s), which can communicate with each other via a communication bus, etc. A computer system may include a non-transitory computer-readable medium on which instructions or program code are stored, which, when executed by a processor, cause the processor to perform the processes described herein with reference to the accompanying drawings.

[0152] The techniques described above can be implemented in dedicated hardwired circuits, software and / or firmware combined with programmable circuits, or a combination thereof. Dedicated hardware or hardwired circuits can take, for example, the form of one or more accelerators (for accelerating computational tasks related to image reconstruction), application-specific integrated circuits (ASICs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), etc. The term "processor" should be understood broadly to include processing units, ASICs, logic units, or programmable gate arrays. The term "accelerator" can generally refer to any suitable hardware or other computing processing unit (e.g., high-performance computing processing unit) used to accelerate computational tasks, such as graphics processing units (GPUs), tensor processing units (TPUs), neural network processing units (NPUs), etc. Any alternative processor(s) architecture, such as a hybrid architecture (called an XPU), can be used, designed to handle a variety of workloads by combining different types of processing units, etc.

[0153] The foregoing detailed description illustrates various embodiments of the device and / or process using block diagrams, flowcharts, and / or examples. As long as such block diagrams, flowcharts, and / or examples contain one or more functions and / or operations, those skilled in the art will understand that each function and / or operation in such block diagrams, flowcharts, or examples can be implemented individually and / or in combination by various hardware, software, firmware, or any combination thereof.

[0154] Those skilled in the art will recognize that some aspects (in whole or in part) of the embodiments disclosed herein can be equivalently implemented in an integrated circuit, implemented as one or more computer programs running on one or more computers (e.g., as one or more programs running on one or more computing systems), implemented as one or more programs running on one or more processors (e.g., as one or more programs running on one or more microprocessors), implemented as firmware, or implemented as virtually any combination thereof, and that designing circuits and / or writing software and / or firmware code according to this disclosure is entirely within the skill of those skilled in the art.

[0155] Software used to implement the techniques described herein can be stored on a non-transient computer-readable storage medium and can be executed by one or more general-purpose or special-purpose programmable microprocessors. "Computer-readable storage medium" (as used herein) includes any mechanism that provides (i.e., stores and / or transmits) information in a machine-accessible form (e.g., a computer, network device, personal digital assistant (PDA), mobile device, manufacturing tool, any device having 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 disk or optical storage media, flash memory devices, etc.).

[0156] The accompanying drawings are merely illustrative, and the units or procedures shown in the drawings are not essential for implementing this disclosure. Those skilled in the art will understand that the units in the illustrated device may be arranged in the illustrated device as described, or alternatively located in one or more devices different from the illustrated device. The units in the described examples may be combined into a module or further divided into multiple sub-units.

[0157] Example Terms

[0158] The subject matter of the following clauses provides further aspects of these teachings (where it can be understood that any of these clauses may be suitably combined with one or more of the other clauses). Clause 2 may be combined with Clause 1, depending on the desired outcome; Clause 3 may be combined with Clauses 1 and / or 2; Clause 4 may be combined with one or more of Clauses 1-3; Clause 5 may be combined with one or more of Clauses 1-4; Clause 6 may be combined with one or more of Clauses 1-5; and Clause 7 may be combined with one or more of Clauses 1-6. This also applies to (a) Clause 8, which may be combined with one or more of Clauses 9-14, and (b) Clause 15, which may be combined with one or more of Clauses 16-21.

[0159] (a) First aspect (reconstruction using iterative images with upsampling)

[0160] Clause 1. A method for a computer system to perform iterative image reconstruction using upsampling for radiotherapy, the method comprising: acquiring projected image data associated with a target structure of a patient; performing iterative image reconstruction for a first number of iterations based on the projected image data to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; upsampling the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; and performing iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data associated with the second resolution level.

[0161] Clause 2. The method according to Clause 1, wherein upsampling the first volumetric image data comprises: generating second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0162] Clause 3. The method according to Clause 1 or Clause 2, wherein upsampling the first volumetric image data comprises: performing bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data.

[0163] Clause 4. The method according to one or more of Clauses 1-3, wherein upsampling the first volumetric image data comprises: applying an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0164] Clause 5. The method according to one or more of Clauses 1-4, wherein the method further comprises: upsampling the output volumetric image data to a third volumetric image data associated with a third resolution level higher than the second resolution level; and performing an iterative image reconstruction for a third number of iterations based on the third volumetric image data to generate final volumetric image data associated with the third resolution level.

[0165] Clause 6. The method according to one or more of Clauses 1-5, wherein performing iterative image reconstruction comprises: performing iterative image reconstruction based on projected image data to generate first volumetric image data, the projected image data being first downsampled projected image data associated with a first resolution level; and performing iterative image reconstruction based on second downsampled projected image data associated with a second resolution level to generate output volumetric image data.

[0166] Clause 7. The method according to one or more of Clauses 1-6, wherein the method further comprises at least one of: (a) generating and displaying a first user interface (UI) view on a display device to allow a user to interact with first volumetric image data, wherein the first UI view specifies at least one of: a first resolution level and a first number of iterations; and (b) generating and displaying a second UI view on a display device to allow a user to interact with output volumetric image data, wherein the second UI view specifies at least one of: a second resolution level and a total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

[0167] Clause 8. A non-transient computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform the following operations: acquire projection image data associated with a target structure of a patient; perform iterative image reconstruction for a first number of iterations based on the projection image data to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; upsample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; and perform iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data associated with the second resolution level.

[0168] Clause 9. The non-transient computer-readable medium according to Clause 8, wherein instructions for upsampling first volumetric image data cause the processor to: generate second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0169] Clause 10. A non-transient computer-readable medium as described in Clause 8 or Clause 9, wherein instructions for upsampling first volumetric image data cause a processor to perform bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data.

[0170] Clause 11. A non-transient computer-readable medium according to one or more of Clauses 8-10, wherein instructions for upsampling first volumetric image data cause a processor to apply an AI engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0171] Clause 12. A non-transient computer-readable medium according to one or more of Clauses 8-11, wherein the instructions further cause the processor to: upsample the output volumetric image data to a third volumetric image data associated with a third resolution level higher than the second resolution level; and perform an iterative image reconstruction for a third number of iterations based on the third volumetric image data to generate final volumetric image data associated with the third resolution level.

[0172] Clause 13. A non-transient computer-readable medium according to one or more of Clauses 8-12, wherein instructions for performing iterative image reconstruction cause a processor to: perform iterative image reconstruction based on projected image data to generate first volumetric image data, the projected image data being first downsampled projected image data associated with a first resolution level; and perform iterative image reconstruction based on second downsampled projected image data associated with a second resolution level to generate output volumetric image data.

[0173] Clause 14. A non-transient computer-readable medium according to one or more of Clauses 8-13, wherein the instructions further cause a processor to perform at least one of the following operations: (a) generating and displaying a first user interface (UI) view on a display device to allow a user to interact with first volumetric image data, wherein the first UI view specifies at least one of: a first resolution level and a first number of iterations; and (b) generating and displaying a second UI view on a display device to allow a user to interact with output volumetric image data, wherein the second UI view specifies at least one of: a second resolution level and a total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations. A computer system is also provided, according to a desired implementation, comprising a processor and a non-transient computer-readable medium according to one or more of Clauses 8-14.

[0174] Clause 15. An imaging system comprising: an imaging source and a detector for acquiring projected image data associated with a target structure of a patient; and a computer system configured to: perform an iterative image reconstruction for a first number of iterations based on the projected image data to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; upsample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; and perform an iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data associated with the second resolution level.

[0175] Clause 16. The imaging system of Clause 15, wherein the computer system is configured to upsample the first volumetric image data by generating second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0176] Clause 17. An imaging system according to Clause 15 or Clause 16, wherein the computer system is configured to upsample first volumetric image data by performing bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data.

[0177] Clause 18. An imaging system according to one or more of Clauses 15-17, wherein the computer system is configured to upsample the first volumetric image data by applying an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0178] Clause 19. An imaging system according to one or more of Clauses 15-18, wherein the computer system is further configured to: upsample output volumetric image data to third volumetric image data associated with a third resolution level higher than the second resolution level; and perform iterative image reconstruction for a third number of iterations based on the third volumetric image data to generate final volumetric image data associated with the third resolution level.

[0179] Clause 20. An imaging system according to one or more of Clauses 15-19, wherein the computer system is configured to perform iterative image reconstruction by: generating first volumetric image data based on projected image data, which is first downsampled projected image data associated with a first resolution level; and generating output volumetric image data based on second downsampled projected image data associated with a second resolution level.

[0180] Clause 21. An imaging system according to one or more of Clauses 15-20, wherein the computer system is further configured to perform at least one of the following operations: (a) generating and displaying a first user interface (UI) view on a display device to allow a user to interact with first volumetric image data, wherein the first UI view specifies at least one of the following: a first resolution level and a first number of iterations; (b) generating and displaying a second UI view on a display device to allow a user to interact with output volumetric image data, wherein the second UI view specifies at least one of the following: a second resolution level and a total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

[0181] Clause 22. A computer program including instructions that, when executed by a computer, cause the computer to perform the following steps: acquiring projected image data associated with a target structure; performing iterative image reconstruction for a first number of iterations based on the projected image data to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; upsampling the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; and performing iterative image reconstruction for a second number of iterations based on the second volumetric image data to generate output volumetric image data associated with the second resolution level.

[0182] Clause 23. The computer program according to Clause 22, wherein the instructions for upsampling the first volumetric image data cause the computer to generate second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0183] Clause 24. A computer program pursuant to Clause 22 or Clause 23, wherein instructions for upsampling first volumetric image data cause the computer to: perform bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data.

[0184] Clause 25. A computer program pursuant to one or more of Clauses 21-24, wherein instructions for upsampling first volumetric image data cause the computer to: apply an AI engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0185] Clause 26. A computer program pursuant to one or more of Clauses 21-25, wherein the instructions further cause the computer to: upsample the output volumetric image data to a third volumetric image data associated with a third resolution level higher than the second resolution level; and perform a third iteration of iterative image reconstruction based on the third volumetric image data to generate final volumetric image data associated with the third resolution level.

[0186] Clause 27. A computer program pursuant to one or more of Clauses 21-26, wherein instructions for performing iterative image reconstruction cause the computer to: perform iterative image reconstruction based on projected image data to generate first volumetric image data, the projected image data being first downsampled projected image data associated with a first resolution level; and perform iterative image reconstruction based on second downsampled projected image data associated with a second resolution level to generate output volumetric image data.

[0187] Clause 28. A computer program pursuant to one or more of Clauses 21-27, wherein the instructions further cause the computer to perform at least one of the following operations: (a) generating and displaying a first user interface (UI) view on a display device to allow a user to interact with first volumetric image data, wherein the first UI view specifies at least one of: a first resolution level and a first number of iterations; and (b) generating and displaying a second UI view on a display device to allow a user to interact with output volumetric image data, wherein the second UI view specifies at least one of: a second resolution level and a total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

[0188] (b) Second aspect (reconstruction using upsampled multi-resolution images)

[0189] Clause 1. A method for a computer system to perform multi-resolution image reconstruction using upsampling, the method comprising: acquiring first volumetric image data associated with a first resolution level, wherein the first volumetric image data is generated based on projection image data associated with a target structure of a patient; upsampling the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; acquiring mask image data that identifies regions of interest (ROIs) associated with the target structure; and generating multi-resolution volumetric image data based on the second volumetric image data and the mask image data, wherein the multi-resolution volumetric image data includes at least a first voxel associated with a second resolution level outside the ROI and a second voxel associated with a third resolution level within the ROI, wherein the third resolution level is higher than the second resolution level.

[0190] Clause 2. The method according to Clause 1, wherein upsampling the first volumetric image data comprises: generating second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0191] Clause 3. The method according to Clause 1 or Clause 2, wherein upsampling the first volumetric image data comprises at least one of: performing bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data; and applying an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0192] Clause 4. The method according to one or more of Clauses 1-3, wherein obtaining the first volumetric image data comprises: performing iterative image reconstruction based on the projected image data to generate the first volumetric image data.

[0193] Clause 5. The method according to one or more of Clauses 1-4, wherein acquiring mask image data comprises: acquiring mask image data generated based on one or more of the following input data: projected image data, first volume image data, second volume image data, and segmentation data associated with the region of interest.

[0194] Clause 6. The method according to one or more of Clauses 1-5, wherein acquiring mask image data includes: acquiring mask image data generated based on one or more of the following: a numerical range associated with a region of interest, a relative value difference associated with a region of interest, and a heuristic method specifying one or more rules for identifying the region of interest.

[0195] Clause 7. The method according to one or more of Clauses 1-6, wherein generating multi-resolution volumetric image data comprises: performing multi-resolution iterative image reconstruction based on second volumetric image data and mask image data, wherein the mask image data is updatable during multi-resolution iterative image reconstruction and is applied during at least one of the following operations: a forward projection operation and a back projection operation.

[0196] Clause 8. A non-transient computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the processor to perform the following operations: acquire first volumetric image data associated with a first resolution level, wherein the first volumetric image data is generated based on projection image data associated with a target structure of a patient; upsample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; acquire mask image data that identifies a region of interest (ROI) associated with the target structure; and generate multi-resolution volumetric image data based on the second volumetric image data and the mask image data, wherein the multi-resolution volumetric image data includes at least a first voxel associated with a second resolution level outside the ROI and a second voxel associated with a third resolution level within the ROI, wherein the third resolution level is higher than the second resolution level.

[0197] Clause 9. The non-transient computer-readable medium according to Clause 8, wherein instructions for upsampling first volumetric image data cause the processor to: generate second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0198] Clause 10. A non-transient computer-readable medium according to Clause 8 or Clause 9, wherein instructions for upsampling first volumetric image data cause a processor to perform at least one of the following operations: perform bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data; and apply an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0199] Clause 11. A non-transient computer-readable medium according to one or more of Clauses 8-10, wherein the instructions for acquiring first volumetric image data include: performing iterative image reconstruction based on projected image data to generate the first volumetric image data.

[0200] Clause 12. A non-transient computer-readable medium according to one or more of Clauses 8-11, wherein instructions for acquiring mask image data cause a processor to: acquire mask image data generated based on one or more of the following input data: projected image data, first volumetric image data, second volumetric image data, and segmentation data associated with a region of interest.

[0201] Clause 13. A non-transient computer-readable medium according to one or more of Clauses 8-12, wherein instructions for acquiring mask image data cause a processor to: acquire mask image data generated based on one or more of the following: a range of values ​​associated with a region of interest, a relative difference of values ​​associated with a region of interest, and a heuristic method specifying one or more rules for identifying the region of interest.

[0202] Clause 14. A non-transient computer-readable medium according to one or more of Clauses 8-13, wherein instructions for generating multi-resolution volumetric image data cause a processor to perform multi-resolution iterative image reconstruction based on second volumetric image data and mask image data, wherein the mask image data is updatable during multi-resolution iterative image reconstruction and is applied during at least one of the following operations: a forward projection operation and a back projection operation. A computer system is also provided, according to a desired implementation, comprising a processor and a non-transient computer-readable medium according to one or more of Clauses 8-14.

[0203] Clause 15. An imaging system comprising: an imaging source and a detector for acquiring projected image data associated with a target structure of a patient; and a computer system configured to: acquire first volumetric image data associated with a first resolution level, wherein the first volumetric image data is generated based on the projected image data; upsample the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; acquire mask image data that identifies regions of interest (ROIs) associated with the target structure; and generate multi-resolution volumetric image data based on the second volumetric image data and the mask image data, wherein the multi-resolution volumetric image data includes at least a first voxel associated with a second resolution level outside the ROI and a second voxel associated with a third resolution level within the ROI, wherein the third resolution level is higher than the second resolution level.

[0204] Clause 16. The imaging system of Clause 15, wherein the computer system is configured to upsample the first volumetric image data by generating second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0205] Clause 17. An imaging system according to Clause 15 or 16, wherein the computer system is configured to upsample first volumetric image data by performing at least one of the following operations: performing bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data; and applying an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0206] Clause 18. An imaging system according to one or more of Clauses 15-17, wherein the computer system is configured to acquire first volumetric image data by performing iterative image reconstruction based on projected image data to generate the first volumetric image data.

[0207] Clause 19. An imaging system according to one or more of Clauses 15-18, wherein the computer system is configured to acquire mask image data by acquiring mask image data generated based on one or more of the following input data: projected image data, first volumetric image data, second volumetric image data, and segmentation data associated with a region of interest.

[0208] Clause 20. An imaging system according to one or more of claims 15-19, wherein the computer system is configured to acquire mask image data by: acquiring mask image data generated based on one or more of: a range of values ​​associated with a region of interest, a relative difference of values ​​associated with a region of interest, and a heuristic method specifying one or more rules for identifying the region of interest.

[0209] Clause 21. An imaging system according to one or more of claims 15-20, wherein the computer system is configured to generate multi-resolution volumetric image data by performing multi-resolution iterative image reconstruction based on second volumetric image data and mask image data, wherein the mask image data is updatable during multi-resolution iterative image reconstruction and is applied during at least one of the following operations: a forward projection operation and a back projection operation.

[0210] Clause 22. A computer program including instructions that, when executed by a computer, cause the computer to perform the following steps: acquiring first volumetric image data associated with a first resolution level, wherein the first volumetric image data is generated based on projection image data associated with a target structure of a patient; upsampling the first volumetric image data associated with the first resolution level to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; acquiring mask image data that identifies regions of interest (ROIs) associated with the target structure; and generating multi-resolution volumetric image data based on the second volumetric image data and the mask image data, wherein the multi-resolution volumetric image data includes at least a first voxel associated with a second resolution level outside the ROI and a second voxel associated with a third resolution level within the ROI, wherein the third resolution level is higher than the second resolution level.

[0211] Clause 23. The computer program according to Clause 22, wherein the instructions for upsampling the first volumetric image data cause the computer to generate second volumetric image data by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

[0212] Clause 24. A computer program pursuant to Clause 22 or Clause 23, wherein instructions for upsampling first volumetric image data cause a computer to perform at least one of the following operations: perform bilinear upsampling or cubic upsampling based on the first volumetric image data to generate second volumetric image data; and apply an artificial intelligence (AI) engine trained to perform upsampling to generate second volumetric image data based on the first volumetric image data.

[0213] Clause 25. A computer program according to one or more of Clauses 22-24, wherein instructions for acquiring first volumetric image data cause the computer to: perform iterative image reconstruction based on projected image data to generate the first volumetric image data.

[0214] Clause 26. A computer program according to one or more of Clauses 22-25, wherein instructions for acquiring mask image data cause the computer to: acquire mask image data generated based on one or more of the following input data: projected image data, first volumetric image data, second volumetric image data, and segmentation data associated with a region of interest.

[0215] Clause 27. A computer program according to one or more of Clauses 22-26, wherein instructions for acquiring mask image data cause the computer to: acquire mask image data generated based on one or more of the following: a range of values ​​associated with a region of interest, a relative difference of values ​​associated with a region of interest, and a heuristic method specifying one or more rules for identifying the region of interest.

[0216] Clause 28. A computer program according to one or more of Clauses 22-27, wherein instructions for generating multi-resolution volumetric image data cause the computer to: perform multi-resolution iterative image reconstruction based on second volumetric image data and mask image data, wherein the mask image data is updatable during multi-resolution iterative image reconstruction and is applied during at least one of the following operations: a forward projection operation and a back projection operation.

[0217] Those skilled in the art will recognize that various modifications, alterations, and combinations can be made to the above embodiments without departing from the scope of the invention, and all such modifications, alterations, and combinations should be considered to be within the scope of the concept of the invention.

Claims

1. A method for a computer system to perform iterative image reconstruction using upsampling, the method comprising: Acquire projected image data associated with the patient's target structures; Based on the projected image data, an iterative image reconstruction is performed for a first number of iterations to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; The first volumetric image data associated with the first resolution level is upsampled to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; as well as Based on the second volumetric image data, perform a second number of iterative image reconstructions to generate output volumetric image data associated with the second resolution level.

2. The method according to claim 1, wherein upsampling the first volumetric image data comprises: The second volumetric image data is generated by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

3. The method according to claim 1, wherein upsampling the first volumetric image data comprises: Bilinear upsampling or cubic upsampling is performed based on the first volumetric image data to generate the second volumetric image data.

4. The method according to claim 1, wherein upsampling the first volumetric image data comprises: An AI engine trained to perform upsampling is used to generate second volumetric image data based on the first volumetric image data.

5. The method according to claim 1, wherein the method further comprises: The output volumetric image data is upsampled to a third volumetric image data associated with a third resolution level higher than the second resolution level; as well as Based on the third volumetric image data, perform a third iteration of iterative image reconstruction to generate final volumetric image data associated with the third resolution level.

6. The method of claim 1, wherein performing the iterative image reconstruction comprises: The first volumetric image data is generated based on the projected image data, wherein the projected image data represents first downsampled projected image data associated with the first resolution level; as well as The output volumetric image data is generated based on the second downsampled projected image data associated with the second resolution level.

7. The method of claim 1, wherein the method further comprises at least one of the following operations: A first user interface view is generated and displayed on a display device to allow a user to interact with the first volumetric image data, wherein the first user interface view specifies at least one of the following: the first resolution level and the first iteration number; and A second user interface view is generated and displayed on the display device to allow the user to interact with the output volumetric image data, wherein the second user interface view specifies at least one of the following: the second resolution level and the total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

8. A non-transitory computer-readable medium having instructions stored thereon, the instructions causing the processor to perform the following operations when executed by a processor: Acquire projected image data associated with the patient's target structures; Based on the projected image data, an iterative image reconstruction is performed for a first number of iterations to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; The first volumetric image data associated with the first resolution level is upsampled to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; as well as Based on the second volumetric image data, perform a second number of iterative image reconstructions to generate output volumetric image data associated with the second resolution level.

9. The non-transient computer-readable medium of claim 8, wherein the instruction for upsampling the first volumetric image data causes the processor to: The second volumetric image data is generated by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

10. The non-transient computer-readable medium of claim 8, wherein the instruction for upsampling the first volumetric image data causes the processor to: Bilinear upsampling or cubic upsampling is performed based on the first volumetric image data to generate the second volumetric image data.

11. The non-transient computer-readable medium of claim 8, wherein the instruction for upsampling the first volumetric image data causes the processor to: An AI engine trained to perform upsampling is used to generate second volumetric image data based on the first volumetric image data.

12. The non-transient computer-readable medium of claim 8, wherein the instructions further cause the processor to: The output volumetric image data is upsampled to a third volumetric image data associated with a third resolution level higher than the second resolution level; and Based on the third volumetric image data, perform a third iteration of iterative image reconstruction to generate final volumetric image data associated with the third resolution level.

13. The non-transient computer-readable medium of claim 8, wherein the instructions for performing the iterative image reconstruction cause the processor to: The first volumetric image data is generated based on the projected image data, wherein the projected image data represents first downsampled projected image data associated with the first resolution level; and The output volumetric image data is generated based on the second downsampled projected image data associated with the second resolution level.

14. The non-transient computer-readable medium of claim 8, wherein the instructions further cause the processor to perform at least one of the following operations: A first user interface view is generated and displayed on a display device to allow a user to interact with the first volumetric image data, wherein the first user interface view specifies at least one of the following: the first resolution level and the first iteration number; and A second user interface view is generated and displayed on the display device to allow the user to interact with the output volumetric image data, wherein the second user interface view specifies at least one of the following: the second resolution level and the total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

15. An imaging system, comprising: Imaging sources and detectors are used to acquire projected image data associated with the patient's target structures; as well as Computer system, the computer system being configured as follows: Based on the projected image data, an iterative image reconstruction is performed for a first number of iterations to generate first volumetric image data, wherein the first volumetric image data is associated with a first resolution level; The first volumetric image data associated with the first resolution level is upsampled to generate second volumetric image data associated with a second resolution level, wherein the second resolution level is higher than the first resolution level; as well as Based on the second volumetric image data, perform a second number of iterative image reconstructions to generate output volumetric image data associated with the second resolution level.

16. The imaging system of claim 15, wherein the computer system is configured to upsample the first volumetric image data in the following manner: The second volumetric image data is generated by upsampling the first volumetric image data at least along the vertical axis, such that the second resolution level is associated with a higher vertical resolution compared to the first resolution level.

17. The imaging system of claim 15, wherein the computer system is configured to upsample the first volumetric image data in the following manner: Bilinear upsampling or cubic upsampling is performed based on the first volumetric image data to generate the second volumetric image data.

18. The imaging system of claim 15, wherein the computer system is configured to upsample the first volumetric image data in the following manner: An AI engine trained to perform upsampling is used to generate second volumetric image data based on the first volumetric image data.

19. The imaging system of claim 15, wherein the computer system is further configured to: The output volumetric image data is upsampled to a third volumetric image data associated with a third resolution level higher than the second resolution level; and Based on the third volumetric image data, perform a third iteration of iterative image reconstruction to generate final volumetric image data associated with the third resolution level.

20. The imaging system of claim 15, wherein the computer system is configured to perform the iterative image reconstruction in such a way that: The first volumetric image data is generated based on the projected image data, wherein the projected image data represents first downsampled projected image data associated with the first resolution level; and The output volumetric image data is generated based on the second downsampled projected image data associated with the second resolution level.

21. The imaging system of claim 15, wherein the computer system is further configured to perform at least one of the following operations: A first user interface view is generated and displayed on a display device to allow a user to interact with the first volumetric image data, wherein the first user interface view specifies at least one of the following: the first resolution level and the first iteration number; and A second user interface view is generated and displayed on the display device to allow the user to interact with the output volumetric image data, wherein the second user interface view specifies at least one of the following: the second resolution level and the total number of iterations, the total number of iterations including the first number of iterations and the second number of iterations.

Images