Artifact-driven data synthesis in computed tomography

Abstract

To provide computer processing techniques for augmenting computed tomography (CT) images with synthetic artifacts for artificial intelligence (AI) applications.SOLUTION: A computer-implemented method can include generating, by a system comprising a processor, synthetic artifact data corresponding to one or more CT image artifacts, where the synthetic artifact data comprises anatomy-agnostic synthetic representations of the one or more CT image artifacts. The method further includes generating, by the system, augmented CT images comprising the one or more CT image artifacts using the synthetic artifact data. In one or more examples, the method can further include training, by the system, a medical image inferencing model to perform an inferencing task using the augmented CT images as training images.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] This application relates to medical image processing, and more particularly to computer processing techniques for enhancing computed tomography (CT) images with synthetic artifacts. [Background technology]

[0002] Machine learning (ML) models are used in many medical image processing and analysis tasks, such as organ segmentation, anomaly detection, diagnosis classification, risk prediction, temporal analysis, and image reconstruction. However, one of the fundamental problems in data-driven machine learning approaches is that the final model inferencing capability is limited by the scope of the training data used for model development. With regard to the medical imaging domain, due to the nature of the data acquisition process, it can be difficult and costly to acquire enough medical images for model training that provide a comprehensive representation of the potential image variants across different patient populations. [Brief description of the drawings]

[0003] [Figure 1] 1 illustrates an exemplary computing system equipped to augment CT images with synthetic artifacts in accordance with one or more embodiments of the disclosed subject matter. [Diagram 2] 1 illustrates several examples of CT image artifacts in accordance with one or more embodiments of the disclosed subject matter. [Figure 3A]FIG. 1 illustrates a simulation example of a CT metal artifact according to a conventional simulation process. [Figure 3B] FIG. 1 shows an example of a CT metal artifact simulation according to the disclosed anatomy-agnostic simulation process. [Figure 4] 1A-1C show examples of Gaussian noise simulation in the CT image domain and the CT sinogram domain using the disclosed anatomy-agnostic simulation process in accordance with one or more embodiments of the disclosed subject matter. [Diagram 5] 1 illustrates an exemplary workflow for generating enhanced CT image data using pre-generated synthetic artifact data in accordance with one or more embodiments of the disclosed subject matter. [Figure 6] FIG. 6 is a diagram illustrating an example of CT image enhancement with Poisson noise, in accordance with one or more embodiments of the disclosed subject matter. [Figure 7] 1A-1C show examples of CT image enhancement with Gaussian noise, in accordance with one or more embodiments of the disclosed subject matter. [Figure 8] 1 illustrates an exemplary CT image enhancement with randomly inserted metal artifacts in accordance with one or more embodiments of the disclosed subject matter. [Figure 9] FIG. 1 illustrates target artifact integration in CT using ground truth labeling, in accordance with one or more embodiments of the disclosed subject matter. [Figure 10] 13A-13C illustrate several examples of metal artifact CT image enhancement operations facilitated by a data enhancement component in accordance with one or more embodiments of the disclosed subject matter. [Figure 11] FIG. 1 illustrates an exemplary computing system equipped to augment CT images with synthetic artifacts and employ the augmented CT images for training a medical image inference model in accordance with one or more embodiments of the disclosed subject matter. [Figure 12]FIG. 1 is a block diagram of an exemplary, non-limiting computer-implemented method for generating an enhanced CT image including a synthesis artifact, in accordance with one or more embodiments of the disclosed subject matter. [Figure 13] FIG. 1 is a block diagram of an exemplary, non-limiting computer-implemented method for developing a medical image inference model that is robust to artifacts, in accordance with one or more embodiments of the disclosed subject matter. [Figure 14] FIG. 1 illustrates a block diagram of an exemplary, non-limiting operating environment in which one or more embodiments described herein may be facilitated. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0004] The following detailed description is merely illustrative and is not intended to limit the embodiments and / or the application or uses of the embodiments, and further, there is no intention to be bound by any express or implied information presented in the preceding "Background" or "Detailed Description" sections.

[0005] The disclosed subject matter is directed to a system, computer-implemented method, apparatus, and / or computer program product that provides computer processing techniques that facilitate artifact-driven data synthesis for computed tomography (CT) artificial intelligence (AI) applications. Synthetic imaging techniques are used in the medical imaging domain to increase the amount and variety of medical images used to train machine learning (ML) models, a process commonly referred to as training data augmentation or simply data augmentation. In contrast to actual medical image data that is derived from direct measurements captured via a medical image acquisition system (e.g., a CT scanner), synthetic imaging techniques use rule-based schemes, statistical modeling, simulation, and other techniques to generate synthetic or artificially generated medical image data.

[0006] One or more embodiments of the disclosed subject matter are directed to synthesizing CT artifacts in the context of employing CT artifacts for data augmentation of CT image training datasets for CTAI / ML applications. The disclosed technique identifies multiple different types of CT image artifacts observed in real CT images and synthesizes these artifacts independently. The synthesized artifacts are then combined with the real CT images in various ways to generate augmented synthetic CT images that can be used as additional training images to increase the diversity of training image datasets for CTAI / ML model development (i.e., artifact-driven data synthesis for CTAI applications). In this regard, there are various types of artifacts observed in real CT images, such as metal objects, Poisson noise, Gaussian noise, streaks, scatter, etc. Many simulation techniques have been developed to simulate artifacts in CT images. However, these simulations are typically performed on real CT images for the purpose of developing artifact removal applications. As a result, conventional CT artifact simulation is anatomically and patient-dependent and cannot be used for augmentation purposes.

[0007] The disclosed technique employs a novel anatomy agnostic simulation method to simulate different CT artifacts in a simulation material that has low radiodensity values ​​as opposed to the anatomical regions of the body represented in the actual CT image data. As a result, the disclosed technique can generate synthetic CT artifacts that are anatomically agnostic (i.e., not tied to any particular anatomical region of the body) and free of any patient dependencies. In various embodiments, the simulation material has a Hounsfield unit (HU) value preferably between about -200 and +200, more preferably between about -100 and +100, and even more preferably between about -50 and +50. In some embodiments, the simulation material consists of water, which has a HU value of zero. Other suitable simulation materials may include, but are not limited to, fat (e.g., having an HU value of about -50), human soft tissue (e.g., having an HU value of about +50), human blood (e.g., having an HU value of about +50), and similar materials. In some embodiments, the disclosed techniques further generate the synthetic CT artifact in the sinogram domain as opposed to the image domain, which further enhances the realistic quality of the synthetic CT artifact. In other embodiments, the synthetic CT artifact may be generated in the image domain and / or the projection domain.

[0008] One or more embodiments of the disclosed subject matter are directed to synthesizing a plurality of different CT artifacts using the above-described anatomy-agnostic simulation method. The different CT artifacts correspond to pre-generated synthetic representations of the different CT artifacts. One or more additional embodiments are directed to combining the pre-generated synthetic CT artifacts with a CT image to generate an augmented CT image that each includes one or more of the synthetic artifacts. One or more additional embodiments are further directed to employing the augmented CT image in conjunction with training a medical image inference model to perform one or more inferencing tasks with improved robustness to artifacts.

[0009] According to one embodiment, a system is provided that includes a memory that stores computer-executable components and a processor that executes the computer-executable components stored in the memory. The computer-executable components include a simulation component that generates synthetic artifact data corresponding to one or more CT image artifacts, the simulation component generating the synthetic artifact data using an anatomically agnostic simulation model, such that the synthetic artifact data is comprised of one or more synthetic representations of the one or more CT image artifacts simulated in a simulation material having a low radiodensity value, rendering the synthetic artifact data anatomically agnostic. In various embodiments, the simulation material has a HU value preferably between about -200 and +200, more preferably between about -100 and +100, and even more preferably between about -50 and +50. In some embodiments, the simulation material is comprised of water, which has a HU value of zero. Other suitable simulation materials may include, but are not limited to, fat (e.g., having an HU value of about -50), human soft tissue (e.g., having an HU value of about +50), human blood (e.g., having an HU value of about +50), and similar materials.

[0010] In one or more embodiments, the simulation component generates the composite artifact data in the sinogram domain. In other embodiments, the simulation component generates the composite artifact data in the projection domain. In yet other embodiments, the simulation component generates the composite artifact data in the image domain.

[0011] The computer-executable components further include a data augmentation component that uses the synthetic artifact data to generate augmented CT images including one or more computed tomography image artifacts. In this regard, the data augmentation component combines the synthetic artifact data with one or more CT images to generate the augmented CT images. In some implementations, the data augmentation component registers one or more CT image artifacts to target locations on the one or more CT images using ground truth region of interest data associated with the one or more CT images. In various implementations, the one or more CT images include medical images corresponding to different anatomical regions. In this regard, because the synthetic artifact data is anatomically agnostic, the data augmentation component can apply corresponding artifacts to any CT scan image, regardless of the body part or body region scanned. The data augmentation component can also combine artifacts from one or more CT images with respective copies of the same CT image at different positions or orientations in connection with generating the augmented CT images.

[0012] In various embodiments, the one or more CT image artifacts are comprised of a plurality of different types of artifacts. For example, the CT image artifacts may include, but are not limited to, one or more metal object artifacts, one or more Poisson noise artifacts (e.g., representing different levels of Poisson noise), and one or more Gaussian noise artifacts (e.g., representing different levels of Gaussian noise). Other types of CT image artifacts are also contemplated. In some implementations, the data augmentation component may combine two or more of the plurality of different types of artifacts with the same CT image in connection with generating the augmented CT image.

[0013] In one or more embodiments, the computer-executable components further comprise a training component using the enhanced CT images as training images in connection with training the medical image inference model to perform an inferencing task.

[0014] In some embodiments, the elements described in the disclosed system may be embodied in different forms, such as a computer-implemented method, a computer program product, or other forms.

[0015] The disclosed technique provides many technical advantages. The disclosed technique provides for efficient augmentation of a CT image training dataset with increased variety in the form of CT images with synthetically integrated artifacts that are synthetically generated and applied to the CT images. The augmented training dataset, which includes not only artifact-free CT images but also additional CT images with synthetic artifacts, significantly improves the robustness (e.g., accuracy, specificity, etc.) of medical image inference models to artifacts when trained with the augmented training dataset. In addition, due to the fact that the synthetic CT artifacts are pre-generated, anatomy-agnostic, and patient-free, the disclosed technique enables rapid and efficient integration of different types of synthetic artifacts at different positions and orientations on any anatomical CT image. Furthermore, the disclosed technique of generating synthetic artifacts using an anatomy-agnostic simulation model significantly reduces the processing speed and computational resources required to generate different types of synthetic artifacts applicable to different anatomical CT images.

[0016] In this specification, the terms "algorithm" and "model" are used interchangeably unless the context requires otherwise. The terms "artificial intelligence (AI) model" and "machine learning (ML) model" are used interchangeably in this specification, unless the context requires otherwise.

[0017] As used herein, the term "image inferencing model" refers to an AI / ML model adapted to perform image processing or analysis tasks on image data. The image processing or analysis tasks may be varied. In various embodiments, the image processing or analysis tasks may include, but are not limited to, a segmentation task, an image reconstruction task, an object recognition task, a motion detection task, a video tracking task, an optical flow task, and the like. The image inferencing models described herein may include not only two-dimensional (2D) image processing models, but also three-dimensional (3D) image processing models. The image processing model may employ various types of AI / ML algorithms, including, but not limited to, deep learning models, neural network models, deep neural network models (DNNs), convolutional neural network models (CNNs), generative adversarial neural network models (GANs), etc. In this specification, the terms "image inferencing model", "image processing model", "image analysis model", etc. are used interchangeably unless otherwise required by the context.

[0018] As used herein, the term "image-based inference output" refers to a decision or prediction that an image processing model is configured to generate. For example, an image-based inference output may include a segmentation mask, a reconstructed image, an adapted image, an annotated image, a classification, a value, etc. Image-based inference output varies based on the type of model and the particular task the model is configured to perform. Image-based inference output may include renderable data objects (e.g., visual data objects), storable data objects, data objects usable as input for another processing task, etc. As used herein, terms such as "image-based inference output," "inference output," "inference result," "inference," "output," "predication," etc. are used interchangeably unless the context specifically requires a distinction between the terms.

[0019] In this specification, a "medical imaging inferencing model" refers to an image inference model that is adapted to perform image processing / analysis tasks on medical image data. For example, medical image processing / analysis tasks include, but are not limited to, disease / condition classification, disease region segmentation, organ segmentation, disease quantification, disease / condition staging, risk prediction, temporal analysis, anomaly detection, anatomical feature characterization, medical image reconstruction, and the like. In this specification, the terms "medical image inferencing model", "medical image processing model", "medical image analysis model", and the like are used interchangeably unless the context requires otherwise.

[0020] The term "image data" is used herein to refer to both digital images and data used to generate digital images (e.g., raw measurement data generated by the imaging acquisition system and associated metadata, simulated measurement data and associated metadata, sinogram data, etc.). As applied to the disclosed technology, the term image data may include CT image data, synthetic CT image data, CT artifact data, synthetic CT artifact data, and augmented CT image data. The term "native" as applied to medical image data is used herein to refer to actual medical image data obtained from direct measurements captured via a medical image acquisition system (e.g., a CT scanner). The term "synthetic" as applied to medical imaging data, and more specifically to CT artifact data, is used to refer to artificially generated medical imaging data that is generated using rule-based schemes, statistical modeling, simulation, and other techniques.

[0021] The system and device embodiments described herein may include one or more machine-executable (i.e., computer-executable) components or instructions embodied in one or more machines (e.g., embodied in one or more computer-readable storage media associated with one or more machines). Such components, when executed by one or more machines (e.g., processors, computers, computing devices, virtual machines, etc.), may cause the one or more machines to perform the operations described. These computer / machine-executable components or instructions (and others described herein) may be stored in a memory associated with the one or more machines. The memory may further be operatively coupled to at least one processor such that the components are executable by the at least one processor to perform the operations described. In some embodiments, the memory may include a non-transitory machine-readable storage medium including instructions that, when executed by the executable components or processors, facilitate performance of the operations described for the respective executable components. Examples of the above and memory and processors, as well as other suitable computer or computing-based elements, can be found with reference to FIG. 14 (e.g., processing unit 1404 and system memory 1406, respectively) and can be used in connection with implementing one or more of the systems or components shown and described in connection with FIG. 1, or other figures disclosed herein.

[0022] One or more embodiments will now be described with reference to the drawings, in which like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a more thorough understanding of one or more embodiments. However, it will be apparent that in various instances, one or more embodiments may be practiced without these specific details.

[0023] FIG. 1 illustrates an exemplary computing system (hereinafter, system 100) with facilities for augmenting CT images with synthetic artifacts in accordance with one or more embodiments of the disclosed subject matter. System 100 includes computer-executable components 102, storage 116, communication components 122, input / output devices 112, processing unit 114, and memory 124. System 100 further includes a system bus 126 coupling computer-executable components 112, storage 116, communication components 122, input / output devices 112, processing unit 114, and memory 124 to one another. In some embodiments, computer-executable components 102 may be stored in memory 124 and executed by processing unit 114 to perform operations described with respect to components corresponding to system 100. In this regard, computing system 100 may correspond to any suitable computing device or machine (e.g., communication devices, desktop computers, personal computers, smartphones, servers, virtual computing devices, etc.) or an interconnected group of computing devices / machines (e.g., interconnected via wired and / or wireless communication technologies).

[0024] The system 100 can be communicatively coupled (e.g., via any suitable wired and / or wireless communication technique) to an external storage device 128 that can provide data to be processed by the system 100, including CT image data 130. In other embodiments, the CT image data 130 can be stored locally in the storage device 116 and / or memory 124. The storage device 116 can store data generated by the system 100, including the composite artifact data 118 and the enhanced CT image data 120. In this regard, the storage device 128 and the storage device 116 can correspond to any suitable machine-readable medium accessible by the computing system 100, including both volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, models, algorithms, program modules, or other data. Computer storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory or other memory technology, digital video disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by computing system 100.

[0025] The communications component 122 may correspond to any suitable communications technology hardware and / or software capable of performing wired and / or wireless communication of data between the computing system 100 and other systems, devices and / or databases (e.g., storage device 128). In this regard, the communications component 122 may provide for receiving data (e.g., CT image data 130) from one or more external systems and / or devices and communicating (e.g., sending, transmitting, etc.) data (e.g., composite artifact data 118, enhanced CT image data 120, etc.) to one or more external systems and / or devices. Examples of suitable communications technology hardware and / or software that may be employed by the communications component 122 are described below with reference to FIG. 14. The input / output devices 112 may correspond to any suitable input and / or output devices (e.g., keyboard, mouse, touch screen, display, etc.) that provide for receiving user input and displaying or rendering information to a user (e.g., in a visual format via a display, as audio, etc.) in connection with use of the features and functionality of the computing system 100. For example, in various embodiments, the input / output devices 112 may include any suitable display device, via which the rendering component 110 may render (e.g., display, present, etc.) either data received by the system 100 (e.g., CT image data 130) and / or data generated by the system 100 (e.g., composite artifact data 118, enhanced CT image data 120). Additional examples of some suitable input / output devices 112 are also described with reference to FIG. 14 with respect to one or more input devices 1428 and one or more output devices 1436).

[0026] The computer-executable components 102 can include a simulation component 104 , a simulation model 106 , a data augmentation component 108 , and a rendering component 110 .

[0027] The simulation component 104 can generate synthetic artifact data 118 corresponding to one or more CT image artifacts. There are many different types of artifacts observed in real or native CT images, such as metal objects, noise (eg, Poisson noise, Gaussian noise, etc.), streaks, scatter, and several others. FIG. 2 illustrates some examples of CT image artifacts in accordance with one or more embodiments of the disclosed subject matter. As illustrated in FIG. 2, the CT image artifacts can include artifacts due to system design (eg, design, model, make operating settings, etc., of the CT scanning and reconstruction system), X-ray tube induced, detector induced, patient induced, operator induced, and the like. Artifacts from one or more of these categories routinely appear in real CT images as visual image features in the form of noise, blur, streaks, objects, and other image characteristics. Due to the nature of the CT imaging process and the various causes of CT artifacts, the amount and variety of different CT artifacts that appear in real CT images is high. Thus, training datasets used in the medical field to train and develop medical image inference models to perform clinical inference tasks on real CT images (on real patients) must account for the wide range of CT artifacts that may be observed. As mentioned in the background section, obtaining real CT image training datasets for training and developing ML / AI models that account for the various artifacts that may be present in real patient images at run time can be difficult and costly.

[0028] With this context in mind, and referring to FIGS. 1-2, the simulation component 104 can generate synthetic representations (e.g., synthetic artifact data 118) of a wide range of different CT artifacts shown in FIG. 2. The data augmentation component 108 can further generate an augmented CT image including one or more of the different CT artifacts using the synthetic representations. To facilitate this purpose, the simulation component 104 can employ a simulation model 106 to generate the synthetic artifact data 118. The simulation model 106 simulates the capture of each CT artifact by the CT scanner in a simulation material that has a low radio density value in contrast to the anatomical regions of the body represented in the actual CT image data, such as a body of water as opposed to the human body. As a result, the simulation model 106 generates synthetic CT artifacts that are anatomy-agnostic (e.g., not tied to any particular anatomical region of the body) and free of any patient dependency, making the simulation model 106 an anatomy-agnostic or anatomically independent simulation model. In various embodiments, the simulation material preferably has an HU value between about -200 and +200, more preferably between about -100 and +100, and even more preferably between about -50 and +50. In some embodiments, the simulation material includes water, which has an HU value of zero. Other suitable simulation materials may include, but are not limited to, fat (e.g., having an HU value of about -50), human soft tissue (e.g., having an HU value of about +50), human blood (e.g., having an HU value of about +50), and similar materials.

[0029] In this regard, CT scanners use a rotating X-ray (XR) tube and an array of detectors located in a gantry to measure XR attenuation by various tissues in the body. Following a conventional CT scanning procedure, the patient is positioned in the center of a motorized scanning table through the circular opening of the CT imaging system. As the patient passes through the CT imaging system via the motorized scanning table (e.g., progressively forward or backward to generate different tube positions relative to the patient's body), the XR tube, which provides the XR source, rotates inside the circular opening. The XR tube generates a narrow fan beam of XR that is used to irradiate a portion of the patient's body. In a typical examination, there are several phases, each consisting of 10 to 50 rotations of the XR tube around the patient in coordination with the table moving through the circular opening. A detector at the patient's exit side records the XR signal emanating from the portion of the patient's body being irradiated as an XR "snapshot" at one position (angle or view) of the XR tube. More specifically, the XR tube projects XR photons through the patient's body, and an array of detectors located on the gantry on the opposite side of the body measures the XR attenuation of the photons by different tissues in the body based on the amount of photons that reach the detector cells, called projection signals or simply projections. Thus, each "snapshot" corresponding to each tube position can be composed of multiple projections. The number of projections obtained for each tube position (e.g., each tube angle or view) depends on the configuration of the detector cells, which may include an array of one or more rows and columns of cells. Many different "snapshots" (angles) are collected during one complete (e.g., 360°) rotation. The collection of projections taken during the scanning process is then sent to a computer, which reconstructs all of the individual "snapshots" into cross-sectional images (slice) of internal organs and tissues for each complete rotation of the XR tube.In this regard, in conventional (e.g., non-helical) CT scanning procedures, the collection of signals captured at each tube position (e.g., angle or view) of a rotation (referred to herein as projections) is then backprojected to create a reconstructed image that corresponds to a cross-sectional slice of the body in that rotation.

[0030] The projection is measured as a function of Hounsfield units (HU), which are a relative quantitative measurement of radiation density used by radiologists in interpreting CT images. The absorption / attenuation coefficient of radiation in tissue is used to generate grayscale images during CT reconstruction. The physical density of tissue is proportional to the absorption / attenuation of the x-ray beam. HU (also called CT units) is calculated based on a linear transformation of the baseline linear attenuation coefficient of the x-ray beam.

[0031] The simulation model 106 is based on the fact that water comprises approximately 72% of human tissue measured by CT, making water a good approximation medium of the human body for data science applications (e.g., training and development of AI models). Also, the HU value (standard temperature, standard pressure) of distilled water is arbitrarily defined as zero. Thus, by using a simulation material of water or other materials with low CT values ​​(e.g., fat, soft tissue, blood, etc.), the simulation model 106 simulates the occurrence of CT artifacts in background attenuation media that are zero (i.e., blank canvas / background) or near zero (e.g., -50 to +50 for fat, soft tissue, blood, and similar materials), making the resulting synthetic CT artifact representation anatomically independent.

[0032] For example, FIG. 3A illustrates an exemplary CT metal artifact simulation according to a conventional simulation process, and FIG. 3B illustrates a corresponding exemplary CT metal artifact simulation according to the disclosed anatomy-agnostic simulation process using a simulated material of water. As shown in FIG. 3A, the conventional method for simulating artifacts such as metal implants in this embodiment is performed on real CT images (i.e., CT image 301 and CT image 302). In this regard, CT image 301 and CT image 302 each correspond to a real CT image of a femur implanted with a synthetically generated metal implant (depicted in white). As shown in FIG. 3A, as a result of simulating a metal implant on / in the real CT image data, the resulting metal artifacts have anatomical and patient dependencies. In particular, the visual appearance of the simulated metal implants is impacted or influenced by the content (i.e., projection values ​​and corresponding pixel characteristics) of the CT images in which they are simulated. Conversely, FIG. 3B presents artifact images 304 and 305 consisting of the same metal implant illustrated in FIG. 3A but generated using the simulation model 106. As illustrated in FIG. 3B, the artifact images 304 and 305 do not contain anatomical dependencies. In theory, one could try to extract the simulated metal implants from the actual CT image of FIG. 3A using segmentation processing techniques to generate synthetic CT artifacts that are removed from the anatomical background data. However, the extracted artifacts would still have visual properties that are affected by the background data from which they were originally created, making them inaccurate for the disclosed data augmentation purposes. In addition, conventional simulations of CT artifacts are tailored to specific anatomical data sets (e.g., brain, abdomen, chest, etc.) and are computationally intensive (and expensive) to perform.Furthermore, each type of artifact (e.g., metal implants, Poisson noise, etc.) simulated using existing simulation techniques requires a different simulation methodology.

[0033] 1-3, in accordance with the disclosed technique, the simulation component 104 can apply the same simulation model 106 to generate synthetic artifacts for a variety of different types of CT artifacts, including metal objects, Poisson noise, Gaussian noise, and various others depicted in FIG. 2. In addition, since the artifacts are simulated for simulation materials with low radiodensity values, such as water, fat, soft tissue, blood, and similar materials (e.g., HU values ​​between 200 and +200, more preferably between about -100 and +100, and even more preferably between about -50 and +50), the synthetic artifacts are not patient-dependent or anatomical-dependent. As a result, the data augmentation component 108 can apply them to any type of CT image corresponding to any anatomical body region or body part (e.g., head, abdomen, chest, etc.).

[0034] In some embodiments, the simulation model 106 can generate the synthetic artifact data 118 in the image domain. Additionally or alternatively, the simulation model 106 can generate the synthetic artifact data 118 in the sinogram domain and / or projection domain as opposed to the image domain, which further enhances the realistic quality of the synthetic CT artifact. Although the artifact images 304 and 305 shown in FIG. 3B depict synthetic metal object artifacts rendered in the image domain, it should be understood that they may be simulated in the image domain, sinogram domain and / or projection domain. Any sinogram domain representation of a CT artifact can be converted to the image domain and / or projection domain, and vice versa, using conventional CT reconstruction algorithms as described below.

[0035] In this regard, as described above, according to the CT scanning procedure, a beam of XR is passed through the anatomical slice of interest, and the resulting XR is captured on a detector opposite the source as a projection image consisting of multiple projections, each corresponding to the amount of photons absorbed by a respective tissue in the body. The source-detector combination is rotated around the object being imaged to capture a multitude of projection images. This is CT image acquisition. A CT sinogram field, or simply a sinogram, refers to a two-dimensional array of data that includes the projections of all projection images. Each reconstructed CT image has a sinogram representation that corresponds to a 2D array of projection values ​​captured / detected at the viewpoint (i.e., position and orientation) of the CT image slice. In this regard, the image acquisition step is followed by a reconstruction step, in which CT reconstruction algorithms (e.g., backprojection and filtered backprojection are examples) are used to get back the anatomical details of the imaged slice using forward projection (or Radon transform) and / or backprojection (or inverse Radon transform).

[0036] In this regard, reference to simulating an artifact in the sinogram domain means modeling the composite artifact using projection data corresponding to simulated projection values ​​(i.e., attenuation intensity) represented by the artifact itself with background simulated materials (e.g., water, fat, soft tissue, blood, and similar materials) having low radiodensity values. In various embodiments, using the simulation model 106, the simulation component 104 can generate a composite sinogram representation of the CT artifact. The simulation component 104 can also convert the composite sinogram representation into a grayscale image corresponding to a composite CT image in the image domain using conventional CT reconstruction algorithms (e.g., back-projection and filtered back-projection). In this regard, the composite artifact data 118 can be composed of multiple composite artifacts, each corresponding to a different CT artifact, and each composite artifact can be composed of a composite sinogram representation and an image domain representation calculated or reconstructed from the sinogram representation. For example, the different composite artifacts may correspond to different types of objects (e.g., different types of implants observed in a patient) and / or objects of different materials (e.g., different types of metals or other materials). The different composite artifacts can also include noise artifacts corresponding to Poisson noise, Gaussian noise, and / or different levels of noise (e.g., different levels of Poisson noise and / or Gaussian noise).

[0037] The data augmentation component 108 may further combine the synthetic artifact data 118 with the CT image data (e.g., CT image data 130) to generate augmented CT image data 120. In this regard, the data augmentation component 108 may insert synthetic artifacts included in the synthetic artifact data 118 onto or within the CT image included in the CT image data 130 to generate an augmented version of the CT image including the synthetic artifacts. The types of CT images represented in the CT image data 130 may vary and correspond to CT images representing any anatomical region of the body, including real CT images as well as synthetic CT images.

[0038] In some embodiments, the data augmentation component 108 can synthesize the augmented CT image in the sinogram domain. In these embodiments, the CT image data 130 can include a sinogram representation of one or more CT images, and / or the data augmentation component 108 can simulate a sinogram representation of the CT image using forward projection processing. In these embodiments, the data augmentation component 108 can combine a first sinogram representation of the CT image with a second sinogram representation of the synthetic artifact to generate a third fused sinogram representation of the augmented CT image including the artifact. The data augmentation component 108 can further convert the third fused sinogram representation to a grayscale CT image in the image domain for rendering. In other embodiments, the data augmentation component 108 can generate the augmented CT image data 120 in the image domain. In these embodiments, the data augmentation component 108 can combine a CT image included in the CT image data (i.e., in the image domain) with a synthetic artifact image included in the synthetic artifact data 118 (i.e., in the image domain) to generate an augmented CT image including the synthetic artifact. In some implementations of these embodiments, the data augmentation component 108 can apply the synthetic artifact image as overlay data to the CT image.

[0039] 4 illustrates an exemplary Gaussian noise simulation in the CT image domain and the CT sinogram domain using the disclosed anatomy-agnostic simulation process in accordance with one or more embodiments of the disclosed subject matter. Image 401 corresponds to an original CT image depicting a cross-sectional region of the pelvis. Image 402 corresponds to a first augmented version of the original CT image 401 where a first synthetic representation of Gaussian noise is added to the original CT image 401, where the first synthetic representation of Gaussian noise was simulated in the image domain using the simulation model 106. Image 404 corresponds to a differential image depicting only the added noise (e.g., image 402 minus image 401). Image 403 corresponds to a second augmented version of the original CT image 401 where a second synthetic representation of Gaussian noise is added to the original CT image 401, where the second synthetic representation of Gaussian noise was simulated in the sinogram domain using the simulation model 106. Image 405 corresponds to a difference image (e.g., image 404 minus image 401) depicting only the added noise. As can be seen by comparing difference image 405 with difference image 404, and with image 403 with image 402, the simulation of the sinogram domain results in a more realistic representation of Gaussian noise in CT due to the consideration of the CT scanning geometry required to compute the sinogram representation.

[0040] FIG. 5 illustrates an example workflow 500 for generating augmented CT image data using pre-generated synthetic artifact data (e.g., by the data augmentation component 108) in accordance with one or more embodiments of the disclosed subject matter. With reference to FIGS. 1-5 , in various embodiments, the synthetic artifact data has been pre-generated (e.g., by the data augmentation component 108). In various embodiments, the synthetic artifact data 118 can include or correspond to one or more collections of pre-generated synthetic CT artifacts (e.g., synthetic representations of CT artifacts in the sinogram domain, the projection domain and / or the image domain) generated by the simulation component 104 using the simulation model 106. For example, the composite artifact data 118 may include Poisson noise data 506 corresponding to different levels of Poisson noise artifacts (e.g., low, medium, high, very high, etc.), different levels of Gaussian noise artifacts (e.g., low, medium, high, very high, etc.), metal artifact data 510 corresponding to different types of metal objects and / or the same object of different types of metal, etc. (e.g., other artifact data 512 may correspond to one or more additional collections of one or more additional types of composite artifacts depicted in FIG. 2 ).

[0041] In accordance with the disclosed techniques, the simulation component generates synthetic artifact data 118 for simulation materials having low radiodensity values ​​(e.g., water, fat, soft tissue, blood, and similar materials), so that each synthetic artifact is anatomy agnostic. As a result, the data augmentation component 108 can apply any of the synthetic artifacts to any type of CT image data corresponding to any anatomical region. For example, in the illustrated embodiment, the workflow 500 includes processing head CT data 502 and pelvis CT data 504 included in the CT image data 130 with the same pre-generated synthetic artifact data 118 to generate corresponding augmented versions thereof (e.g., augmented head CT data 514 and augmented pelvis CT data, respectively). It should be understood that the types of head CT data and pelvis CT data are merely exemplary, and the particular anatomical regions depicted may include any anatomical region. Additionally, the data augmentation component 108 can apply one or more (e.g., two or more) artifacts to the same CT image in conjunction with generating the augmented CT image data 120. For example, the circular plus sign icons above and below each composite artifact collection indicate an add function that may be selectively activated (or deactivated) by the data augmentation component 108 (and / or in response to user input) to selectively add artifacts from the corresponding collection to the corresponding CT data. In this regard, the augmented head CT data 514 may include the addition of a selected level of Poisson noise, the addition of a selected level of Gaussian noise, one or more selected metal artifacts, etc. Similarly, the augmented pelvis CT data 516 may include the addition of a selected level of Poisson noise, the addition of a selected level of Gaussian noise, one or more selected metal artifacts, etc. Furthermore, the composite artifact data 118 may be pre-generated because it is not tied to a particular anatomical structure or CT image data set, and the artifact-enhancing workflow 500 may be executed at high processing speeds in a computationally efficient manner.

[0042] FIG. 6 illustrates an example of CT image enhancement with Poisson noise, in accordance with one or more embodiments of the disclosed subject matter. With reference to FIGS. 1-6, sinogram images 601-604 correspond to synthetic artifact representations, respectively. With reference to FIGS. 1-6, sinogram images 601-604 correspond to synthetic artifact representations of different levels (e.g., low, medium, high, and very high) of Poisson noise artifacts generated by simulation component 104 in the sinogram domain using water simulation model 104. In various embodiments, each of the synthetic artifact representations illustrated in sinogram images 601-604 may be included in synthetic artifact data 118 (or more specifically, Poisson noise data 506 collection of FIG. 5) as a candidate artifact to apply to any CT image data for augmentation purposes. In the illustrated embodiment, each Poisson noise artifact was added (e.g., by data augmentation component 106) to a CT image of a pelvis. Image 605 corresponds to an original CT image without added noise. Image 606 depicts original CT image 605 with a low Poisson noise artifact added (e.g., corresponding to sinogram image 601), image 607 depicts original CT image 605 with a medium Poisson noise artifact added (e.g., corresponding to sinogram image 602), image 608 depicts original CT image 605 with a high Poisson noise artifact added (e.g., corresponding to sinogram image 603), and image 609 depicts original CT image 605 with a very high Poisson noise artifact added (e.g., corresponding to sinogram image 604). In this regard, each of images 606-609 corresponds to an exemplary enhanced CT image that may be generated by data augmentation component 108 using synthetic artifact data 118 and included in enhanced CT image data 120.

[0043] FIG. 7 illustrates an example of CT image enhancement with Gaussian noise in accordance with one or more embodiments of the disclosed subject matter. With reference to FIGS. 1-5 and 7, sinogram images 701-704 correspond to synthetic artists, respectively. With reference to FIGS. 1-5 and 7, sinogram images 701-704 correspond to synthetic artifact representations of different levels (e.g., low, medium, high, and very high) of Gaussian noise artifacts generated by simulation component 104 in the sinogram domain using simulation model 106. In various embodiments, each of the synthetic artifact representations illustrated in sinogram images 701-704 can be included in synthetic artifact data 118 (or more specifically, Gaussian noise data 508 collection of FIG. 5) as a candidate artifact to apply to any CT image data for enhancement purposes. In the illustrated embodiment, each Gaussian noise artifact was added (e.g., by data augmentation component 108) to a CT image of a pelvis. Image 705 corresponds to an original CT image without added noise. Image 706 depicts original CT image 705 with a low Gaussian noise artifact added (e.g., corresponding to sinogram image 701), image 707 depicts original CT image 705 with a medium Gaussian noise artifact added (e.g., corresponding to sinogram image 702), image 708 depicts original CT image 705 with a high Gaussian noise artifact added (e.g., corresponding to sinogram image 703), and image 709 depicts original CT image 705 with a very high Gaussian noise artifact added (e.g., corresponding to sinogram image 704). In this regard, each of images 706-709 corresponds to an exemplary enhanced CT image that may be generated by data augmentation component 108 using synthetic artifact data 118 and included in enhanced CT image data 120.

[0044] 8 illustrates an exemplary CT image enhancement with randomly inserted metal artifacts in accordance with one or more embodiments of the disclosed subject matter. FIG. 8 illustrates a plurality of exemplary enhanced CT images comprised of synthetic metal artifacts generated using the simulation model 106 and integrated onto or into the CT image by the data augmentation component 108. The enhanced CT images illustrated in FIG. 8 correspond to additional exemplary enhanced CT images that may be generated by the data augmentation component 108 using synthetic artifact data 118 (e.g., metal object synthetic artifact data in particular) and included in the enhanced CT image data 120.

[0045] Each enhanced CT image is different with respect to either the depicted anatomical region (e.g., including different slices of the pelvis, abdomen, and head), the type of inserted metal object, and / or the location / orientation of the inserted metal object. The synthetic metal object artifacts in each image correspond to white objects. As shown in FIG. 8, in some embodiments, the data augmentation component 108 can randomly insert (e.g., apply, combine, add, etc.) synthetic metal artifacts on or within the CT images at different positions and / or locations, regardless of the object type and CT image type (i.e., depicted anatomical region), to create an augmented set of CT images with artifact variations. Although some randomly augmented CT images have artifact variations that are not observed in clinical practice (e.g., see the bottom right image with a metallic dental implant outside the body and the second row of center images with a triangular metallic dot arrangement outside the body), integrating these random artifact variations into a training dataset for training a medical image inference model has been shown to improve the robustness and accuracy of the model.

[0046] Additionally, in some embodiments, the data augmentation component 108 can employ ground truth (GT) label data associated with the CT images to facilitate target positioning of the composite artifact in relation to combining the artifact with the CT images at a desired location and / or orientation. In this regard, the data augmentation component 108 can align one or more CT image artifacts to a target position on one or more CT images using GT region of interest (ROI) data associated with the one or more CT images.

[0047] FIG. 9 illustrates targeted artifact integration in CT using ground truth labeling according to one or more embodiments of the disclosed subject matter. With reference to FIGS. 1-9, image 901 depicts ground truth (GT) labels indicating regions of interest (ROIs) associated with a corresponding CT image depicted in image 902. The GT labels in this example are illustrated against a black background to show that the GT label data is separable (e.g., as metadata, etc.) from the underlying CT image data. For example, many medical image inference models employ training datasets annotated with GT labels in conjunction with the supervised or semi-supervised machine learning training paradigms. One example of this includes segmentation model training datasets employing GT labels applied to target ROIs to be segmented by a segmentation model, such as an organ segmentation model, a lesion segmentation model, etc. GT labels are typically applied by a medical professional in conjunction with viewing training CT images and applying manual markups to rendered CT images via a medical image annotation and viewing application. Once applied, the location and shape of the GT labels relative to the corresponding image data can be calculated in 2D and / or 3D and can be decoupled from the underlying image while serving as a relational guide indicating the target ROI for segmentation, etc. In the example shown in FIG. 9, the target RIO indicates the location of the femur for segmentation. In general, these GT labels indicate locations and / or objects that are relevant to the inference task of the model being trained. Thus, these locations and / or objects correspond to critical locations where the presence of artifacts has been shown to adversely affect the accuracy and specificity of the model, making it more difficult for the model to identify discrepancies caused by artifacts.In this regard, incorporating augmented training images with synthetic artifacts inserted at or near GT-labeled locations can significantly improve the robustness of the model to artifacts. In the example shown in FIG. 9, the GT labels indicating the relative location of the femur in the corresponding CT image shown in image 902 were used (e.g., by the data augmentation component 108) as a location guide for inserting a synthetic metal object 903 into a target ROI (e.g., a hip prosthetic where the patient is expected to have a femur bone). Thus, in some embodiments where the CT image data to be augmented (e.g., included in the CT image data 130) includes or is otherwise associated with GT labels indicating one or more ROIs, the data augmentation component 108 can employ the GT labels as a guide for aligning and positioning one or more synthetic artifacts relative to the GT labels.

[0048] FIG. 10 illustrates several exemplary metal artifact CT image enhancement operations facilitated by the data augmentation component in accordance with one or more embodiments of the disclosed subject matter. With reference to FIGS. 1-10, as described above, metal artifact CT image enhancement is facilitated by the metal artifact CT image enhancement component. As described above, because the synthetic artifact data 118 is anatomy agnostic and pre-generated, the data augmentation component 108 can easily and efficiently apply one or more of the corresponding synthetic CT artifacts contained therein to any CT image regardless of the anatomy depicted. Additionally, the data augmentation component 108 can enhance the diversity of the augmented CT image by adjusting the position, orientation, and / or size (among other image manipulation techniques) of the synthetic artifact. In this regard, the data augmentation component 108 can combine the same synthetic artifact with each copy of the same CT image, but integrate artifacts at different positions, orientations, and / or different sizes in each copy in conjunction with generating the augmented CT image.

[0049] For example, FIG. 10 illustrates a swapping and rotating operation of a metal object artifact facilitated by the data augmentation component 108. Images 1001A, 1001B, and 1001C correspond to different augmented versions of a first CT image, and images 1002A, 1002B, and 1002C correspond to different augmented versions of a second CT image. In particular, reference is again made to FIGS. 3A and 3B. Referring to FIGS. 3A and 3B, in view of FIG. 10, image 1001A corresponds to a version of CT image 301 generated according to the disclosed anatomically agnostic simulation technique, as opposed to conventional simulation techniques. In this regard, image 1001A incorporates a simulated version of the metal artifact shown in image 304 onto a native CT image corresponding to image 301. Similarly, image 1002A corresponds to a version of CT image 302 generated according to the disclosed anatomically agnostic simulation technique, as opposed to conventional simulation techniques. In this regard, image 1002A incorporates a simulated version of the metal artifact depicted in image 305 onto the native CT image corresponding to image 302. As can be seen by comparing images 1001A and 1002A with the conventional simulated images 301 and 302, respectively, the disclosed anatomical artifact agnostic artifact simulation and data augmentation techniques produce much more realistic and higher quality augmented CT images compared to conventional simulation techniques. In addition, since the metal artifact is simulated independently of the actual CT image, the metal artifact can be easily swapped, rotated, and resized to add more variety for data augmentation purposes. For example, images 1001B and 1002B swap the respective metal artifacts depicted in images 1001A and 1002B.In another example, image 1001C depicts a version of image 1001B with the metal artifact (corresponding to the pair of open circles) rotated approximately 100 degrees counterclockwise relative to the depicted anatomical structure. In another example, image 1002C depicts a version of image 1002A with the metal artifact (corresponding to the pair of open circles) rotated approximately 20 degrees counterclockwise relative to the depicted anatomical structure.

[0050] 11 illustrates an exemplary computing system 1100 (hereinafter, system 1100) equipped to augment CT images with synthetic artifacts and employ the augmented CT images for training a medical image inference model in accordance with one or more embodiments of the disclosed subject matter. System 1100 is similar to system 100 with the addition of some specific subcomponents, such as data augmentation component 108, artifact selection component 1102, positioning component 1104, insertion component 1106, and editing component 1110, and ML model data 1112 to external storage 128. Repetitive descriptions of elements as employed in each embodiment are omitted for brevity.

[0051] In various embodiments, the artifact selection component 1102 can facilitate selecting a particular pre-generated synthetic artifact included in the synthetic artifact data 118 for addition to one or more CT images (e.g., included in the CT image data 112), as described with reference to FIG. 5. In this regard, in some embodiments, one or more features and functionality of the data augmentation component 108 can implement a user-facing training data augmentation tool (e.g., in the form of a local software application / program, web application, mobile application, etc.) that allows a user (e.g., an expert, ML model developer, clinician, etc.) to review and augment training data sets for ML model training and / or other AI applications. In these embodiments, the rendering component 110 can present the user with a suitable graphical user interface that provides for selecting CT images and / or training data sets to augment, selecting and / or controlling artifact augmentation parameters regarding which synthetic artifacts to integrate into some or all of the training images and where, and reviewing the results of the augmented CT images. The positioning component 1104 can provide positioning control tools that allow a user to perform operations to control the positioning of the composite artifact within the CT image, including swapping operations, rotating operations, resizing operations, etc. The positioning component 1104 can further perform targeted artifact insertion using the corresponding GT label, as described above with reference to FIGURE 9. The editing component 1108 can provide additional editing operations on the enhanced CT image, such as various fine-tuning editing operations and other post-processing operations.The insertion component 1106 may perform any operation related to adding, inserting, overlaying and / or otherwise combining a synthetic representation of an artifact included in the synthetic artifact data 118 onto or within the CT image data 112 (e.g., within the sinogram domain and / or image domain) to generate enhanced CT image data 120.

[0052] The training component 1110 can further use the augmented CT image data 120 in connection with training one or more ML models corresponding to one or more medical image inference models. In this regard, in various embodiments, the augmented CT image data 120 can include or be otherwise associated with a training dataset that includes both native CT images and augmented CT images that include one or more synthetic artifacts applied by the data augmentation component 108 in accordance with the disclosed techniques. The training component 1110 can further train and develop ML models using the training dataset according to various suitable machine learning processes (e.g., supervised training, unsupervised training, semi-supervised training, etc.) and protocols (e.g., training, testing and validation phases).

[0053] 12 illustrates a block diagram of an exemplary, non-limiting computer-implemented method 1200 for generating an enhanced CT image consisting of synthetic artifacts in accordance with one or more embodiments of the disclosed subject matter. Repetitive descriptions of elements as employed in each embodiment are omitted for brevity.

[0054] Method 1200 includes, at 1202, generating (e.g., via simulation component 104) by a system including a processor (e.g., computing system 100 or computing system 1100) using an anatomical agnostic simulation model (e.g., simulation model 106) synthetic artifact data (e.g., synthetic artifact data 118), the synthetic artifact data corresponding to one or more CT image artifacts. At 1204, method 1200 further includes generating, by the system, an enhanced CT image (e.g., via data augmentation component 108) using the synthetic artifact data including the one or more computed tomography image artifacts (e.g., enhanced CT image data 120). In some implementations, method 1200 may further include rendering or displaying the enhanced CT image via a display associated with computing system 100 and / or transmitting (e.g., via a wired or wireless communication network) the enhanced CT image to a remote display device.

[0055] 13 shows a block diagram of an exemplary non-limiting computer-implemented method 1300 for developing a medical image inference model with artifact robustness according to one or more embodiments of the disclosed subject matter. Repeated descriptions of similar elements employed in each embodiment are omitted for brevity.

[0056] The method 1300, at 1302, includes integrating, by a system including a processor (e.g., computing system 1100), one or more synthetic artifacts with a CT image (e.g., via data augmentation component 108), resulting in an augmented CT image. The one or more synthetic artifacts include one or more representations of the one or more CT artifacts simulated in a simulated material having low radiation intensity (e.g., water, fat, soft tissue, blood, or similar material). At 1304, the method 1300 further includes training, by the system (e.g., via training component 1110), a medical image inferencing model to perform an inferencing task using the augmented CT images as training images. In various embodiments, the training images can include enhanced CT images (e.g., with synthetic artifacts) as well as CT images without synthetic artifacts, which can correspond to the original CT images before enhancement and / or additional CT images. As a result of incorporating the enhanced CT images in the training set, the resulting trained model has improved robustness (e.g., accuracy and / or specificity) to CT images with artifacts compared to a version of the model trained without the enhanced CT images.

[0057] In this regard, the effectiveness of the disclosed technique was tested and verified in relation to improving the robustness of the CT femur left-right segmentation bone model. In this experiment, a training dataset consisting of multiple different CT images of the femur labeled with GT ROI data was supplemented with different synthetic artifacts generated using the simulation model 106 in the sinogram domain. The synthetic artifacts included synthetic metal objects that were targeted and inserted at or near the femur as described with reference to FIG. 9. The synthetic artifacts also included different levels and combinations of Poisson noise and Gaussian noise artifacts. The resulting training dataset included both augmented images (e.g., with synthetic artifacts generated using the disclosed technique) and native CT images. The training, testing, and validation stages were performed with both augmented and native images. The performance results of the segmentation model on the augmented CT images containing Poisson noise, Gaussian noise, and metal artifacts showed that the integration of the augmented CT data resulted in improved model performance.

[0058] One or more embodiments may be a system, a method, and / or a computer program product at any possible level of technical detail integration. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to execute one or more aspects of the present embodiments.

[0059] A computer-readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), static random access memories (SRAMs), portable compact disk read-only memories (CD-ROMs), digital versatile disks (DVDs), memory sticks, floppy disks, mechanically encoded devices such as punch cards or ridge structures in grooves with instructions recorded thereon, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not intended to be understood as a transitory signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., light pulses passing through a fiber optic cable), or electrical signals transmitted over wires.

[0060] The computer readable program instructions described herein can be downloaded to a respective computing / processing device from a computer readable storage medium or to an external computer or external storage device over a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include a copper transmission cable, an optical transmission fiber, a wireless transmission, a router, a firewall, a switch, a gateway computer and a network adapter in the local area network computing / processing device. The network interface transfers the computer readable program instructions from the network to the computer readable storage medium in the wide area network / processing device for storage.

[0061] The computer readable program instructions for carrying out the operations of the present invention can be either assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, configuration data for an integrated circuit, or source or object code written in any combination of one or more programming languages, including object oriented programming languages ​​such as Smalltalk, C++, and procedural programming languages ​​such as the "C" programming language. The computer readable program instructions can be executed entirely on the entity's computer, partially on the entity's computer, as a stand-alone software package, partially on the entity's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer can be connected to the entity's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be to an external computer (e.g., via the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) can utilize state information of computer readable program instructions to execute computer readable program instructions and personalize the electronic circuitry to perform aspects of the present invention.

[0062] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0063] These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to manufacture a machine, such that the instructions, executed via the processor of the computer or other programmable data processing apparatus, create means for performing the functions / acts specified in the flowchart and / or block diagram blocks or blocks. These computer readable program instructions may also be stored on a computer readable storage medium capable of directing a computer, programmable data processing apparatus, and / or other apparatus to function in a particular manner, such that the computer readable storage medium having instructions stored therein constitutes an article of manufacture including instructions implementing aspects of the functions / acts specified in the flowchart and / or block diagram blocks or blocks.

[0064] The computer readable program instructions may be loaded into a computer, other programmable data processing apparatus, or other device to cause the computer, other programmable apparatus, or other device to execute a series of operational steps to generate a computer-implemented process, such that the instructions executing on the computer, other programmable apparatus, or other device perform the functions / acts specified in the flowchart and / or block diagram blocks or blocks.

[0065] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of instructions constituting one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may be executed in the reverse order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flowchart illustrations, as well as combinations of blocks in the block diagrams and / or flowchart illustrations, may be implemented by a special purpose hardware-based system that performs the specified functions or acts, or executes a combination of special purpose hardware and computer instructions.

[0066] 14, the systems and processes described below may be embodied in hardware, such as a single integrated circuit (IC) chip, multiple ICs, application specific integrated circuits (ASICs), etc. Additionally, the order in which some or all of the process blocks appear in each process should not be considered limiting. Rather, it should be understood that some of the process blocks may be performed in various orders, not all of which may be explicitly illustrated herein.

[0067] Referring to Figure 14, an exemplary environment 1400 for implementing various aspects of the claimed subject matter includes a computer 1402. The computer 1402 includes a processing unit 1404, a system memory 1406, a codec 1435, and a system bus 1408. The system bus 1408 couples system components, including but not limited to the system memory 1406, to the processing unit 1404. The processing unit 1404 may be any of a variety of processors available. Dual microprocessors and other multi-processor architectures may also be employed as the processing unit 1404.

[0068] The system bus 1408 can be any of several types of bus structures, including a memory bus or memory controller, a peripheral bus or external bus, or a local bus using a variety of available bus architectures, including, but not limited to, Industrial Standard Architecture (ISA), Micro-Channel Architecture (MSA), Extended ISA (EISA), Intelligent Drive Electronics (IDE), VESA Local Bus (VLB), Peripheral Component Interconnect (PCI), CardBus, Universal Serial Bus (USB), Advanced Graphics Port (AGP), Personal Computer Memory Card International Association Bus (PCMCIA), and the like. Association), Firewire (IEEE13144), and Small Computer Systems Interface (SCSI: Small Computer Systems Interface).

[0069] The system memory 1406 includes a volatile memory 1410 and a non-volatile memory 1412, and in various embodiments may employ one or more of the disclosed memory architectures. A basic input / output system (BIOS), containing the basic routines for transferring information between elements within the computer 1402, such as during start-up, is stored in the non-volatile memory 1412. Further, in accordance with the present invention, the codec 1435 may include at least one of an encoder or a decoder, which may be configured in hardware, software, or a combination of hardware and software. Although the codec 1435 is depicted as a separate component, the codec 1435 may be included within the non-volatile memory 1412. By way of example and not limitation, the non-volatile memory 1412 may include a read only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, a 3D flash memory, or a resistive memory, such as a resistive random access memory (RRAM). The non-volatile memory 1412, in at least some embodiments, can employ one or more of the disclosed memory devices. Moreover, the non-volatile memory 1412 can be computer memory (e.g., physically integrated with the computer 1402 or its main board), or removable memory. Examples of suitable removable memory with which the disclosed embodiments can be implemented can include Secure Digital (SD) cards, Compact Flash (CF) cards, Universal Serial Bus (USB) memory sticks, and the like. The volatile memory 1410 includes random access memory (RAM) that acts as an external cache memory, and can also employ one or more of the disclosed memory devices in various embodiments.By way of example and not limitation, RAM is available in many forms, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), and Enhanced SDRAM (ESDRAM).

[0070] The computer 1402 may also include removable / non-removable, volatile / non-volatile computer storage media. FIG. 14 illustrates, for example, a disk storage device 1416. The disk storage device 1416 includes, but is not limited to, devices such as a magnetic disk drive, a solid-state disk (SSD), a flash memory card, or a memory stick. Additionally, the disk storage device 1416 may include storage media separately or in combination with other storage media, including, but not limited to, an optical disk drive, such as a compact disk ROM drive (CD-ROM), a CD recordable drive (CD-R drive), a CD rewriteable drive (CD-RW drive), or a digital versatile disk ROM drive (DVD-ROM). To facilitate connection of the disk storage device 1416 to the system bus 1408, a removable or non-removable interface, such as the interface 1414, is typically used. It is understood that the disk storage device 1416 may store information related to an entity. Such information may be stored on a server or provided to an application running on the entity device. In one embodiment, the entity may be notified (e.g., by one or more output devices 1436) of the type of information to be stored on disk storage 1416 or transmitted to the server or application. The entity may be provided with an opportunity to opt-in or opt-out of such information being collected or shared with the server or application (e.g., by input from one or more input devices 1428).

[0071] It should be appreciated that Figure 14 describes software that acts as an intermediary between the entities described in the preferred operating environment 1400 and the basic computer resources. Such software includes an operating system 1418. The operating system 1418 may be stored on disk storage 1416 and serves to control and allocate resources of the computer system 1402. Applications 1420 take advantage of the management of resources by the operating system 1418 through computer system program modules 1424 in the system memory 1406 or on the disk storage 1416, and program data 1426, such as boot / shutdown transaction tables. The operating system may be understood as a combination of the operating system and the system memory of the computer system.

[0072] An entity may provide commands or information to the computer 14 operating system via one or more input devices 1428, including, but not limited to, pointing devices such as a mouse, trackball, stylus, touch pad, keyboard, microphone, joystick, game pad, satellite dish, scanner, TV tuner card, digital camera, digital video camera, web camera, and the like. These and other input devices connect to the processing unit 1404 through the system bus 1408 via one or more interface ports 1430. The one or more interface ports 1430 may include, for example, a serial port, a parallel port, a game port, and a universal serial bus (USB). The one or more output devices 1436 may use some of the same types of ports as the one or more input devices 1428. Thus, for example, a USB port may be used to provide input to the computer 1402 and to output information from the computer 1402 to the output device 1436. One or more output adapters 1434 are provided to illustrate that there are output devices 1436 that require special adapters, such as monitors, speakers, printers, among other output devices 1436. The output adapters 1434 include, by way of example and not limitation, video cards and sound cards that provide a means of connection between the output device(s) 1436 and the system bus 1408. It should be noted that the other device or system of devices provides both input and output capabilities, such as the remote computer 1438.

[0073] The computer 1402 can operate in a networked environment using logical connections to one or more remote computers, such as one or more remote computers 1438. The remote computer 1438 can be a personal computer, a server, a router, a network PC, a workstation, a microprocessor based appliance, a peer device, a smart phone, a tablet, or other network node, and typically includes many of the elements described in conjunction with the computer 1402. For simplicity, only memory storage 1440 is illustrated with the remote computer 1438. The remote computer 1438 is logically connected to the computer 1402 through a network interface 1442 and is connected via a communication connection 1444. The network interface 1442 encompasses wired or wireless communication networks, such as a local area network (LAN), a wide area network (WAN), and a cellular network. LAN technologies include Fiber Distributed Data Interface (FDDI), Copper Distributed Data Interface (CDDI), Ethernet, Token Ring, etc. WAN technologies include, but are not limited to, point-to-point links, circuit-switched networks such as Integrated Services Digital Networks (ISDN), packet-switched networks, and Digital Subscriber Lines (DSL).

[0074] The one or more communications connections 1444 refer to the hardware / software employed to connect the network interface 1442 to the bus 1408. The communications connections 1444 are shown internal to the computer 1402 for illustrative clarity, but may also be external to the computer 1402. The hardware / software required to connect to the network interface 1442 includes, by way of example only, internal and external technology such as modems, including ordinary telephone-grade modems, cable modems and DSL modems, ISDN adapters, and wired and wireless Ethernet cards, hubs, and routers.

[0075] The illustrated aspects of the disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

[0076] Although the subject matter has been described above in the general context of computer-executable instructions for a computer and / or a computer program product executing on a computer, those skilled in the art will recognize that the present disclosure can also be implemented or can be implemented in combination with other program modules. Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks and / or implement particular abstract data types. Furthermore, those skilled in the art will appreciate that the computer-implemented method of the present invention can be implemented with other computer system configurations, including single-processor or multi-processor computer systems, minicomputer devices, mainframe computers, as well as computers, handheld computer devices (e.g., PDAs, phones), microprocessor-based or programmable consumer or industrial electronic devices, and the like. The illustrated aspects can also be implemented in distributed computing environments where tasks are performed by remote processing devices linked through a communications network. However, some, if not all aspects of the present disclosure can be implemented on stand-alone computers. In a distributed computing environment, program modules can be located in both local and remote memory storage devices.

[0077] As used herein, terms such as "component," "system," "subsystem," "platform," "layer," "gateway," "interface," "service," "application," "device," and the like, can refer to and / or include one or more computer-related entities, or entities associated with an operating machine having one or more specific functionalities. The entities disclosed herein can be either hardware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, a thread of execution, a program, and / or a computer. By way of illustration, both an application running on a server and the server can be a component. One or more components can reside within a process and / or thread of execution, and a component can be localized on one computer and / or distributed among two or more computers. In another example, each component can execute from various computer-readable media having various data structures stored thereon. The components can communicate via local and / or remote processes, such as following a signal having one or more data packets (e.g., data from one component interacting with another system over a network such as the Internet via a local system, a distributed system, and / or a signal). As another example, a component may be a device having a particular function provided by mechanical parts operated by electrical or electronic circuits that are operated by a software or firmware application executed by a processor. In such a case, the processor may be internal or external to the device and may execute at least a portion of the software or firmware application.As yet another example, a component may be a device that provides a particular functionality through electronic components without mechanical components, where the electronic components may include a processor or other means for executing software or firmware that provides at least a portion of the functionality of the electronic component. In an aspect, a component may emulate an electronic component via a virtual machine, for example in a cloud computing system.

[0078] Furthermore, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the natural inclusive permutations. That is, if X employs A, if X employs B, if X employs both A and B, then "X employs A or B" applies to any of the foregoing. Furthermore, the articles "a" and "an" as used herein and in the accompanying drawings should generally be construed to mean "one or more" unless otherwise specified or clear from the context to be directed to the singular form. As used herein, the terms "example" and / or "exemplary" are utilized to mean serving as an example, instance, or illustration and are intended to be non-limiting. For the avoidance of doubt, the subject matter disclosed herein is not limited by such examples. Additionally, any aspect or design described herein as "example" and / or "exemplary" is not necessarily to be construed as preferred or advantageous over other aspects or designs, and is not intended to exclude equivalent exemplary structures and techniques known to those skilled in the art.

[0079] As used herein, the term "processor" may refer to substantially any computing processing unit or device, including, but not limited to, single-core processors, single processors with software multithreaded execution capabilities, multi-core processors, multi-core processors with software multithreaded execution capabilities, multi-core processors with hardware multithreading techniques, parallel platforms, and parallel platforms with distributed shared memory. Additionally, a processor may refer to an integrated circuit, an application specific integrated circuit (ASIC), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic controller (PLC), a complex programmable logic device (CPLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. Additionally, a processor may utilize nanoscale architectures, including, but not limited to, molecular and quantum dot-based transistors, switches, and gates to optimize space usage or improve the performance of the entity equipment. A processor may also be implemented as a combination of computing processing units. In this disclosure, terms such as "memory," "storage," "data storage," "database," and substantially any other information storage component associated with the operation and functionality of a component are utilized to refer to a "memory component," an entity embodied in a "memory," or a component that comprises a memory. It should be understood that the memory and / or memory components described herein can be either volatile or non-volatile memory, or can include both volatile and non-volatile memory.By way of example and not limitation, non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), flash memory, or non-volatile random access memory (RAM) (e.g., ferroelectric RAM (FeRAM)). Volatile memory may include, for example, a RAM acting as an external cache memory. By way of example and not limitation, RAM is available in many forms, such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), SyncLink DRAM (SLDRAM), Direct Rambus RAM (DRRAM), Direct Rambus Dynamic RAM (DRDRAM), and Rambus Dynamic RAM (RDRAM). Additionally, it is intended that the memory components of the systems or computer-implemented methods disclosed herein include, without being limited to, these and any other suitable types of memory.

[0080] What has been described above includes merely examples of systems and computer-implemented methods. Of course, it is not possible to describe every conceivable combination of components or computer-implemented methods for purposes of describing this disclosure, but one of ordinary skill in the art can recognize that many further combinations and permutations of the present disclosure are possible. Furthermore, to the extent that terms such as "includes," "has," "possesses," and the like are used in the detailed description, claims, appendices, and drawings, such terms are intended to be inclusive in a manner similar to the term "comprising" interpreted when employed as transitional words in the claims. The description of the various embodiments is presented for illustrative purposes, but is not intended to be an exhaustive or limiting description of the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terms used herein have been selected to best explain the principles of the embodiments, practical applications or technical improvements to the art found in the marketplace, or to enable those of ordinary skill in the art to understand the embodiments disclosed herein. [Explanation of symbols]

[0081] 100: Computing system 102: Computer executable component 104: Simulation component 106: Simulation model 108: Data augmentation component 110: Rendering component 112: Input / output device 114: Processing unit 116: Storage device 118: Synthetic artifact data 120: Augmented CT image data 122: Communication component 124: Memory 126: System bus 128: External storage device 130: CT image data 500: Workflow 502: Head CT data 504: Pelvis CT data 506: Poisson noise data 508: Gaussian noise data 510: Metal artifact data 512: Other artifact data 514: Augmented head CT data 516: Augmented pelvis CT data 1100: Computing system 1102: Artifact selection component 1104: Positioning component 1106: Insertion component 1108: Editing component 1110: Training component 1112:ML model data 1400:Environment 1402:Computer 1404:Processing unit 1406:System memory 1408:System bus 1410:Volatile memory 1412:Non-volatile memory 1414:Interface 1416:Disk storage 1418:Operating system 1420:Application 1424:Module 1426:Data 1428:One or more input devices 1430:One or more interface ports 1434:One or more output adapters 1435:Codec 1436:One or more output devices 1438:One or more remote computers 1440:Memory storage device 1442:Online store interface 1444:One or more communication connections

Claims

1. a memory for storing computer executable components; a processor for executing the computer-executable components stored in the memory, The computer executable components include: a simulation component for generating synthetic artifact data corresponding to artifacts of one or more computed tomography images, the simulation component generating the synthetic artifact data in the sinogram domain using a simulation model that is agnostic to anatomical regions of the body; a data augmentation component that uses the composite artifact data to generate an augmented computed tomography image that includes artifacts of the one or more computed tomography images; Including, The augmented computed tomography images are used as training images to train a medical image inference model.

2. The system of claim 1 , wherein the synthetic artifact data comprises a synthetic representation of one or more simulated computed tomography image artifacts in a material selected from the group consisting of a water material, a fat material, a soft tissue material, and a blood material.

3. The system described in claim 1, wherein the sinogram domain representation of the synthetic artifact data is converted to an image domain representation of the synthetic artifact data using a CT reconstruction algorithm.

4. The system of claim 1 , wherein the simulation component generates the synthetic artifact data in an image domain.

5. The system of claim 1 , wherein the data augmentation component combines the synthetic artifact data with one or more computed tomography images to generate the augmented computed tomography images.

6. 6. The system of claim 5, wherein the data augmentation component aligns artifacts of the one or more computed tomography images to target locations on the one or more computed tomography images using ground truth region of interest data associated with the one or more computed tomography images.

7. The system of claim 5 , wherein the composite artifact data is agnostic to anatomical regions of the body, and the one or more computed tomography images include a plurality of medical images corresponding to different anatomical regions of the body.

8. 6. The system of claim 5, wherein the artifacts of the one or more computed tomography images include a plurality of different types of artifacts, and the data augmentation component combines two or more of the plurality of different types of artifacts into a same computed tomography image in connection with generating the augmented computed tomography image.

9. 6. The system of claim 5, wherein the data augmentation component combines artifacts of the one or more computed tomography images with computed tomography images acquired during the same examination but at different locations or orientations in generating the augmented computed tomography image.

10. The system of claim 1 , wherein the artifacts in the one or more computed tomography images include a plurality of different types of artifacts.

11. The system of claim 10 , wherein the plurality of different types of artifacts are selected from the group consisting of metal object artifacts, Poisson noise artifacts, and Gaussian noise artifacts.

12. The system of claim 1 , wherein the artifacts in the one or more computed tomography images include a plurality of noise artifacts each corresponding to a different noise level.

13. The computer executable components further include: The system of claim 1 further comprising a training component that trains the medical image inference model.

14. a system including a processor generating synthetic artifact data in a sinogram domain corresponding to artifacts of one or more computed tomography images, the synthetic artifact data including a synthetic representation of the artifacts of the one or more computed tomography images that is agnostic to anatomical regions of the body; the system using the composite artifact data to generate an augmented computed tomography image including artifacts of the one or more computed tomography images; Including, The method, wherein the augmented computed tomography images are used as training images to train a medical image inference model.

15. 15. The method of claim 14, wherein the anatomical region-agnostic synthetic representation of the body represents one or more simulated computed tomography image artifacts in a material selected from the group consisting of water material, fat material, soft tissue material, and blood material.

16. The method of claim 14, comprising converting the sinogram domain representation of the synthetic artifact data into an image domain representation of the synthetic artifact data using a CT reconstruction algorithm.

17. 15. The method of claim 14, wherein generating the augmented computed tomography image comprises combining the composite artifact data with one or more computed tomography images to generate the augmented computed tomography image.

18. 20. The method of claim 17, wherein generating the augmented computed tomography images further comprises the system using ground truth region of interest data associated with the one or more computed tomography images to align artifacts of the one or more computed tomography images to target locations on the one or more computed tomography images.

19. The method of claim 14 , wherein the system includes training a medical image inference model to perform an inference task using the augmented computed tomography images as training images.

20. A non-transitory machine-readable storage medium containing executable instructions, The executable instructions, when executed by a processor, generating synthetic artifact data in the sinogram domain using an anatomical region agnostic simulation model of the body, the synthetic artifact data corresponding to artifacts of one or more computed tomography images; using the composite artifact data to generate an augmented computed tomography image that includes artifacts of the one or more computed tomography images; It facilitates the execution of several operations, including A non-transitory machine-readable storage medium, wherein the augmented computed tomography images are used as training images to train a medical image inference model.

Images