Systems and methods for autonomous identification of heterogeneous body module regions

Abstract

A computer-implemented method includes obtaining, at a processor, a tomographic image of a phantom, where the phantom includes a heterogeneous region having a plurality of materials of varying attenuation coefficients. The method also includes automatically segmenting, via the processor, the heterogeneous region from the tomographic image to generate a segmented image. The method further includes automatically identifying, via the processor, a plurality of regions of interest within the tomographic image having varying attenuation coefficients based on the segmented image. The method still further includes automatically labeling, via the processor, each of the plurality of regions of interest as representing a particular material of the plurality of materials. The method even further includes outputting, via the processor, a labeled image of the tomographic image. The method can be used for quality assurance testing and calibration.

Description

Background Technology

[0001] The subject of this paper relates to systems and methods for autonomous identification of heterogeneous phantom regions based on deep learning.

[0002] Volumetric medical imaging techniques utilize various methods to collect three-dimensional information about the body. For example, computed tomography (CT) imaging systems measure the attenuation of X-ray beams passing through a patient from multiple angles. Based on these measurements, the computer is able to reconstruct images of parts of the patient's body that contribute to the radiation attenuation. As those skilled in the art will understand, these images are based on individual examinations of a series of angularly shifted measurements. It should be noted that CT systems generate data representing the distribution of linear attenuation coefficients of the scanned object. The data is then reconstructed to produce images typically displayed on a screen and can be printed or reproduced on film.

[0003] In CT systems, phantoms are commonly used during calibration and quality assurance (QA) testing. For homogeneous phantoms (e.g., constructed using a single material where the attenuation coefficient does not vary within the phantom's cross-section), regions within such phantoms are readily identifiable. Heterogeneous phantoms (e.g., constructed using multiple materials where the attenuation coefficient varies depending on the material within the phantom image) are widely used in QA for both single-energy and dual-energy CT scanners to determine and conform to various image quality parameters. Currently, manual intervention is required to identify regions of interest (ROIs) for different materials as an intermediate step in QA testing. Additionally, heterogeneous phantoms may be advantageous for calibration purposes in multi-energy photon-counting CT scanners. Therefore, accurate and autonomous identification of heterogeneous phantom ROIs is necessary to perform such tasks. Summary of the Invention

[0004] The following outlines certain embodiments commensurate with the scope of the initially claimed subject matter. These embodiments are not intended to limit the scope of the claimed subject matter, but rather are intended only to provide a brief overview of possible implementations. In practice, the invention may include various forms that are similar to or different from the embodiments described below.

[0005] In one embodiment, a computer-implemented method is provided. The method includes obtaining a tomographic image of a phantom at a processor, wherein the phantom includes heterogeneous regions having multiple materials with varying attenuation coefficients. The method further includes automatically segmenting the heterogeneous regions from the tomographic image via the processor to generate a segmented image. The method further includes automatically identifying multiple regions of interest (ROIs) with varying attenuation coefficients within the tomographic image based on the segmented image via the processor. The method further includes automatically labeling each of the multiple ROIs via the processor as representing a specific material among the multiple materials. The method even further includes outputting a labeled image of the tomographic image via the processor.

[0006] In one embodiment, a computer-implemented method is provided. The method includes initiating calibration of a tomographic imaging system via a processor. The method further includes acquiring a tomographic image of a phantom using the tomographic imaging system via the processor, wherein the phantom includes heterogeneous regions having multiple materials with varying attenuation coefficients. The method further includes automatically segmenting the heterogeneous regions from the tomographic image via the processor to generate a segmented image. The method further includes automatically identifying multiple regions of interest (ROIs) with varying attenuation coefficients within the tomographic image based on the segmented image via the processor. The method further includes automatically labeling each of the multiple ROIs via the processor as representing a specific material among the multiple materials. The method further includes calculating a corresponding calibration vector for one or more of the multiple ROIs via the processor based on the label of each ROI.

[0007] In one embodiment, a computer-implemented method is provided. The method includes initiating a quality assurance test of a tomographic imaging system via a processor. The method includes obtaining a tomographic image of a phantom using the tomographic imaging system via the processor, wherein the phantom includes heterogeneous regions having multiple materials with varying attenuation coefficients. The method further includes automatically segmenting the heterogeneous regions from the tomographic image via the processor to generate a segmented image. The method further includes automatically identifying multiple regions of interest (ROIs) with varying attenuation coefficients within the tomographic image based on the segmented image via the processor. The method further includes automatically labeling each of the multiple ROIs via the processor as representing a specific material among multiple materials. The method further includes calculating an image quality metric via the processor based on the labeling of each ROI. Attached Figure Description

[0008] These and other features, aspects, and advantages of the subject matter of the invention will be better understood when reading the following detailed description with reference to the accompanying drawings, in which like reference numerals denote like parts throughout the drawings, wherein:

[0009] Figure 1 A schematic diagram of an embodiment of a computed tomography (CT) system configured to acquire and process CT images of a patient, according to various aspects of this disclosure;

[0010] Figure 2 A schematic diagram of an X-ray source and a multi-row X-ray detector (e.g., as observed in the XY plane) for an imaging heterophantom according to various aspects of this disclosure;

[0011] Figure 3 A flowchart of a method for identifying regions within a heterogeneous phantom according to various aspects of this disclosure;

[0012] Figure 4 Examples of tomographic images with static lookup tables and markings or annotations according to various aspects of this disclosure;

[0013] Figure 5 Examples of tomographic images with dynamic lookup tables and markings or annotations according to various aspects of this disclosure;

[0014] Figure 6 Examples of charts including coordinates, indices, and material descriptions according to various aspects of this disclosure;

[0015] Figure 7 For the purposes of this disclosure Figure 3 A schematic diagram of the method described in the diagram;

[0016] Figure 8 A flowchart of a method for training a deep neural network to identify regions within a heterogeneous model, according to various aspects of this disclosure;

[0017] Figure 9 This is a schematic diagram illustrating the generation of training data from unenhanced images according to various aspects of this disclosure;

[0018] Figure 10 A schematic diagram illustrating various aspects of this disclosure for generating further training data;

[0019] Figure 11 Examples of training data for image label pairs generated from unenhanced images according to various aspects of this disclosure;

[0020] Figure 12 A flowchart illustrating a method for performing a calibration process utilizing an ROI location algorithm, according to various aspects of this disclosure; and

[0021] Figure 13 This is a flowchart of a method for performing a QA testing process utilizing an ROI location algorithm, according to various aspects of this disclosure. Detailed Implementation

[0022] One or more specific implementations will be described below. To provide a concise description of these implementations, not all features of an actual implementation will be described in this specification. It should be understood that in the development of any such actual implementation, as in any engineering or design project, many implementation-specific decisions must be made to achieve the developer's specific objectives, such as complying with system-related and business-related constraints that may differ from implementation to implementation. Furthermore, it should be understood that such development efforts may be complex and time-consuming, but remain routine tasks of design, fabrication, and manufacturing for those skilled in the art who benefit from this disclosure.

[0023] When describing the elements of various embodiments of the subject matter of this invention, the articles “a,” “an,” “the,” and “the” are intended to indicate the presence of one or more of the stated elements. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that additional elements may be present in addition to the listed elements. Furthermore, any numerical examples in the following discussion are intended to be non-limiting, and therefore the additional numerical values, ranges, and percentages are within the scope of the disclosed embodiments.

[0024] Some general information is provided to give general background to the various aspects of this disclosure and to facilitate understanding and interpretation of certain technical concepts described herein.

[0025] The deep learning (DL) methods discussed herein can be based on artificial neural networks and therefore can include one or more of deep neural networks, fully connected networks, convolutional neural networks (CNNs), perceptrons, encoder-decoders, recurrent networks, wavelet filter banks, u-nets, generative adversarial networks (GANs), or other neural network architectures. Neural networks may include shortcuts, activations, batch normalization layers, and / or other features. These techniques are referred to as deep learning techniques in this paper, but the term may also be used specifically with reference to the use of deep neural networks, which are neural networks with multiple layers.

[0026] As discussed in this paper, deep learning techniques (also known as deep machine learning, hierarchical learning, or deep structured learning) are a branch of machine learning that employs mathematical representations of data and artificial neural networks used to learn and process such representations. For example, deep learning methods can be characterized as using one or more algorithms to extract or model highly abstract concepts of a class of data of interest. This can be accomplished using one or more processing layers, where each layer typically corresponds to a different level of abstraction and thus may take or utilize different aspects of the initial data or the output of the previous layer (i.e., the hierarchical or cascaded structure of the layers) as the target of the process or algorithm for a given layer. In the context of image processing or reconstruction, this can be characterized as different layers corresponding to different feature levels or resolutions in the data. Generally, processing from one representation space to the next level of representation space can be viewed as a “stage” of a process. Each stage of the process can be performed by a single neural network or by different parts of a larger neural network.

[0027] This disclosure provides systems and methods for autonomous identification (e.g., via region of interest location algorithms) of regions in a heterogeneous phantom based on deep learning. The disclosed embodiments include obtaining or acquiring (e.g., via a CT scanner) tomographic images of a phantom. The phantom comprises heterogeneous regions of multiple materials (e.g., within a rod) with varying or different attenuation coefficients. The disclosed embodiments also include automatically segmenting (e.g., via a trained deep neural network) the heterogeneous regions from the tomographic images to generate segmented images (e.g., segmentation masks). The disclosed embodiments further include automatically identifying multiple regions of interest with varying attenuation coefficients within the tomographic images based on the segmented images. Specifically, the disclosed embodiments include automatically determining the rotation angle of the phantom relative to a correctly positioned phantom (via a comparison of the segmentation mask with a reference image of the correctly positioned phantom (e.g., a reference or template segmentation mask used as ground truth)). The disclosed embodiments further include automatically labeling each of the multiple regions of interest to represent a specific material among multiple materials. The disclosed embodiments further include outputting a labeled or annotated image of the tomographic images with the labeled regions of interest. In some embodiments, the rotation angle of the phantom within the tomographic image and / or a graph containing at least labels and specific materials for each region of interest. The disclosed embodiments enable the use of multi-structure phantoms during calibration and QA processes without modifying any phantom hardware (e.g., adding tracers / tags to identify regions). Furthermore, deep neural networks can be trained with any phantom under any scanning protocol. The ability to automatically identify regions of interest within heterogeneous phantoms eliminates the need for manual intervention while ensuring the effectiveness of region identification.

[0028] Considering the foregoing discussion, Figure 1 Embodiments of an imaging system 10 for acquiring and processing image data according to various aspects of this disclosure are shown. Although the following embodiments are discussed in relation to a computed tomography (CT) imaging system, the embodiments may also be used in conjunction with other imaging systems, such as X-ray, PET, CT / PET, SPECT, nuclear CT, MRI, etc. In the illustrated embodiment, system 10 is a CT system designed to acquire X-ray projection data, reconstruct the projection data into tomographic images, and process the image data for display and analysis. The CT imaging system 10 includes an X-ray source 12. As detailed herein, source 12 may include one or more X-ray sources, such as X-ray tubes or solid-state emission structures. According to this embodiment, X-ray source 12 is configured to emit an X-ray beam 20 at one or more energies.

[0029] In some implementations, source 12 may be positioned adjacent to collimator 22, which defines the size and shape of one or more X-ray beams 20 that pass through an area located by subject 24 (e.g., a patient) or object of interest. Subject 24 attenuates at least a portion of the X-rays. The resulting attenuated X-rays 26 affect a detector array 28 formed by multiple detector elements. Each detector element generates an electrical signal representing the intensity of the X-ray beam incident at the location of the detector element when the beam irradiates detector 28. The electrical signals are acquired and processed to generate one or more scan datasets or reconstructed images.

[0030] System controller 30 commands the operation of imaging system 10 to execute examination and / or calibration protocols and process acquired data. Relative to X-ray source 12, system controller 30 provides power, focus position, control signals, etc., for X-ray examination sequences. Detector 28 is coupled to system controller 30, which commands the acquisition of signals generated by detector 28. Furthermore, via motor controller 36, system controller 30 can control the operation of linear positioning subsystem 32 and / or rotation subsystem 34 for moving components of imaging system 10 and / or subject 24. System controller 30 may include signal processing circuitry and associated memory circuitry. In such embodiments, the memory circuitry may store programs, routines, and / or coding algorithms executed by system controller 30 to operate imaging system 10 (including X-ray source 12) and process data acquired by detector 28 according to the steps and processes discussed herein. In one embodiment, system controller 30 may be implemented as all or part of a processor-based system such as a general-purpose or special-purpose computer system.

[0031] Source 12 may be controlled by an X-ray controller 38 included within system controller 30. X-ray controller 38 may be configured to provide power and timing signals to source 12. System controller 30 may include a data acquisition system (DAS) 40. DAS 40 receives data collected by readout electronics of detector 28, such as sampled analog signals from detector 28. DAS 40 may then convert the data into digital signals for further processing by a processor-based system (such as computer 42). In other embodiments, detector 28 may convert the sampled analog signals into digital signals before transmitting them to data acquisition system 40. Computer may include processing circuitry 44 (e.g., image processing circuitry). Computer 42 may include or communicate with one or more non-transitory memory devices 46 that may store data to be processed by computer 42, data to be processed by computer 42, or instructions to be executed by the processor of computer 42 (e.g., processing circuitry 44). For example, the processing circuitry 44 of computer 42 can execute one or more instruction sets stored in memory 46, which may be the memory of computer 42, the memory of a processor, firmware, or a similar instance. According to this embodiment, memory 46 stores instruction sets that, when executed by a processor, perform image processing methods as discussed herein. Memory 46 also stores one or more algorithms and / or neural networks 47 that can be used for autonomous identification of heterogeneous phantom regions, as detailed below.

[0032] Computer 42 may also be adapted to control features enabled by system controller 30 (i.e., scanning operations and data acquisition), such as responding to commands and scanning parameters provided by the operator via operator workstation 48. System 10 may also include a display 50 coupled to operator workstation 48, which allows the operator to view relevant system data, observe reconstructed images, control imaging, etc. Additionally, system 10 may include a printer 52 coupled to operator workstation 48 and configured to print images. Display 50 and printer 52 may also be connected directly or via operator workstation 48 to computer 42. Additionally, operator workstation 48 may include or be coupled to a Picture Archiving and Communication System (PACS) 54. PACS 54 may be coupled to a remote system 56, a Radiology Information System (RIS), a Hospital Information System (HIS), or to an internal or external network, enabling other personnel in different locations to access image data.

[0033] Furthermore, computer 42 and operator workstation 48 can be coupled to other output devices, which may include standard or dedicated computer monitors and associated processing circuitry. One or more operator workstations 40 can be further linked within the system for outputting system parameters, requesting checks, viewing images, etc. Generally, the monitors, printers, workstations, and similar devices provided within the system may be local to the data acquisition components, or may be located remotely, such as elsewhere within the institution or hospital, or in entirely different locations, linked to the image acquisition system via one or more configurable networks (such as the Internet, VPN, etc.).

[0034] While the foregoing discussion has addressed various exemplary components of the imaging system 10 individually, these various components may be provided within a common platform or in an interconnected platform. For example, computer 30, memory 38, and operator workstation 40 may be provided collectively as a general-purpose or special-purpose computer or workstation configured to operate according to various aspects of this disclosure. In such embodiments, the general-purpose or special-purpose computer may be provided as a separate component relative to the data acquisition components of system 10, or may be provided in a common platform having such components. Similarly, system controller 30 may be provided as part of such a computer or workstation, or as part of a separate system dedicated to image acquisition.

[0035] Considering the previous discussion of the overall imaging system 10, and turning to... Figure 2 Examples of X-ray source 12 and detector 28 (e.g., having multiple rows 58) are shown in the XY plane. As depicted, collimator 22 is disposed between X-ray source 12 and detector 28 and defines the shape of X-ray beam 20. Heterogeneous phantom 60 is disposed between X-ray source 12 and detector 28 and can be used in CT imaging systems (e.g., Figure 1 The calibration process and QA testing of system 10) are described below. The heterogeneous phantom 60 includes heterogeneous regions 62, which have various materials with varying or different attenuation coefficients. Specifically, each heterogeneous region 62 can be a rod 64 made of different materials with different attenuation coefficients. Some rods may have the same material. Some rods may have the same material but at different concentrations. Examples of materials may be calcium, iodine, adipose tissue, water, iron, tissue-like materials (e.g., brain, adipose tissue, etc.), and blood-like materials.

[0036] As described herein, techniques (e.g., region of interest location algorithms) can be used to automatically identify regions (e.g., rods) within heterogeneous phantoms that have varying or different attenuation coefficients. Specifically, these techniques utilize a holistic approach that combines system hardware and deep learning-based models to identify these regions within heterogeneous phantoms. Figure 3This is a flowchart of method 66 for identifying regions within a heterogeneous phantom. One or more steps of method 66 can be performed by... Figure 1 It is performed by one or more components (e.g., processing circuitry) of the CT imaging system or by means of a device that is separate from or remote from the CT imaging system.

[0037] Method 66 includes obtaining or acquiring tomographic images of a heterogeneous phantom (box 68). The tomographic images can be obtained using predefined protocols (e.g., kV, mA, filter, rotation speed, aperture, etc.).

[0038] Method 66 also includes automatically segmenting heterogeneous regions (e.g., multiple materials with varying or different attenuation coefficients) from tomographic images to generate segmented images (e.g., segmentation masks) (box 70). A trained deep neural network can be used to perform the segmentation of heterogeneous regions. In some embodiments, alternative techniques (e.g., if / then statements) can be used to perform the segmentation of heterogeneous regions.

[0039] Method 66 further includes automatically determining the rotation angle (e.g., angular rotation) of the heterophantom relative to a correctly positioned heterophantom (or a heterophantom at a predetermined location) (box 72). Determining the rotation angle involves comparing a segmented image derived from a tomographic image with a reference image of the correctly positioned heterophantom. The reference image can be a segmented image or a segmentation mask derived from a tomographic image of the correctly positioned heterophantom. A similarity index can be used to determine the rotation angle based on the comparison between the segmented image and the reference image. There are no limitations on the phantom rotation angle that can be determined.

[0040] Method 66 further includes automatically identifying multiple Regions of Interest (ROIs) with varying attenuation coefficients within a tomographic image based on the segmented image (i.e., based on a determined rotation angle) (box 74). The multiple ROIs correspond to heterogeneous regions within a heterogeneous phantom. Identifying ROIs includes identifying multi-connected (e.g., 8) pixel regions within the tomographic image. Identifying ROIs also includes identifying pixel regions with more than a predetermined number (e.g., 100) of connected pixels.

[0041] Method 66 further includes automatically labeling each of the multiple ROIs as representing a specific material among materials with varying or different attenuation coefficients (box 76). In some embodiments, labeling ROIs includes generating a label matrix for the multiple ROIs and comparing the label matrix with a static lookup table to label / identify each ROI as representing a specific material. Figure 4 This is an example of a static lookup table 78 and a tomographic image 80 with tags or annotations. In Figure 4 In the labeled image 80, each ROI 82 is associated with a label 84 (e.g., R1, R2, R3, etc.) in the label matrix. Figure 4In the static lookup table 78, each label 84 corresponds to a specific material. In some embodiments, labeling an ROI involves calculating the average number of CTs for each ROI (e.g., in Heinz units (HU)) and comparing the corresponding CT number with a dynamic lookup table to determine the material. The dynamic lookup table associates a specific range of CT numbers with a specific material. Figure 5 This is an example of a dynamic lookup table 78 and a tomographic image 88 with tags or annotations. In Figure 5 In this context, the average number of CT scans is determined for each ROI 90. Figure 5 In the dynamic lookup table 86, each material is associated within a specific range of CT numbers (e.g., the HU range).

[0042] return Figure 3 Method 66 even further includes outputting labeled or annotated images of the tomographic images (e.g., Figure 4 , Figure 5 The tomographic images marked in the images (80, 88) (box 92) are also output. In some embodiments, the rotation angle of the heterophantom is also output. In some embodiments, a graph is also output, which includes at least labels and markers for the specific material (or material concentration or density) of each ROI in the images. The graph may also include the coordinates of each ROI. Figure 6 The document provides an example of a chart that can be output, 94.

[0043] Figure 7 yes Figure 3 A schematic diagram of method 66 (e.g., the ROI location algorithm). Figure 3 As depicted, an image 96 of a heterogeneous phantom (e.g., an unlabeled tomographic image) is input into a trained deep neural network 98 (e.g., a trained convolutional neural network). The trained deep neural network 98 generates a segmented image 100 (e.g., a segmentation mask) from image 96, which has a heterogeneous region 102 having various materials with varying or different attenuation coefficients. The segmented image 100 is compared with a reference image or template image from which a similarity index 104 is used to determine the rotation angle of the heterogeneous phantom in image 96 relative to a correctly positioned heterogeneous phantom (or a heterogeneous phantom at a predetermined position). Regions of interest (ROIs) are identified in image 96 by recognizing connectivity regions (as indicated by reference numeral 106) and pixel regions greater than 100 pixels (as indicated by reference numeral 108). Additionally, a label matrix (as indicated by reference numeral 110) is created for the ROIs in image 96. Regions of interest in image 96 can be identified / labeled by comparing the label matrix and lookup table (as indicated by reference numeral 112 in the attached figure). The labeled or annotated image 114 of image 96 is output. All steps of the described method can be automated.

[0044] Figure 8 For use in training deep neural networks (e.g., Figure 7 The flowchart of method 116, which uses a deep neural network 98 in a matrix to identify regions within a heterogeneous model, is shown. One or more steps of method 116 can be performed by... Figure 1 It is performed by one or more components (e.g., processing circuitry) of a CT imaging system or by a device that is separate from or remote from the CT imaging system.

[0045] Method 116 includes generating training data for a deep neural network (e.g., a convolutional neural network) (box 118). Figure 9 As depicted, the initial training data can be generated by acquiring an unenhanced image 120 of a heterogeneous phantom (e.g., a tomographic image) and randomly rotating the central portion 122 of image 120 (as shown in images 123, 124) while keeping the peripheral portion 126 of image 120 intact (as shown in image 128) to generate an enhanced image or data 130 including the central portion 122 rotated relative to image 120. Multiple enhanced images 130 can be generated from image 120 using different degrees of rotation of the central portion 122. These enhanced images can be manually processed to generate labels (e.g., segmentation images or segmentation masks) for the heterogeneous regions in each enhanced image. Therefore, image label pairs (including enhanced images and segmentation images) can be generated from the enhanced images.

[0046] From these image label pairs, additional training data (different from the original training data) can be generated while training a deep neural network (e.g., a convolutional neural network). For example, in-situ data augmentation or real-time data augmentation (e.g., using the ImageDataGenerator class in the Keras deep learning network library) can be used to generate additional training data. Figure 10 As depicted, a batch of images 132 is obtained from a store of enhanced image label pairs 134, and data augmentation (as indicated by reference numeral 136) is performed to generate a batch of randomly transformed images 138, which are provided to form a larger training dataset 140 (enhanced image label pairs) for training a deep neural network. Data augmentation may include random translation and rotation, horizontal flipping, vertical flipping, width shifting, and height shifting. Additionally, noise (e.g., Gaussian noise) may be introduced into the images. Figure 11 Provide different examples of augmented image label pairs 142 obtained from unaugmented image 143. Figure 11 Each augmented image label pair 142 includes an augmented image 144 and a segmentation image 146 (e.g., a segmentation mask for heterogeneous regions).

[0047] return Figure 8Method 116 also includes obtaining training data (e.g., image label pairs) (box 148). Method 116 further includes training a deep neural network to generate a trained deep neural network (box 150) that can segment heterogeneous regions from an input image of a heterogeneous phantom. It should be noted that training can be performed for any scanning protocol.

[0048] As mentioned above, ROI location algorithms (e.g., Figure 3 Method 66) can be used during the calibration process. Figure 12 This is a flowchart of method 152 for performing a calibration process utilizing an ROI location algorithm. One or more steps of method 152 can be performed by... Figure 1 It is performed by one or more components (e.g., processing circuitry) of a CT imaging system or by a device that is separate from or remote from the CT imaging system.

[0049] Method 152 includes initiating the CT scanner system ( Figure 1 The calibration of the CT imaging system in the image is described in box 154. Method 152 further includes performing a scan of the heterophantom as described above (box 156). Method 152 further includes collecting or acquiring scan data (box 158) and performing reconstruction on the scan data to generate a tomographic image of the heterophantom (box 160). Method 152 further includes identifying one or more heterogeneous regions within the tomographic image of the heterophantom (box 162). The identification of the heterogeneous regions is performed using the ROI location algorithm described above. Method 152 further includes calculating a desired value (e.g., CT number) from one or more identified heterogeneous regions (box 164). Method 152 further includes calculating and storing the corresponding calibration vector for one or more heterogeneous regions (box 166). Method 152 includes determining whether the calibration process is complete (box 168). If not complete, method 152 returns to box 156. If complete, method 152 includes ending the calibration process (box 170).

[0050] As mentioned above, ROI location algorithms (e.g., Figure 3 Method 66) can be used in the QA testing process. Figure 13 This is a flowchart of method 172 for performing a QA testing process utilizing an ROI location algorithm. One or more steps of method 172 can be performed by... Figure 1 It is performed by one or more components (e.g., processing circuitry) of a CT imaging system or by a device that is separate from or remote from the CT imaging system.

[0051] Method 172 includes initiating the CT scanner system ( Figure 1The method includes QA testing of the CT imaging system (box 174). Method 172 also includes manually positioning the heterophantom between the radiation source and the detector of the CT scanner (box 176). Method 172 also includes performing a scan of the heterophantom (box 178). Method 172 further includes collecting or acquiring scan data (box 180) and performing reconstruction on the scan data to generate a tomographic image of the heterophantom (box 182).

[0052] Method 172 includes identifying the central region of the heterogeneous phantom (box 184). Method 172 also includes determining, based on the identification of the central region, whether the heterogeneous phantom is severely misaligned (box 186). If the heterogeneous phantom is severely misaligned, method 172 includes determining whether the misalignment can be automatically corrected (box 188). If the misalignment cannot be automatically corrected, method 172 returns to box 176. If the misalignment can be automatically corrected, method 172 includes automatically adjusting the table position in the axial plane (box 190) and / or automatically adjusting the table position in the z-direction (box 192).

[0053] If the heteromorphic phantom is not severely misaligned, method 172 includes passing the identified central region as input to an image analysis for autonomous ROI placement (e.g., utilizing the ROI location algorithm described above) (box 194). The ROI location algorithm can correct for any minor misalignment (i.e., provide fine-tuning). Method 172 also includes calculating an image quality (IQ) metric (box 196). Method 172 further includes ending the QA testing process (box 198).

[0054] The technical advantages of the disclosed embodiments include providing a system and method for autonomous identification of regions of interest (e.g., via region of interest location algorithms) based on deep learning for heterogeneous phantoms. The disclosed embodiments enable the use of multiple structures during calibration and QA processes without modifying any phantom hardware (e.g., adding tracers / tags to identify regions). Furthermore, deep neural networks can be trained on any phantom under any scanning protocol. The ability to automatically identify regions of interest within heterogeneous phantoms eliminates the need for manual intervention while ensuring the effectiveness of region identification.

[0055] Referring to the technology presented herein and protected by the claims, and applying it to physical objects and concrete examples of practical nature, which explicitly improves the present art, it is therefore not abstract, intangible, or purely theoretical. Furthermore, if any claim appended to the end of this specification contains one or more elements designated as “means for [performing]...the function” or “steps for [performing]...the function,” such elements are intended to be interpreted pursuant to Section 35, Section 112(f) of the USC. However, for any claim containing elements designated in any other manner, such elements are not intended to be interpreted pursuant to Section 35, Section 112(f) of the USC.

[0056] This written description uses examples to disclose the subject matter, including best practices, and also enables those skilled in the art to practice the subject matter, including making and using any apparatus or system and performing any included methods. The patent scope of this subject matter is defined by the claims and may include other examples that would occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that are not indistinguishable from the literal language of the claims, or if they include equivalent structural elements that have minor differences from the literal language of the claims.

Claims

1. A computer-implemented method, comprising: Tomographic images of a heterogeneous phantom are obtained at a processor, wherein the heterogeneous phantom includes a heterogeneous region having multiple materials with varying attenuation coefficients; The processor automatically segments the heterogeneous region from the tomographic image to generate a segmented image; Based on the segmented image, the processor automatically identifies multiple regions of interest with varying attenuation coefficients within the tomographic image, wherein the multiple regions of interest correspond to the heterogeneous regions within the heterogeneous phantom; The processor automatically labels each of the plurality of regions of interest as representing a specific material among the plurality of materials; as well as The processor outputs a labeled image of the tomographic image.

2. The computer-implemented method according to claim 1, comprising automatically determining the rotation angle of the phantom relative to a correctly positioned phantom via the processor.

3. The computer-implemented method according to claim 2, wherein automatically determining the rotation angle includes comparing the segmented image with a reference image of a correctly positioned phantom.

4. The computer-implemented method according to claim 2, comprising outputting the rotation angle of the phantom relative to the correctly positioned phantom via the processor.

5. The computer-implemented method of claim 1, comprising outputting a table via the processor of the specific material having at least a label and each of the plurality of regions of interest.

6. The computer-implemented method of claim 1, wherein each of the plurality of regions of interest represents a different rod within the phantom.

7. The computer-implemented method of claim 1, wherein automatically segmenting the heterogeneous region from the tomographic image comprises using a trained deep neural network to segment the heterogeneous region from the tomographic image to generate the segmented image.

8. The computer-implemented method according to claim 7, comprising: Training data comprising multiple image label pairs is obtained via the processor, wherein each image label pair comprises an enhanced tomographic image of the heterogeneous region having rotated relative to a correctly positioned phantom and a corresponding segmented image of the enhanced tomographic image; as well as The processor uses the training data to train a deep neural network to generate the trained deep neural network.

9. The computer-implemented method of claim 8, comprising generating the training data via the processor by performing enhancement on an unenhanced tomographic image, wherein the enhancement includes randomly rotating the heterogeneous region and adding noise.

10. The computer-implemented method of claim 1, wherein automatically identifying the plurality of regions of interest includes identifying multi-connected pixel regions and pixel regions with more than a predetermined number of connected pixels within the tomographic image.

11. The computer-implemented method of claim 1, wherein automatically labeling each region of interest comprises generating a label matrix for the plurality of regions of interest and comparing the label matrix with a static lookup table.

12. The computer-implemented method of claim 1, wherein automatically labeling each region of interest comprises calculating the average number of computed tomography images for each region of interest and comparing it with a dynamic lookup table, wherein the dynamic lookup table associates a particular range of computed tomography images with a particular material.

13. The computer-implemented method of claim 1, further comprising calculating, via the processor, a corresponding calibration vector for one or more of the plurality of regions of interest based on the markers of each region of interest.

14. The computer-implemented method of claim 1, comprising calculating an image quality metric via the processor based on the markers of each region of interest as part of a quality assurance test.

15. A computer-implemented method, comprising: The calibration of the tomographic imaging system begins via the processor; The processor uses the tomographic imaging system to obtain tomographic images of a heterogeneous phantom, wherein the heterogeneous phantom includes a heterogeneous region, the heterogeneous region having multiple materials with varying attenuation coefficients; The processor automatically segments the heterogeneous region from the tomographic image to generate a segmented image; Based on the segmented image, the processor automatically identifies multiple regions of interest with varying attenuation coefficients within the tomographic image, wherein the multiple regions of interest correspond to the heterogeneous regions within the heterogeneous phantom; The processor automatically labels each of the plurality of regions of interest as representing a specific material among the plurality of materials; as well as Based on the label of each region of interest, the processor calculates a corresponding calibration vector for one or more of the plurality of regions of interest.

16. The computer-implemented method of claim 15, comprising storing the corresponding calibration vector via the processor.

17. The computer-implemented method of claim 15, wherein automatically segmenting the heterogeneous region from the tomographic image comprises utilizing a trained deep neural network to segment the heterogeneous region from the tomographic image to generate the segmented image.

18. The computer-implemented method of claim 15, wherein automatically marking each region of interest includes calculating the average number of computed tomographic images for each region of interest.

19. A computer-implemented method, comprising: Quality assurance testing of the tomographic imaging system begins via the processor; The processor uses the tomographic imaging system to obtain tomographic images of a heterogeneous phantom, wherein the heterogeneous phantom includes a heterogeneous region, the heterogeneous region having multiple materials with varying attenuation coefficients; The heterogeneous region is automatically segmented from the tomographic image to generate a segmented image; Based on the segmented image, multiple regions of interest with varying attenuation coefficients are automatically identified within the tomographic image, wherein the multiple regions of interest correspond to the heterogeneous regions within the heterogeneous phantom; Each of the plurality of regions of interest is automatically labeled to represent a specific material among the plurality of materials; as well as The processor calculates an image quality metric based on the markers for each region of interest.

20. The computer-implemented method of claim 19, wherein automatically segmenting the heterogeneous region from the tomographic image comprises utilizing a trained deep neural network to segment the heterogeneous region from the tomographic image to generate the segmented image.

Images