X-ray computed tomography equipment, learning equipment, and ultrasound diagnostic equipment

Abstract

To easily reduce artifacts.SOLUTION: An X-ray computerized tomographic apparatus according to an embodiment includes a generation part and a display control part. The generation part generates a first reduced image with reduced metal artifacts from a first sinogram collected by 3-dimensional positioning imaging on a subject or from a positioning image based on the first sinogram, using a first machine learning model trained by machine learning to reduce metal artifacts. The display control part displays the first reduced image.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The embodiments disclosed in the present specification and drawings relate to an X-ray computed tomography apparatus, a learning apparatus, and an ultrasound diagnostic apparatus. [Background technology]

[0002] When the subject contains metal, metal artifacts appear in the reconstructed CT image. Because metal artifacts can affect diagnosis, a process called Metal Artifact Reduction (MAR) is performed to reduce metal artifacts, generating images that minimize the impact on diagnosis.

[0003] When the subject contains metal, it is impossible to know what kind of artifacts will occur and how much artifacts can be reduced by performing MAR processing without actually performing the reconstruction and MAR processing. Therefore, it is difficult for users to determine in advance whether to apply MAR processing during reconstruction. Even if MAR processing is performed every time, there are issues with the calculation cost and the risk of image quality degradation due to MAR processing. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-10378 [Patent Document 2] Japanese Patent Application Publication No. 2018-89301 [Patent Document 3] Japanese Patent Publication No. 2020-179031 [Patent Document 4] Special Publication No. 2008-505694 Summary of the Invention [Problem to be solved by the invention]

[0005] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to easily reduce artifacts. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0006] An X-ray computed tomography apparatus according to an embodiment includes a generator and a display controller. The generator generates a first reduced image in which metal artifacts have been reduced from a first sinogram acquired by 3D positioning imaging of a subject or a positioning image based on the first sinogram, using a first machine learning model trained by machine learning to reduce metal artifacts. The display controller displays the first reduced image. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram showing an example of the arrangement of an X-ray computed tomography apparatus according to the first embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of the learning device according to the first embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of input and output of the DL-MAR model according to the first embodiment. [Figure 4] FIG. 4 is a diagram schematically illustrating a first example of a process for generating a DL-MAR model by the learning device according to the first embodiment. [Figure 5] FIG. 5 is a diagram schematically illustrating a second example of the process of generating a DL-MAR model by the learning device according to the first embodiment. [Figure 6] FIG. 6 is a diagram showing an example of the flow of a first CT examination performed by the X-ray computed tomography apparatus according to the first embodiment. [Figure 7] FIG. 7 shows an example of the preview display confirmation screen displayed in step SC4. [Figure 8]FIG. 8 shows an example of the preview screen displayed in step SC8. [Figure 9] FIG. 9 is a diagram showing an example of input and output of the DL-MAR model related to the second CT examination of the first embodiment. [Figure 10] FIG. 10 is a diagram showing an example of the flow of the second CT examination by the X-ray computed tomography apparatus according to the first embodiment. [Figure 11] FIG. 11 is a diagram showing an example of the arrangement of an X-ray computed tomography apparatus according to the second embodiment. [Figure 12] FIG. 12 is a diagram showing an example of input and output of the cross-section restoration model according to the second embodiment. [Figure 13] FIG. 13 is a diagram illustrating an example of input and output of the DL-MAR model according to the second embodiment. [Figure 14] FIG. 14 is a diagram showing an example of the flow of a CT examination using the X-ray computed tomography apparatus according to the second embodiment. [Figure 15] FIG. 15 is a diagram showing an example of the configuration of an ultrasonic diagnostic apparatus according to the third embodiment. [Figure 16] FIG. 16 is a diagram illustrating an example of input and output of the artifact reduction model according to the third embodiment. [Figure 17] FIG. 17 is a diagram schematically illustrating an example of a process for generating an artifact reduction model by the learning device according to the third embodiment. [Figure 18] FIG. 18 is a diagram showing an example of the flow of an ultrasonic examination performed by the ultrasonic diagnostic apparatus according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, embodiments of an X-ray computed tomography apparatus, a learning apparatus, and an ultrasonic diagnostic apparatus will be described in detail with reference to the drawings.

[0009] Hereinafter, embodiments of an X-ray computed tomography apparatus and an X-ray diagnostic apparatus will be described in detail with reference to the drawings.

[0010] (First embodiment) FIG. 1 is a diagram showing an example of the configuration of an X-ray computed tomography apparatus 1 according to the first embodiment. As shown in FIG. 1, the X-ray computed tomography apparatus 1 includes a gantry 10, a bed 30, and a console 40. Although FIG. 1 illustrates multiple gantry 10s for ease of explanation, the apparatus may actually include one or multiple gantry 10s. The gantry 10 is a scanning device configured to perform X-ray CT scans of a subject P. The bed 30 is a transport device on which the subject P to be scanned for X-ray CT scans is placed and which positions the subject P. The console 40 is a computer that controls the gantry 10. For example, the gantry 10 and the bed 30 are installed in a CT examination room, and the console 40 is installed in a control room adjacent to the CT examination room. The gantry 10, the bed 30, and the console 40 are connected to each other by wire or wirelessly so that they can communicate with each other. The console 40 does not necessarily have to be installed in the control room. For example, the console 40 may be installed in the same room as the gantry 10 and the bed 30. The console 40 may also be incorporated into the cradle 10 .

[0011] As shown in FIG. 1, the gantry 10 includes an X-ray tube 11, an X-ray detector 12, a rotating frame 13, an X-ray high voltage device 14, a control device 15, a wedge 16, a collimator 17, and a data acquisition system (DAS) 18.

[0012] The X-ray tube 11 irradiates the subject P with X-rays. Specifically, the X-ray tube 11 includes a cathode that generates thermoelectrons, an anode that generates X-rays upon receiving thermoelectrons flying from the cathode, and a vacuum tube that holds the cathode and anode. The X-ray tube 11 is connected to the X-ray high voltage device 14 via a high-voltage cable. A tube voltage is applied between the cathode and the anode by the X-ray high voltage device 14. The application of the tube voltage causes thermoelectrons to fly from the cathode toward the anode. A tube current flows as the thermoelectrons fly from the cathode toward the anode. X-rays are generated when the thermoelectrons collide with the anode.

[0013] The X-ray detector 12 detects X-rays emitted from the X-ray tube 11 and passing through the subject P, and outputs an electrical signal corresponding to the detected X-ray dose to the DAS 18. The X-ray detector 12 has a structure in which a plurality of X-ray detection element rows, each of which has a plurality of X-ray detection elements arranged in the channel direction, are arranged in the slice direction (row direction). The X-ray detector 12 is, for example, an indirect conversion type detector having a grid, a scintillator array, and a photosensor array. The scintillator array has a plurality of scintillators. The scintillator outputs light with an amount of light corresponding to the amount of incident X-rays. The grid is arranged on the X-ray incident surface side of the scintillator array and has an X-ray shielding plate that absorbs scattered X-rays. The grid is sometimes called a collimator (one-dimensional collimator or two-dimensional collimator). The photosensor array converts light from the scintillator into an electrical signal corresponding to the amount of light. For example, a photodiode is used as the photosensor. The X-ray detector 12 may also be a direct conversion type detector.

[0014] The rotating frame 13 is an annular frame that supports the X-ray tube 11 and the X-ray detector 12 rotatably around a rotation axis (Z-axis). Specifically, the rotating frame 13 supports the X-ray tube 11 and the X-ray detector 12 so that they face each other. In addition to the X-ray tube 11 and the X-ray detector 12, the rotating frame 13 also supports the X-ray high-voltage generator 14 and the DAS 18. The rotating frame 13 is supported on a fixed frame (not shown) so that it can rotate around the rotation axis. The rotation mechanism includes, for example, a motor that generates a rotational driving force and a bearing that transmits the rotational driving force to the rotating frame 13 to rotate it. The motor is provided on the fixed frame, and the bearing is physically connected to the rotating frame 13 and the motor, so that the rotating frame 13 rotates in response to the rotational force of the motor. The rotation of the rotating frame 13 around the rotation axis causes the X-ray tube 11 and the X-ray detector 12 to rotate around the rotation axis. The rotating frame 13 is an example of a rotating unit.

[0015] This embodiment is applicable to supine CT and / or upright CT. In the case of supine CT, the longitudinal direction of the rotation axis of the rotating frame 13 or the top board 33 of the bed 30 in a non-tilted state is defined as the Z-axis direction, the axial direction perpendicular to the Z-axis direction and horizontal to the floor surface is defined as the X-axis direction, and the axial direction perpendicular to the Z-axis direction and perpendicular to the floor surface is defined as the Y-axis direction. In the case of application only to upright CT or to both supine CT and upright CT, the longitudinal direction of the rotation axis of the rotating frame 13 in a non-tilted state is defined as the Z-axis direction, the direction perpendicular to the Z-axis direction and from the center of rotation toward the support column supporting the rotating frame 13 is defined as the X-axis direction, and the direction perpendicular to the Z-axis direction and the X-axis direction is defined as the Y-axis direction.

[0016] The X-ray high voltage device 14 has a high voltage generator and an X-ray control device. The high voltage generator has electrical circuits such as a transformer and a rectifier, and generates a high voltage to be applied to the X-ray tube 11 and a filament current to be supplied to the X-ray tube 11. The X-ray control device controls the output voltage according to the X-rays emitted by the X-ray tube 11. The high voltage generator may be of a transformer type or an inverter type. The X-ray high voltage device 14 may be provided on the rotating frame 13 in the gantry 10, or on a fixed frame (not shown) in the gantry 10.

[0017] The wedge 16 adjusts the dose of X-rays irradiated onto the subject P. Specifically, the wedge 16 attenuates the X-rays so that the dose of X-rays irradiated from the X-ray tube 11 onto the subject P has a predetermined distribution. For example, the wedge 16 is made of a metal plate such as aluminum, such as a wedge filter or a bow-tie filter.

[0018] The collimator 17 limits the irradiation range of the X-rays that have passed through the wedge 16. The collimator 17 slidably supports multiple lead plates that shield the X-rays, and adjusts the shape of the slits formed by the multiple lead plates. The collimator 17 is sometimes called an X-ray aperture.

[0019] The DAS 18 reads out an electrical signal from the X-ray detector 12 corresponding to the X-ray dose detected by the X-ray detector 12. The DAS 18 amplifies the read-out electrical signal and integrates the electrical signal over a view period to collect detection data having a digital value corresponding to the X-ray dose over the view period. The detection data is also called projection data. The DAS 18 is realized, for example, by an application specific integrated circuit (ASIC) equipped with circuit elements capable of generating projection data. The projection data is transmitted to the console 40 via a non-contact data transmission device or the like.

[0020] In this embodiment, an integral type X-ray detector 12 and an X-ray computed tomography apparatus 1 equipped with an integral type X-ray detector 12 are described as examples, but the technology according to this embodiment can also be applied to a photon counting type X-ray detector.

[0021] The rotating frame 13 and the fixed frame are each provided with a non-contact or contact communication circuit, and these communication circuits enable communication between the units supported on the rotating frame 13 and devices external to the fixed frame or gantry 10. For example, if optical communication is used as the non-contact communication method, the detection data generated by the DAS 18 is transmitted by optical communication from a transmitter having a light-emitting diode (LED) provided on the rotating frame 13 to a receiver having a photodiode provided on the fixed frame of the gantry 10, and the data is then transferred from the fixed frame to the console 40 by the transmitter. Note that, other communication methods may also be used, such as non-contact data transmission methods such as capacitive coupling and radio wave methods, as well as contact data transmission methods using slip rings and electrode brushes.

[0022] The control device 15 controls the X-ray high-voltage generator 14 and the DAS 18 to perform X-ray CT imaging in accordance with the imaging control function 441 of the processing circuit 44 of the console 40. The control device 15 includes a processing circuit having a central processing unit (CPU) or a micro processing unit (MPU), etc., and a drive mechanism such as a motor and an actuator. The processing circuit includes, as hardware resources, a processor such as a CPU and memory such as a read-only memory (ROM) or a random access memory (RAM). The control device 15 executes various functions using a processor that executes programs loaded in the memory. Note that various functions are not limited to being implemented by a single processing circuit. A processing circuit may be configured by combining multiple independent processors, and each processor may execute a program to implement each function. The control device 15 may also be implemented using an ASIC or a field programmable gate array (FPGA). Furthermore, the control device 15 may be realized by other complex programmable logic devices (CPLDs) or simple programmable logic devices (SPLDs).

[0023] The control device 15 has a function of receiving input signals from an input interface 43 (described later) attached to the console 40 or the gantry 10 and controlling the operation of the gantry 10 and the bed 30. For example, the control device 15 receives input signals and controls the rotation of the rotating frame 13, the tilt of the gantry 10, and the operation of the bed 30 and the tabletop 33. The control of tilting the gantry 10 is realized by the control device 15 rotating the rotating frame 13 around an axis parallel to the X-axis direction based on inclination angle (tilt angle) information input via an input interface attached to the gantry 10. The control device 15 may be provided in the gantry 10 or in the console 40.

[0024] The bed 30 includes a base 31, a support frame 32, a top plate 33, and a bed driving device 34. The base 31 is placed on the floor. The base 31 is a housing that supports the support frame 32 so that it can move vertically (in the Y-axis direction) relative to the floor. The support frame 32 is a frame provided on top of the base 31. The support frame 32 supports the top plate 33 so that it can slide along the rotation axis (Z-axis). The top plate 33 is a flexible plate on which the subject P is placed.

[0025] The bed driving device 34 is housed in the housing of the bed 30. The bed driving device 34 is a motor or actuator that generates power to move the support frame 32 on which the subject P is placed and the tabletop 33. The bed driving device 34 operates under the control of the console 40 or the like.

[0026] The console 40 has a memory 41, a display 42, an input interface 43, and a processing circuit 44. Data communication between the memory 41, the display 42, the input interface 43, and the processing circuit 44 is performed via a bus (BUS). Note that although the console 40 will be described as being separate from the gantry 10, the gantry 10 may include the console 40 or some of the components of the console 40.

[0027] The memory 41 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), or an integrated circuit storage device that stores various information. The memory 41 may be a portable storage medium such as a compact disc (CD), a digital versatile disc (DVD), a Blu-ray (registered trademark) disc (BD), or a flash memory, in addition to an HDD or SSD. The memory 41 may also be a drive device that reads and writes various information from and to a semiconductor memory element such as a flash memory or a RAM. The storage area of ​​the memory 41 may be located within the X-ray computed tomography apparatus 1, or may be located in an external storage device connected via a network.

[0028] The display 42 displays various types of information. Any of a variety of displays can be used as the display 42, as appropriate. For example, a liquid crystal display (LCD), a cathode ray tube (CRT) display, an organic electroluminescence display (OLED), or a plasma display can be used as the display 42. The display 42 may be installed anywhere in the control room. The display 42 may also be installed on the stand 10. The display 42 may be a desktop type, or may be configured as a tablet terminal or the like capable of wireless communication with the main body of the console 40. The display 42 may also be configured as one or more projectors.

[0029] The input interface 43 accepts various input operations from the operator, converts the accepted input operations into electrical signals, and outputs the electrical signals to the processing circuitry 44. Examples of the input interface 43 that can be used include a mouse, keyboard, trackball, switch, button, joystick, touchpad, and touch panel display, as appropriate. In this embodiment, the input interface 43 is not limited to a device equipped with physical operating components such as a mouse, keyboard, trackball, switch, button, joystick, touchpad, and touch panel display. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs the electrical signal to the processing circuitry 44 is also included as an example of the input interface 43. The input interface 43 may also be provided on the pedestal 10. The input interface 43 may also be configured as a tablet terminal or the like that is capable of wireless communication with the console 40 main body.

[0030] The processing circuitry 44 controls the overall operation of the X-ray computed tomography apparatus 1 in response to electrical signals of input operations output from the input interface 43. The processing circuitry 44 generates image data based on electrical signals output from the X-ray detector 12. For example, the processing circuitry 44 has, as hardware resources, a processor such as a CPU, MPU, or GPU, and memory such as ROM or RAM. The processing circuitry 44 executes an imaging control function 441, a sinogram acquisition function 442, an image generation function 443, a metal position identification function 444, a metal artifact estimation function 445, a DL-MAR function 446, a MAR function 447, a display control function 448, and the like, by using a processor that executes programs loaded in the memory.

[0031] Note that each of the functions 441 to 448 is not limited to being realized by a single processing circuit, but may be realized by combining a plurality of independent processors to form a processing circuit, and each processor may execute a program to realize each of the functions 441 to 448.

[0032] In the imaging control function 441, the processing circuitry 44 controls the X-ray high voltage device 14, the control device 15, and the DAS 18 in accordance with the imaging conditions to perform X-ray CT imaging. In this embodiment, 3D positioning imaging and main imaging are performed as X-ray CT imaging. The 3D positioning imaging is low-dose imaging for collecting 3D images (volume data) of the subject P, which is performed prior to the main imaging in order to determine the region of the subject P to be imaged in the main imaging. The main imaging is high-dose imaging with a higher dose than the 3D positioning imaging. Note that in the first embodiment, both the 3D positioning imaging and the main imaging may be performed, or only one of the 3D positioning imaging or the main imaging may be performed.

[0033] In the sinogram acquisition function 442, the processing circuitry 44 aligns the projection data collected via the DAS 18 during X-ray CT imaging to acquire a sinogram. The sinogram is, for example, two-dimensional spatial data whose horizontal axis is the channel direction of the X-ray detector 12 and whose vertical axis is the rotation angle direction (view direction) of the X-ray tube 11, and each point is assigned a value of the projection data or a value based thereon. The sinogram is generated for each X-ray detection element row of the X-ray detector 12. Note that the sinogram may take a variety of formats, and may be, for example, three-dimensional spatial data whose horizontal axis is the channel direction, whose vertical axis is the view direction, and whose depth axis is the row direction. Alternatively, the sinogram may be two-dimensional spatial data whose horizontal axis is the channel direction and whose vertical axis is the view direction, and whose points are assigned statistical values ​​of the projection data values ​​of the X-ray detection element row.

[0034] In the image generation function 443, the processing circuitry 44 performs reconstruction processing on a sinogram related to the subject P to generate a CT image. Specifically, the processing circuitry 44 performs preprocessing on the sinogram, such as logarithmic conversion processing, offset correction processing, inter-channel sensitivity correction processing, and beam hardening correction. The processing circuitry 44 performs reconstruction processing on the preprocessed sinogram to generate a CT image related to the subject P. As the reconstruction processing, a filtered back projection method or an iterative reconstruction method is used. Alternatively, as the reconstruction processing, a reconstruction processing incorporating a denoising process using machine learning into these methods may be used. The processing circuitry 44 converts the CT image into a cross-sectional image of an arbitrary cross section or a rendering image of an arbitrary viewpoint direction. The conversion is performed based on an input operation received from the operator via the input interface 43. For example, the processing circuitry 44 performs three-dimensional image processing such as volume rendering, surface volume rendering, pixel value projection processing, MPR (Multi-Planer Reconstruction) processing, and CPR (Curved MPR) processing on the reconstructed image data to generate a rendering image in any viewpoint direction.

[0035] In this embodiment, a sinogram acquired by 3D positioning imaging is called a positioning sinogram, and a CT image based on the positioning sinogram is called a positioning image. Also, a sinogram acquired by actual imaging is called an actual imaging sinogram, and a CT image based on the actual imaging sinogram is called an actual imaging image.

[0036] In the metal position identification function 444, the processing circuitry 44 detects a metal region included in the positioning image. The metal region is an image region included in the positioning image that corresponds to the metal contained in the subject P. The processing circuitry 44 identifies the position where the detected metal region exists. The position of the metal region is represented by the position of the bed 30 in the Z-axis direction (slice direction).

[0037] In the metal artifact estimation function 445, the processing circuitry 44 generates a reconstructed image (hereinafter referred to as a pseudo-main image) that is estimated to be generated by the actual imaging based on the positioning sinogram or the positioning image. The pseudo-main image includes metal artifacts that are estimated to be included in the reconstructed image generated by the actual imaging.

[0038] In the DL-MAR function 446, the processing circuitry 44 uses a machine learning model (hereinafter referred to as a DL-MAR model) trained by machine learning to reduce metal artifacts to generate an image (hereinafter referred to as a DL-MAR image) from a positioning sinogram or positioning image acquired by 3D positioning imaging of the subject P, in which metal artifacts contained in the positioning image have been reduced. The DL-MAR model is generated by a learning device, which will be described later. For example, when a metal region is detected by the metal localization function 444, the processing circuitry 44 generates a DL-MAR image from the positioning sinogram.

[0039] In the MAR function 447, the processing circuitry 44 executes metal artifact reduction processing (hereinafter referred to as MAR processing) without using the DL-MAR model. For example, the processing circuitry 44 performs MAR processing on the actual captured sinogram or the actual captured image to generate a CT image in which metal artifacts are reduced (hereinafter referred to as MAR image).

[0040] In the display control function 448, the processing circuitry 44 displays various information on the display 42. For example, the display 42 displays a positioning sinogram, a positioning image, a main image, a DL-MAR image, a MAR image, and the like.

[0041] Although the console 40 has been described as a single console that executes multiple functions, multiple functions may be executed by separate consoles. The processing circuitry 44 is not limited to being included in the console 40, but may also be included in an integrated server that collectively processes projection data acquired by multiple medical image diagnostic devices. Post-processing may be performed by either the console 40 or an external workstation. Furthermore, processing may be performed simultaneously by both the console 40 and the workstation.

[0042] There are various types of X-ray computed tomography apparatuses (CT apparatuses), such as third-generation CT and fourth-generation CT, and any of these types can be applied to this embodiment. Here, the third-generation CT is a rotate / rotate-type in which the X-ray tube and detector rotate together around the subject. The fourth-generation CT is a stationary / rotate-type in which a large number of X-ray detection elements arranged in a ring shape are fixed, and only the X-ray tube rotates around the subject.

[0043] Fig. 2 is a diagram showing an example of the configuration of a learning device 5 according to the first embodiment. The learning device 5 is a computer for generating a DL-MAR model. As shown in Fig. 2, the learning device 5 has a memory 51, a display 52, an input interface 53, a communication interface 54, and a processing circuit 55. Data communication between the memory 51, the display 52, the input interface 53, the communication interface 54, and the processing circuit 55 is performed via a BUS.

[0044] The memory 51 is a storage device such as an HDD, SSD, or integrated circuit storage device that stores various information. In addition to an HDD or SSD, the memory 51 may be a portable storage medium such as a CD, DVD, BD, or flash memory. The memory 51 may also be a drive device that reads and writes various information to and from semiconductor memory elements such as flash memory and RAM. The storage area of ​​the memory 51 may be located within the learning device 5 or in an external storage device connected via a network.

[0045] The display 52 displays various types of information. Any of a variety of displays can be used as the display 52. ​​For example, an LCD, a CRT display, an OLED, or a plasma display can be used as the display 52. ​​Furthermore, one or more projectors may be used as the display 52.

[0046] The input interface 53 accepts various input operations from an operator, converts the accepted input operations into electrical signals, and outputs the electrical signals to the processing circuit 55. As the input interface 53, for example, a mouse, keyboard, trackball, switch, button, joystick, touchpad, touch panel display, etc. can be used as appropriate. Note that in this embodiment, the input interface 53 is not limited to one equipped with physical operation components such as a mouse, keyboard, trackball, switch, button, joystick, touchpad, and touch panel display. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the device and outputs the electrical signal to the processing circuit 55 is also included as an example of the input interface 53.

[0047] The communication interface 54 performs data communication between the X-ray computed tomography apparatus 1 and various hospital information systems such as a Hospital Information System (HIS), a Radiology Information System (RIS), a Picture Archiving and Communication System (PACS), etc. For example, a LAN card, a network adapter, a network interface card, etc. are used as the communication interface 54.

[0048] The processing circuitry 55 controls the overall operation of the learning device 5 in response to electrical signals of input operations output from the input interface 53. For example, the processing circuitry 55 has, as hardware resources, a processor such as a CPU, MPU, or GPU, and memory such as a ROM or RAM. The processing circuitry 55 executes an acquisition function 551, a reconstruction function 552, an embedding function 553, an inverse reconstruction function 554, an MAR function 555, a machine learning function 556, a display control function 557, and the like, using a processor that executes a program deployed in the memory.

[0049] Note that each of the functions 551-557 is not limited to being realized by a single processing circuit, but may be realized by combining multiple independent processors to form a processing circuit, and each processor may execute a program to realize each of the functions 551-557.

[0050] The processing circuitry 55 acquires various pieces of information in the acquisition function 551. For example, the processing circuitry 55 acquires sinograms or reconstructed images based on the sinograms from the X-ray computed tomography apparatus 1, HIS, RIS, PACS, etc. The processing circuitry 55 acquires sinograms or reconstructed images of various subjects.

[0051] In the reconstruction function 552, when the acquisition function 551 acquires a sinogram, the processing circuitry 55 performs reconstruction processing on the sinogram to generate a reconstructed image.

[0052] In the embedding function 553, the processing circuitry 55 embeds a pseudo metal region (hereinafter referred to as a pseudo metal region) in the reconstructed image. Hereinafter, the reconstructed image in which the pseudo metal region is embedded will be referred to as a pseudo metal image.

[0053] In the inverse reconstruction function 554, the processing circuitry 55 performs inverse reconstruction processing on the reconstructed image to generate a sinogram. The inverse reconstruction processing is a term used in contrast to the reconstruction processing that generates a reconstructed image from a sinogram, and refers to the processing of generating a sinogram from a reconstructed image. As an example, the processing circuitry 55 performs inverse reconstruction processing on a pseudo-metal image to generate a sinogram including a metal region. Hereinafter, a sinogram generated by performing inverse reconstruction processing on a reconstructed image or a pseudo-metal image will be referred to as a calculated sinogram.

[0054] In MAR function 555, processing circuitry 55 applies MAR processing to the sinogram to produce a sinogram with reduced metal artifacts.

[0055] In the machine learning function 556, the processing circuitry 55 uses the calculated sinogram to train a machine learning model (DL-MAR model) for reducing metal artifacts. More specifically, the processing circuitry 55 trains the DL-MAR model based on supervised learning using the calculated sinogram and a sinogram generated by applying a metal artifact reduction process to the calculated sinogram or a sinogram used to reconstruct a reconstructed image.

[0056] In the display control function 557, the processing circuit 55 displays various information on the display 52. ​​For example, a reconstructed image, a pseudo-metal image, a sinogram, a calculated sinogram, etc. are displayed on the display 42.

[0057] Next, the process of generating a DL-MAR model by the learning device 5 will be described.

[0058] FIG. 3 is a diagram showing an example of input and output of the DL-MAR model according to the first embodiment. As shown in FIG. 3, the DL-MAR model is a machine learning model trained to input a positioning sinogram and output a sinogram after DL-MAR processing (hereinafter referred to as a DL-MAR sinogram). A deep neural network is used as the machine learning model. The positioning sinogram and the DL-MAR sinogram are assumed to be three-dimensional spatial data. The positioning sinogram is a sinogram including metal components. The DL-MAR sinogram is a sinogram in which metal components have been reduced.

[0059] FIG. 4 is a diagram schematically illustrating a first example of a process for generating a DL-MAR model by the learning device 5 according to the first embodiment. First, the processing circuitry 55 acquires a sinogram GS1 collected in a previous scan from a medical image database, such as an X-ray computed tomography apparatus, HIS, RIS, or PACS. The previous scan may be a positioning scan or a real scan, as long as it is a 3D scan. The scanned area is not particularly limited. The subject to be scanned for the sinogram GS1 is assumed to contain no metal. Therefore, the sinogram GS1 is assumed to contain no projection components derived from metal (hereinafter referred to as metal components). Once the sinogram GS1 is acquired, the processing circuitry 55 performs a reconstruction process on the sinogram GS1 to generate a reconstructed image IR1 (step SA1). The reconstructed image IR1 does not contain any metal regions or metal artifacts.

[0060] The processing circuitry 55 may obtain the reconstructed image IR1 from a medical image database, in which case step SA1 is omitted.

[0061] Next, the processing circuit 55 embeds a pseudo-metal region M1 in the reconstructed image IR1 (step SA2). In step SA2, the processing circuit 55 embeds the pseudo-metal region at an arbitrary location in the reconstructed image IR1. The embedding is performed by replacing pixel values ​​at the embedding target position in the reconstructed image IR1 with arbitrary pixel values ​​corresponding to pseudo-metal. The embedding position may be determined randomly. Alternatively, a region where metal is likely to be embedded in surgery or the like may be extracted using an anatomical landmark detection (ALD) algorithm or the like. Examples of metals embedded in surgery include bolts, stents, and implants. For example, a blood vessel region may be extracted from the reconstructed image IR1 using ALD or the like, and a pseudo-metal region equivalent to a stent may be embedded around the blood vessel region.

[0062] The shape of the pseudo-metal region to be embedded may be the shape of a common metal (bolt, stent, implant, etc.) implanted in the human body, or it may be a random shape, or it may be a slightly modified version of any of these. The shapes of commercially available metals may be recorded in advance as a database as the shape of the pseudo-metal region, or they may be made importable in a format similar to object data for 3D printers. The number of pseudo-metal regions to be embedded may be one or multiple.

[0063] After step SA2, the processing circuit 55 performs inverse reconstruction processing on the reconstructed image IR2 in which the pseudo-metal region M1 is embedded to generate a calculated sinogram GS2 (step SA3). The inverse reconstruction processing simulates (predicts) CT imaging of a subject in which a metal corresponding to the pseudo-metal region is embedded. In the simulation, a forward projection process is performed on the reconstructed image IR2. For example, the simulation takes into account the CT value of the reconstructed image IR1, the material and density of the assumed embedded metal, the amount of irradiated photons, and the attenuation and scattered rays of X-rays when they pass through a material. This estimates the calculated sinogram GS2 that would be obtained by the CT imaging. The calculated sinogram GS2 includes a metal component M2 based on the pseudo-metal region.

[0064] If there is a difference between a sinogram obtained by actually imaging a subject with metal and a sinogram obtained by inverse reconstruction with a pseudo-metal, i.e., if the estimation accuracy of the calculated sinogram GS2 is poor, the accuracy of the calculated sinogram GS2 may be improved using a machine learning model. The machine learning model is a neural network trained to input a sinogram generated by inverse reconstruction and output a sinogram acquired by CT imaging. In this case, the processing circuitry 55 applies the machine learning model to the calculated sinogram GS2 and outputs a sinogram. This allows the calculated sinogram GS2 to have image quality closer to that of a sinogram acquired by CT imaging.

[0065] When step SA3 is performed, the processing circuitry 55 performs MAR processing on the calculated sinogram GS2 to generate a sinogram GS3 after MAR processing (step SA3). The MAR processing may be any image processing that can reduce metal artifacts and does not use a DL-MAR model. Such MAR processing may be threshold processing, an image filter, or a metal artifact reduction process incorporated into a reconstruction process such as iterative reconstruction. Hereinafter, MAR processing will be referred to as metal artifact reduction processing incorporated into a reconstruction process. Examples of such MAR processing include processing a sinogram to generate a sinogram, processing a sinogram to generate a reconstructed image, processing a reconstructed image to generate a reconstructed image, and processing a reconstructed image to generate a sinogram. Hereinafter, MAR processing will be referred to as processing a sinogram to generate a sinogram.

[0066] Here, a brief description of the MAR processing by the processing circuit 55 will be given. First, (1) a metal region is extracted from an original reconstructed image containing metal regions and metal artifacts. Next, (2) the metal region is used to extract metal components from the original sinogram. Next, (3) the metal components are removed from the original sinogram by interpolation using the metal components. Next, (4) the sinogram from which the metal components have been removed is back-projected to generate a reconstructed image, and image processing is performed on the reconstructed image to further reduce the metal artifacts. Next, (5) the reconstructed image is forward-projected to generate a sinogram, and image processing is performed on the sinogram to further remove the metal components. This generates a sinogram GS3 from which the metal components have been removed. Note that the sinogram from which the metal components have been extracted from the original sinogram in step (2) may be used as the sinogram GS3 from which the metal components have been removed.

[0067] The processing circuitry 55 executes steps SA1 to SA4 described above on sinograms or reconstructed images of various subjects, collecting sinograms GS2 containing the metal component M2 and sinograms GS3 not containing the metal component M2 for various subjects. The sinograms GS2 containing the metal component M2 and the sinograms GS3 not containing the metal component M2 are used as training samples for machine learning. The sinograms GS2 are used as input training samples, and the sinograms GS3 are used as output training samples. The embedding process of pseudo-metal regions in step SA2 can be executed countless times for a single reconstructed image IR1, making it easy to collect a large number of training samples.

[0068] The processing circuit 55 generates a DL-MAR model through supervised learning using a sinogram GS2 containing the metal component M2 as an input training sample and a sinogram GS3 not containing the metal component M2 as an output training sample. More specifically, the processing circuit 55 first applies a machine learning model to the input training sample to perform forward propagation processing to generate an estimated sinogram. Next, the processing circuit 55 applies a machine learning model that evaluates the difference (error) between the output training sample and the estimated sinogram to perform backpropagation processing to calculate a gradient vector, which is the differential coefficient of an error function, which is a function of learning parameters such as weights and biases. Next, the processing circuit 55 updates the learning parameters based on the gradient vector to minimize the error function. These forward propagation processing, backpropagation processing, and parameter update processing are repeated while changing the learning sample until a stopping condition is satisfied. The stopping condition may be set when the number of updates reaches a predetermined number, the value of the error function falls below a threshold, or the like. The machine learning model to which the learning parameters are assigned when the stopping condition is satisfied is the DL-MAR model. The DL-MAR model is supplied to the X-ray computed tomography apparatus 1.

[0069] This concludes the description of the first example of the generation process for the DL-MAR model. Note that the first generation process can be modified in various ways.

[0070] For example, in the above embodiment, the imaging sites of the learning samples used for machine learning are not particularly limited. In this case, the imaging sites to which the DL-MAR model can be applied are not limited. However, the DL-MAR model may be trained using learning samples of only a specific imaging site. This makes it possible to train a DL-MAR model specialized for the specific imaging site. In this case, it is possible to reduce metal artifacts in reconstructed images related to the specific imaging site with high accuracy.

[0071] As another example, a sinogram GS2 to which arbitrary noise has been added in addition to the metal component M2 may be used as the input training sample. The noise may be added to a sinogram GS2 generated by performing an inverse reconstruction process on a reconstructed image in which a pseudo-metal region has been embedded. Alternatively, a pseudo-metal artifact may be added to the reconstructed image in addition to the pseudo-metal region. In this case, a sinogram generated by performing an inverse reconstruction process on the reconstructed image is used as the input training sample. Adding arbitrary noise in this way is expected to improve the estimation accuracy using machine learning.

[0072] 5 is a diagram schematically illustrating a second example of a generation process of a DL-MAR model by the learning device 5 according to the first embodiment. In the following description, components having substantially the same functions as those in the first generation process are given the same reference numerals and will be described only when necessary.

[0073] As shown in FIG. 5, in the second generation process, similar to the first generation process, a calculated sinogram GS2 generated by performing an inverse reconstruction process on a reconstructed image IR2 in which a pseudo-metal region M1 is embedded is used as an input training sample. Unlike the first generation process, the second generation process uses an original sinogram GS1 as an output training sample. The original sinogram GS1 is collected by CT scanning of a metal-free subject and does not contain any metal components. The processing circuit 55 generates a DL-MAR model based on supervised learning using the sinogram GS2 containing the metal component M2 as an input training sample and the sinogram GS1 not containing the metal component as an output training sample. Note that a sinogram generated by performing an inverse reconstruction process on the reconstructed image IR1 may also be used as the input training sample.

[0074] Next, an example of the operation of the X-ray computed tomography apparatus 1 according to the first embodiment will be described.

[0075] FIG. 6 is a diagram showing an example of the flow of a first CT examination by the X-ray computed tomography apparatus 1 according to the first embodiment. As shown in FIG. 6, the processing circuitry 44, by implementing the imaging control function 441, executes 3D positioning imaging to acquire a positioning sinogram (step SC1). The 3D positioning imaging in step SC1 is assumed to be a low-dose helical scan. The low-dose helical scan involves moving the tabletop 33 on which the subject P is placed along the Z axis while rotating the rotating frame 13, and repeating low-dose X-ray irradiation by the X-ray tube 11 and X-ray detection by the X-ray detector 12, thereby performing X-ray CT imaging of the entire body of the subject P. The X-ray exposure range is preferably limited to one or several rows of X-ray detection elements to reduce exposure of the subject P. The DAS 18 acquires projection data via the X-ray detector 12, and the processing circuitry 44, by implementing the sinogram acquisition function 442, aligns the projection data and acquires a positioning sinogram.

[0076] After step SC1 is performed, the processing circuitry 44 generates a positioning image based on the positioning sinogram by implementing the image generation function 443 (step SC2). The positioning image is a projection image from the X-ray tube 11 onto the X-ray detector 12, and depicts the whole body of the subject P.

[0077] After step SC2 is performed, processing circuitry 44 detects metal regions included in the positioning image generated in step SC2 by implementing metal position identification function 444 (step SC3). In step SC3, processing circuitry 44 performs threshold processing on the positioning image, for example, using a CT value for detecting metal regions as a threshold. A group of pixels having a CT value greater than the threshold is detected as a metal region.

[0078] After step SC3, the processing circuit 44 determines whether or not to display a preview of the MAR processing result (step SC4). In step SC4, the processing circuit 44, by implementing the display control function 448, displays a GUI screen (hereinafter referred to as a preview display confirmation screen) for instructing whether or not to display a preview. The preview display confirmation screen is displayed on the display 42.

[0079] FIG. 7 is a diagram showing an example of a preview display confirmation screen I1 displayed in step SC4. As shown in FIG. 7, the preview display confirmation screen I1 displays the positioning image I11 generated in step SC2. The metal region M3 detected in step SC3 is drawn in the positioning image I11. As shown in FIG. 7, the preview display confirmation screen I1 displays a selection window I12. The selection window I12 displays a pull-down menu I13 for selecting whether or not to perform MAR processing and a GUI button I14 for instructing preview display. For example, if the user determines that the positioning image I11 includes the metal region M3, the user presses the GUI button I14. If the GUI button I14 is pressed, the processing circuit 44 determines to execute preview display. If the GUI button I14 is not pressed and execution of MAR processing is selected, the processing circuit 44 skips steps SC5 to SC8, determines to execute MAR processing in step SC9 (step SC9: YES), and proceeds to step SC10. If the GUI button I14 is not pressed and the selection is made not to execute the MAR processing, the processing circuitry 44 ends the CT examination.

[0080] If it is determined in step SC4 that a preview of the MAR processing result is to be displayed (step SC4: YES), the processing circuitry 44 identifies the position of the metal region detected in step SC3 (hereinafter referred to as the metal position) by implementing the metal position identifying function 444 (step SC5). The metal position is represented by the position of the top board 33 in the Z-axis direction, i.e., the slice direction. The range between the upper and lower ends of the metal region in the Z-axis direction included in the positioner image is identified as the metal position.

[0081] After step SC5 is performed, the processing circuitry 44 generates a pseudo-main image estimated to be generated by the main imaging by implementing the metal artifact estimation function 445 (step SC6). The pseudo-main image is typically generated as a CT image of the same cross section as the reconstructed image, i.e., the scanned cross section. The pseudo-main image includes a metal artifact estimated to be generated in the reconstructed image obtained by the main imaging.

[0082] There are two methods for generating pseudo-main images: a simple method that uses the positioning image and a method that uses machine learning. In the simple method, the processing circuit 44 uses the metal artifacts contained in the positioning image as metal artifacts that are estimated to occur during the main imaging. That is, the processing circuit 44 extracts image data at the metal position from the positioning image and sets the extracted image data as the pseudo-main image. If the extracted image data is three-dimensional image data, the pseudo-main image can be generated by projecting the image data in the column direction.

[0083] In the generation method using machine learning, a machine learning model is used to estimate metal artifacts that are estimated to occur due to the actual radiography. The machine learning model uses a neural network that is trained to input low-dose images and output high-dose images. This machine learning model is called a high-dose image estimation model. The training samples used to train the high-dose image estimation model may be, for example, low-dose images and high-dose images obtained by dual-energy radiography or the like, or low-dose positioning radiography images and high-dose images obtained by the actual radiography. The low-dose images are used as input training samples, and the high-dose images are used as input training samples. The high-dose image estimation model is generated by training the machine learning model based on supervised learning, in which the low-dose images are used as input training samples and the high-dose images are used as output training samples.

[0084] The positioning image generated in step SC1 is a low-dose image because it is generated by 3D positioning imaging, which is a low-dose CT scan. Since the actual scan is a high-dose CT scan, the reconstructed image generated by the actual scan is a high-dose image. Therefore, the processing circuitry 44 applies the positioning image generated in step SC1 to the high-dose image estimation model to generate a pseudo-actual image. The pseudo-actual image includes metal artifacts estimated to be caused by the actual scan.

[0085] To improve the accuracy of the metal artifact estimation process in step SC6, it is recommended that some of the imaging conditions for the positioning imaging in step SC1 be matched with the imaging conditions for the actual imaging. The imaging conditions to be matched include, for example, imaging conditions related to the trajectory of the X-ray tube 11, such as the helical pitch and the tube position, which affect the shape of the metal artifact. Similarly, when a variable helical pitch is used for the actual imaging, it is recommended that the imaging conditions related to the trajectory be matched between the positioning imaging and the actual imaging. Note that the variable helical pitch is an imaging technique that changes the moving speed of the top 33 depending on the imaging region, such as moving slowly near the heart and moving quickly in other regions.

[0086] After step SC6, the processing circuitry 44, by implementing the DL-MAR function 446, performs DL-MAR processing on the positioning sinogram acquired in step SC1 to generate a DL-MAR image (step SC7). In step SC7, the processing circuitry 44 applies a DL-MAR model to the positioning sinogram acquired in step SC1 to generate a DL-MAR-processed sinogram in which metal components have been reduced. Next, the processing circuitry 44 performs reconstruction processing on the DL-MAR-processed sinogram to generate a reconstructed image in which metal regions and metal artifacts have been reduced (hereinafter referred to as a DL-MAR image). At this time, the processing circuitry 44 may extract a sinogram portion necessary for reconstructing the imaging section at the metal position identified in step SC5 from the DL-MAR-processed sinogram and perform reconstruction processing on the extracted sinogram portion to generate a DL-MAR image of the imaging section. This makes it possible to generate a DL-MAR image of the imaging section containing metal. The processing circuitry 44 may also embed the metal region in the DL-MAR image. The metal region may be generated by extracting a metal component sinogram by subtracting the positioning sinogram input to the DL-MAR model from the DL-MAR sinogram output from the DL-MAR model, and then back-projecting the extracted metal component sinogram.

[0087] When step SC7 is performed, the processing circuitry 44 displays the pseudo-actual image generated in step SC6 and the DL-MAR image generated in step SC7 (step SC8) by implementing the display control function 448. In step SC8, the processing circuitry 44 displays a display screen (hereinafter referred to as a preview screen) including the pseudo-actual image and the DL-MAR image on the display 42.

[0088] FIG. 8 is a diagram showing an example of the preview screen I2 displayed in step SC8. As shown in FIG. 8, the preview screen I2 displays the positioning image I11 generated in step SC2. The metal region M3 detected in step SC3 is depicted in the positioning image I11. As shown in FIG. 8, the preview screen I2 displays a preview window I21. The preview window I21 displays a pseudo-actually captured image I22 generated in step SC6 and a DL-MAR image I23 generated in step SC7 side by side. The pseudo-actually captured image I22 depicts metal artifacts that may occur due to the actual photography. The DL-MAR image I23 depicts reduced metal artifacts.

[0089] According to this embodiment, metal artifact reduction processing is performed using a DL-MAR model, which has a lower calculation cost than MAR processing, making it possible to quickly confirm the effect of the metal artifact reduction processing. If the accuracy of the DL-MAR model is high, metal artifacts may be removed and not even rendered.

[0090] As described above, the positioning image I11 is displayed together with the pseudo-actual image I22 and the DL-MAR image I23. An arrow pointing to the metal region M3 on the positioning image I11 is drawn in the preview window I21. This arrow makes it easy to grasp the positions of the cross sections of the pseudo-actual image I22 and the DL-MAR image I23 in the positioning image I11.

[0091] Displaying the preview window I21 allows the user to compare and observe the pseudo-actual image I22 and the DL-MAR image I23. This comparative observation allows the user to determine whether MAR processing is necessary for the actual imaging. For example, if the accuracy of the DL-MAR image I23 is insufficient and the actual imaging image needs to be observed using MAR processing with higher accuracy, the user determines that MAR processing is necessary for the actual imaging. Also, if the metal area contained in the pseudo-actual image I22 is small or there are few metal artifacts, the user determines that MAR processing is unnecessary for the actual imaging.

[0092] If step SC8 has been performed or if it is determined in step SC4 that the preview display of the MAR processing result is not to be performed (step SC4: NO), the processing circuit 44 determines whether or not to perform MAR processing (step SC9).

[0093] As shown in Fig. 8, the preview window I21 displays a selection button I24. The selection button I24 includes, for example, an "ON" radio button indicating that MAR processing is required for the actual scan, and an "OFF" radio button indicating that MAR processing is not required. The preview window I21 also displays an OK button I25 and a cancel button I26. When the cancel button I26 is pressed, the processing circuitry 44 ends the CT scan.

[0094] If the OK button 125 is pressed with the "ON" radio button selected, the processing circuitry 44 determines to execute MAR processing (step SC9: YES). In this case, the processing circuitry 44, by implementing the imaging control function 441, executes the actual imaging and collects the actual imaging sinogram (step SC10). The actual imaging in step SC10 is a high-dose CT imaging that targets the metal position identified in step SC5. The actual imaging may be a 3D scan (volume scan) that does not involve movement of the top 33, or a 3D scan (helical scan) that involves movement of the top 33. The processing circuitry 44, by implementing the sinogram acquisition function 442, aligns the projection data collected by the DAS 18 in the actual imaging and acquires a sinogram (hereinafter referred to as the actual imaging sinogram).

[0095] When step SC10 is performed, the processing circuitry 44 performs MAR processing by implementing the MAR function 447 to generate a MAR image from the actual imaging sinogram acquired in step SC10 (step SC11). In step S11, the processing circuitry 44 performs MAR processing by incorporating metal artifact reduction processing in the process of iterative reconstruction.

[0096] Here, we will briefly explain the MAR process. First, (1) extract the metal regions from the original reconstructed image, which contains metal regions and metal artifacts. Next, (2) use the metal regions to extract the metal components from the original sinogram. Next, (3) use the metal components to perform interpolation to remove the metal components from the original sinogram. Next, (4) back-project the sinogram from which the metal components have been removed to generate a reconstructed image, and then apply image processing to the reconstructed image to further reduce the metal artifacts. Next, (5) forward-project the reconstructed image to generate a sinogram, and then apply image processing to the sinogram to further remove the metal components. Next, (6) back-project the sinogram to generate a reconstructed image in which the metal regions and metal artifacts have been reduced. Finally, (7) add the metal regions to the reconstructed image. This generates a MAR image, which is a reconstructed image in which the metal artifacts have been reduced and the metal regions have been added.

[0097] When step SC11 is performed, the processing circuitry 44 displays the MAR image generated in step SC11 (step SC12) by implementing the display control function 448. The processing circuitry 44 displays the MAR image on the display 42. Since the metal artifacts are reduced with high precision in the MAR image, the user can make a more accurate diagnosis.

[0098] If the OK button I25 is pressed with the "OFF" radio button selected in step SC9, the processing circuitry 44 determines not to perform MAR processing (step SC9: NO). In this case, the processing circuitry 44 performs the actual imaging, performs reconstruction processing without MAR processing, generates a normal reconstructed image, and displays it on the display 42. In this way, when MAR processing is not performed, a simple actual imaging is performed.

[0099] This completes the CT examination using the X-ray computed tomography apparatus 1 according to the first embodiment.

[0100] The above embodiment can be modified in various ways without departing from the spirit of the invention.

[0101] In the above embodiment, steps SC5 to SC8 are performed when it is determined in step SC4 that a preview display is to be performed. However, steps SC5 to SC8 may be performed without making a determination. That is, the processing circuitry 55 may generate a DL-MAR image from the positioning sinogram regardless of whether a metal region is included in the positioning sinogram or the positioning image. By omitting this determination process, it is possible to simplify the workflow of the CT examination. In this case, when a metal region is detected in step SC3, steps SC5 to SC7 are automatically performed, and a preview display is automatically performed in step SC8.

[0102] In the above embodiment, the DL-MAR model may be selected depending on the shape of the metal contained in the subject P. Such a DL-MAR model can be generated by embedding a pseudo-metal region having a specific shape during the embedding process (step SA2). This allows for the generation of DL-MAR models specialized for specific shapes. For example, DL-MAR models for bolts, stents, dental metals, implants, etc. can be generated. When using a DL-MAR model, the processing circuitry 44 identifies the shape of the detected metal region, selects a DL-MAR model corresponding to the identified shape, and applies the localization sinogram to the selected DL-MAR model. This embodiment allows for the use of a DL-MAR model appropriate for the shape of the embedded metal, which is expected to improve the accuracy of metal artifact reduction.

[0103] In the above embodiment, a pseudo-main image is generated in step SC6, but it is not necessary to generate a pseudo-main image. By omitting the pseudo-main image, it is possible to simplify the workflow of the CT examination. In this case, only the DL-MAR image is displayed in step SC8.

[0104] There may be cases where the accuracy of the DL-MAR model is high and you want to use the DL-MAR model in the actual imaging as well. To prepare for this, in step SC9, a button for selecting to perform DL-MAR processing in the actual imaging may be provided in the preview window or the like. When the button is pressed, the DL-MAR model is applied to the actual imaging sinogram, and an actual imaging image with reduced metal artifacts can be generated. This makes it possible to easily perform metal artifact reduction processing on the actual imaging image with low computational cost.

[0105] According to the first CT examination described above, when the subject contains metal, DL-MAR processing using a low-computational-cost machine learning model can be performed on the positioning sinogram acquired in the positioning scan before the actual scan to generate and present a DL-MAR image with reduced metal artifacts. This allows the user to easily know the results of MAR processing before the actual scan without performing the computationally expensive MAR processing. This makes it possible to determine with high accuracy whether or not to perform the computationally expensive MAR processing. Furthermore, by presenting a pseudo-actual image containing metal artifacts estimated to occur during the actual scan together with the DL-MAR image, the user can determine with even higher accuracy whether or not to perform MAR processing by comparing the pseudo-actual image with the DL-MAR image. This allows only the MAR processing that is truly necessary to be performed, thereby improving the workflow of CT examinations.

[0106] Furthermore, if the accuracy of DL-MAR processing improves, the DL-MAR processing itself can be used as MAR processing, which will enable MAR processing to be achieved at low computational cost, further improving the workflow of CT examinations.

[0107] In the first CT examination described above, the DL-MAR model takes a sinogram as input and outputs a sinogram. However, the DL-MAR model according to this embodiment is not limited to this. The second CT examination will be described below. In the second CT examination, the DL-MAR model is applied to images that require real-time processing.

[0108] Imaging methods for capturing images that require real-time performance (real-time CT imaging) include volume CT fluoroscopy, monitoring scan, and TDC (Time Density Curve) scan. Volume CT fluoroscopy is an imaging method in which a three-dimensional imaging region is repeatedly CT-imaged with low-dose X-rays without moving the tabletop 33. Monitoring scan and TDC scan are almost synonymous. Monitoring scan and TDC scan are imaging methods in which CT imaging is repeatedly performed with low-dose X-rays before contrast agent injection, a CT value time change curve (TDC) of the region of interest is generated in real time to monitor the inflow of contrast agent, and when the CT value exceeds a threshold, the imaging method switches to main imaging with high-dose X-rays. The imaging method in the second CT examination can be any of the above methods, but as an example, volume CT fluoroscopy will be used.

[0109] FIG. 9 is a diagram showing an example of input and output of the DL-MAR model for the second CT examination of the first embodiment. As shown in FIG. 9, the DL-MAR model for the second CT examination is a machine learning model trained to input a reconstructed image and output a reconstructed image (DL-MAR image) after DL-MAR processing. A deep neural network is used as the machine learning model. The input reconstructed image includes metal regions and metal artifacts, and the DL-MAR image includes the metal regions and reduces the metal artifacts. Like the DL-MAR model for the first CT examination, the DL-MAR model for the second CT examination is also generated by the learning device 5. The learning device 5 may generate the DL-MAR model for the second CT examination by training the machine learning model based on supervised learning, in which a reconstructed image including metal regions and metal artifacts is used as an input training sample and a reconstructed image including metal regions but not metal artifacts is used as an output training sample.

[0110] 10 is a diagram showing an example of the flow of the second CT examination by the X-ray computed tomography apparatus 1 according to the first embodiment. As shown in Fig. 10, the processing circuitry 44 starts volume CT fluoroscopy by implementing the imaging control function 441 (step SD1). In the volume CT fluoroscopy, the DAS 18 collects projection data on the subject P for each view.

[0111] After step SD1 is performed, the processing circuitry 44 prepares a sinogram of one volume by implementing the sinogram acquisition function 442 (step SD2). Specifically, the processing circuitry 44 prepares a sinogram of one volume by arranging projection data for 360 degrees going back from the most recent view.

[0112] After step SD2, the processing circuitry 44 generates a real-time reconstructed image based on the sinograms prepared in step SD2 by implementing the image generation function 443 (step SD3). Any method, such as filtered back projection or iterative reconstruction, can be used as the reconstruction method for real-time reconstruction, but a method with high speed is preferable. The real-time reconstructed image includes metal regions and metal artifacts.

[0113] After step SD3, the processing circuitry 44 performs DL-MAR processing on the real-time reconstructed image generated in step SD3 by implementing the DL-MAR function 446 to generate a DL-MAR image (step SD4). In step SD4, the processing circuitry 44 applies the real-time reconstructed image to the DL-MAR model related to the second CT examination to generate a DL-MAR image with reduced metal artifacts.

[0114] After step SD4 is performed, the processing circuitry 44 displays the DL-MAR image generated in step SD4 (step SD5) by implementing the display control function 448. In step SD5, the processing circuitry 44 instantly displays the real-time reconstructed image generated in step SD4 on the display 42. This allows for the display of a moving image of the DL-MAR image.

[0115] When step SD5 is performed, the processing circuitry 44 determines whether or not to end the volume CT fluoroscopy (step SD6). If the user determines to end the volume CT fluoroscopy, the user presses the imaging end button provided on the gantry 10 or the imaging end button displayed on the display 42. If the imaging end button is pressed, the processing circuitry 44 determines to end the volume CT fluoroscopy. If the imaging end button is not pressed, the processing circuitry 44 determines not to end the volume CT fluoroscopy (step SD6: NO). The processing circuitry 44 repeats steps SD2 to SD6 until the imaging end button is pressed.

[0116] Then, if it is determined in step SD6 that the volume CT fluoroscopy should be terminated (step SD6: YES), the processing circuitry 44 terminates the volume CT fluoroscopy (step SD7).

[0117] This completes the description of the second CT examination. Note that the second CT examination described above is merely an example, and various modifications are possible as long as they do not change the gist of the invention.

[0118] For example, in the above embodiment, the DL-MAR model inputs a reconstructed image and outputs a reconstructed image, but it may also input a sinogram and output a sinogram. In this case, processing circuitry 44 applies the sinogram obtained in step SD2 to the DL-MAR model to generate a DL-MAR sinogram with reduced metal components, and then performs real-time reconstruction on the sinogram to generate a DL-MAR image.

[0119] According to the second CT examination, a DL-MAR image with reduced metal artifacts can be generated by applying the real-time reconstructed image to the DL-MAR model. If the accuracy of metal artifact reduction by the DL-MAR model is relatively high, the DL-MAR model can be used as a simple MAR processing tool, rather than as a pseudo-MAR processing tool as in the first CT examination. Therefore, even in CT examinations that require immediacy, such as volume CT fluoroscopy, it becomes possible to generate and display DL-MAR images in real time.

[0120] As described above, the X-ray computed tomography apparatus 1 according to the first embodiment includes a processing circuit 44. The processing circuit 44 uses a DL-MAR model trained by machine learning to reduce metal artifacts to generate a DL-MAR image in which metal artifacts contained in the CT image are reduced from a sinogram acquired by CT scanning of a subject or a CT image based on the sinogram. The processing circuit 44 displays the DL-MAR image.

[0121] In the first CT examination, "CT imaging" refers to 3D positioning imaging, "sinogram" refers to the positioning sinogram, and "CT image" refers to the positioning image. In the second CT examination, "CT imaging" refers to real-time imaging such as volume CT fluoroscopy, "sinogram" refers to the sinogram collected by real-time imaging, and "CT image" refers to the real-time reconstructed image.

[0122] According to the above configuration, it is possible to generate a DL-MAR image in which metal artifacts are reduced by using a DL-MAR model, which has a lower calculation cost than MAR processing that does not use a machine learning model. Therefore, according to the first embodiment, it is possible to simply reduce metal artifacts.

[0123] In the above embodiment, the DL-MAR model may be of an input / output type, such as (1) inputting a positioning sinogram and outputting a DL-MAR sinogram, as in the first CT examination, or (2) inputting a positioning image and outputting a DL-MAR image, (3) inputting a positioning sinogram and outputting a DL-MAR image, or (4) inputting a positioning image and outputting a DL-MAR sinogram. In the case of (1), the processing circuitry 44 applies the DL-MAR model to the positioning sinogram to generate a DL-MAR sinogram, and then performs reconstruction processing on the DL-MAR sinogram to generate a DL-MAR image. In the case of (2), the processing circuitry 44 applies the DL-MAR model to the positioning image to generate a DL-MAR image. In the case of (3), the processing circuitry 44 applies the DL-MAR model to the positioning sinogram to generate a DL-MAR image. In the case of (4), the processing circuitry 44 applies the DL-MAR model to the positioning image to generate a DL-MAR sinogram, and then performs reconstruction processing on the DL-MAR sinogram to generate a DL-MAR image. Regardless of the input / output format, the DL-MAR model can be generated based on supervised learning by the learning device 5.

[0124] (Second embodiment) The X-ray computed tomography apparatus according to the second embodiment performs two-dimensional positioning imaging instead of three-dimensional positioning imaging. The X-ray computed tomography apparatus according to the second embodiment will be described below.

[0125] Fig. 11 is a diagram showing an example of the configuration of an X-ray computed tomography apparatus 2 according to the second embodiment. As shown in Fig. 11, a processing circuitry 44 of the X-ray computed tomography apparatus 2 realizes an imaging control function 441, a sinogram acquisition function 442, an image generation function 443, a metal position identification function 444, a DL-MAR function 446, a MAR function 447, a display control function 448, and a cross-section restoration function 449.

[0126] In the metal location identification function 444, the processing circuitry 44 detects metal regions included in positioning images acquired by two-dimensional positioning imaging of the subject.

[0127] In the cross-section reconstruction function 449, the processing circuitry 44 reconstructs a cross-sectional image including a metal region and a metal artifact related to the metal region based on the positioning image. The processing circuitry 44 reconstructs the cross-sectional image using a machine learning model. This machine learning model is called a cross-section reconstruction model, and a cross-sectional image reconstructed using the cross-section reconstruction model is called a reconstructed cross-sectional image.

[0128] In the DL-MAR function 446, the processing circuitry 44 generates a cross-sectional image with reduced metal artifacts (hereinafter referred to as a DL-MAR image) from the restored cross-sectional image using a machine learning model trained by machine learning to reduce metal artifacts.

[0129] An example of the operation of the X-ray computed tomography apparatus 2 according to the second embodiment will be described below. The X-ray computed tomography apparatus 2 uses a cross-sectional reconstruction model and a DL-MAR model.

[0130] Fig. 12 is a diagram showing an example of input and output of a cross-section restoration model according to the second embodiment. As shown in Fig. 12, the cross-section restoration model is a machine learning model trained to input a two-dimensional positioning image and output a restored cross-section image. A deep neural network is used as the machine learning model.

[0131] Fig. 13 is a diagram showing an example of input and output of the DL-MAR model according to the second embodiment. As shown in Fig. 13, the DL-MAR model is a machine learning model trained to input a restored cross-sectional image and output a DL-MAR image. A deep neural network is used as the machine learning model.

[0132] FIG. 14 is a diagram showing an example of a CT examination performed by the X-ray computed tomography apparatus 2 according to the second embodiment. As shown in FIG. 14, the processing circuitry 44 executes two-dimensional positioning imaging to collect two-dimensional cross-sectional images by implementing the imaging control function 441 (step SE1). The two-dimensional positioning imaging in step SE1 is assumed to be low-dose scanogram imaging. In low-dose scanogram imaging, the tabletop 33 on which the subject P is placed is moved along the Z axis while the rotation angle of the X-ray tube 11 is fixed, and low-dose X-ray irradiation by the X-ray tube 11 and X-ray detection by the X-ray detector 12 are repeated to perform X-ray imaging of the entire body of the subject P. The X-ray exposure range is preferably limited to one or several rows of X-ray detection elements to reduce radiation exposure to the subject P. The rotation angle of the X-ray tube 11 is set to any angle, such as 0 degrees or 90 degrees. The DAS 18 collects projection data via the X-ray detector 12, and the processing circuitry 44 generates a two-dimensional position image representing a projection of the entire body of the subject P by implementing an image generation function 443.

[0133] After step SE1 is performed, processing circuitry 44 detects metal regions included in the two-dimensional positioning image collected in step SE1 by implementing metal position identification function 444 (step SE2). In step SE2, processing circuitry 44 performs threshold processing on the positioning image, for example, using a CT value for detecting metal regions as a threshold. A group of pixels having a CT value greater than the threshold is detected as a metal region.

[0134] After step SE2 is performed, the processing circuit 44 determines whether or not to perform a preview display of the MAR processing result (step SE3). Step SE3 may be performed in the same manner as step SC4.

[0135] If it is determined in step SE3 that a preview of the MAR processing result is to be displayed (step SE3: YES), the processing circuit 44 implements the metal position identifying function 444 to identify the position of the metal region detected in step SE2 (step SE4).

[0136] When step SE4 is performed, the processing circuit 44, by implementing the cross-section reconstruction function 449, performs cross-section reconstruction processing on the two-dimensional positioning image acquired in step SE1 to generate a restored cross-sectional image relating to the metal position identified in step SE4 (step SE5). The cross-section of the restored cross-sectional image is an imaging cross-section that passes through the metal position. The restored cross-sectional image includes metal artifacts that are estimated to occur in the reconstructed image obtained by the actual imaging.

[0137] When step SE5 is performed, the processing circuitry 44 performs DL-MAR processing on the restored cross-sectional image generated in step SE5 by implementing the DL-MAR function 446 to generate a DL-MAR image (step SE6). In step SE6, the processing circuitry 44 applies the DL-MAR model to the restored cross-sectional image generated in step SE5 to generate a DL-MAR image, which is a cross-sectional image in which metal artifacts have been reduced.

[0138] When step SE6 is performed, the processing circuitry 44, by implementing the display control function 448, displays the restored cross-sectional image generated in step SE5 and the DL-MAR image generated in step SE6 (step SE7). In step SE7, the processing circuitry 44 displays the restored cross-sectional image and the DL-MAR image on the display 42. This allows the user to compare and observe the restored cross-sectional image and the DL-MAR image. This comparative observation allows the user to determine whether MAR processing is necessary for the actual imaging. For example, if the accuracy of the DL-MAR image is insufficient and the user determines that the actual imaging image needs to be observed using high-accuracy MAR processing, the user determines that MAR processing is necessary for the actual imaging. Furthermore, if the metal region included in the restored cross-sectional image is small or if there are few metal artifacts, the user determines that MAR processing is unnecessary for the actual imaging.

[0139] If step SE7 is performed or if it is determined in step SE3 that the preview display of the MAR processing result is not to be performed (step SE3: NO), the processing circuit 44 determines whether to perform MAR processing (step SE8). The processing of step SE8 is the same as that of step SC9.

[0140] If it is determined that MAR processing is to be performed (step SE8: YES), the processing circuitry 44, by implementing the imaging control function 441, performs the actual imaging and collects the actual imaging sinogram (step SE9). When step SE9 is performed, the processing circuitry 44, by implementing the MAR function 447, performs MAR processing and generates a MAR image from the actual imaging sinogram collected in step SE9 (step SE10). When step SE10 is performed, the processing circuitry 44, by implementing the display control function 448, displays the MAR image generated in step SE10 (step SE11). The processing circuitry 44 displays the MAR image on the display 42. Since metal artifacts are reduced with high precision in the MAR image, the user can make a more accurate diagnosis.

[0141] If it is determined in step SE8 that MAR processing is not to be performed (step SE8: NO), the processing circuitry 44 performs the actual imaging, performs reconstruction processing without MAR processing, generates a normal reconstructed image, and displays it on the display 42. In this way, when MAR processing is not performed, simple actual imaging is performed.

[0142] This completes the CT examination using the X-ray computed tomography apparatus 2 according to the second embodiment.

[0143] As described above, the X-ray computed tomography apparatus 2 according to the second embodiment includes a processing circuit 44. The processing circuit 44 detects metal regions included in positioning images acquired by two-dimensional positioning imaging of a subject. Based on the positioning images, the processing circuit 44 reconstructs cross-sectional images (reconstructed cross-sectional images) including metal regions and metal artifacts related to the metal regions. From the reconstructed cross-sectional images, the processing circuit 44 uses a DL-MAR model trained by machine learning to reduce metal artifacts, to generate DL-MAR images in which metal artifacts included in the reconstructed cross-sectional images have been reduced. The processing circuit 44 displays the DL-MAR images.

[0144] According to the above configuration, the first embodiment can also be used for two-dimensional positioning imaging, thereby improving the workflow of CT examinations including two-dimensional positioning imaging.

[0145] (Third embodiment) In the third embodiment, the DL-MAR processing according to the first and second embodiments is applied to an ultrasonic diagnostic apparatus. The ultrasonic diagnostic apparatus according to the third embodiment will be described below.

[0146] 15 is a diagram showing an example of the configuration of an ultrasound diagnostic apparatus 7 according to the third embodiment. As shown in Fig. 15, the ultrasound diagnostic apparatus 7 includes an ultrasound probe 71, a transmission / reception circuit 72, a B-mode processing circuit 73, a Doppler processing circuit 74, an image processing circuit 75, a display 76, a memory 77, a control circuit 78, and an input interface 79.

[0147] The ultrasonic probe 71 is a device (probe) responsible for transmitting and receiving ultrasonic waves emitted and reflected from the subject, and is formed of an electromechanical reversible conversion element. The ultrasonic probe 71 is configured, for example, as a phased array type having a plurality of elements arranged in an array at its tip. As a result, the ultrasonic probe 71 converts the pulse drive voltage of the supplied drive signal into an ultrasonic pulse signal, transmits it in a desired direction within the scan area of ​​the subject, and converts the ultrasonic signal reflected from the subject into an echo signal with a corresponding voltage.

[0148] With regard to ultrasonic signal transmission, the transmission / reception circuit 72 supplies a drive signal to the ultrasonic probe 71. Specifically, the transmission / reception circuit 72 includes a trigger generation circuit, a delay circuit, a pulser circuit, and the like. The pulser circuit repeatedly generates rate pulses at a predetermined rate frequency to form transmitted ultrasonic waves. The delay circuit provides each rate pulse generated by the pulser circuit with a delay time for each piezoelectric transducer required to focus the ultrasonic waves generated from the ultrasonic probe 71 into a beam and determine the transmission directivity. The trigger generation circuit applies a drive signal (drive pulse) to the ultrasonic probe 71 at a timing based on the rate pulse. In other words, the delay circuit changes the delay time provided to each rate pulse to arbitrarily adjust the transmission direction from the piezoelectric transducer surface.

[0149] The transmitter / receiver circuit 72 has the ability to instantaneously change the transmission frequency, transmission drive voltage, etc. in order to execute a predetermined scan sequence based on instructions from the control circuit 78. In particular, the transmission drive voltage can be changed by an oscillator circuit that can instantaneously switch its value, or by a mechanism that electrically switches between multiple power supply units.

[0150] With regard to ultrasonic signal reception, the transmission / reception circuit 72 performs various processes on echo signals corresponding to reflected wave signals received by the ultrasonic probe 71, converting the echo signals into reflected wave data corresponding to the reception directivity. Specifically, the transmission / reception circuit 72 includes an amplifier circuit, an A / D converter, an adder, and the like. The amplifier circuit amplifies the reflected wave signals for each channel and performs gain correction processing. The A / D converter A / D converts the gain-corrected reflected wave signals and provides the digital data with a delay time required to determine the reception directivity. The adder performs addition processing on the reflected wave signals processed by the A / D converter to generate reflected wave data. The addition processing by the adder emphasizes the reflected wave components from the direction corresponding to the reception directivity of the reflected wave signals.

[0151] The B-mode processing circuit 73 performs logarithmic amplification, envelope detection processing, logarithmic compression, etc. on the reflected wave data from the transmission / reception circuit 72, and generates B-mode information in which the signal strength of each of multiple sample points is expressed as brightness.

[0152] The Doppler processing circuit 74 performs color Doppler analysis on the reflected wave data from the transmission / reception circuit 72 to calculate blood flow information, i.e., Doppler information. In color Doppler analysis, ultrasonic waves are transmitted and received multiple times along the same scan line, and a moving target indicator (MTI) filter is applied to the data sequence at the same position to suppress signals (clutter signals) originating from stationary or slow-moving tissues and extract signals originating from blood flow. Then, color Doppler analysis estimates Doppler information such as blood flow velocity, blood flow dispersion, and blood flow power from this blood flow signal.

[0153] The image processing circuit 75 is a processor that performs image processing. The image processing circuit 75 executes a program stored in the memory 77 to realize functions corresponding to the program. The image processing circuit 75 realizes, for example, an image generation function 751, a reflector position identification function 752, an artifact reduction function 753, and a display control function 754. The functions 751 to 754 do not need to be realized by a single image processing circuit 75, and may be realized by multiple image processing circuits 75 in a shared manner. The functions 751 to 754 may be implemented as hardware rather than as a program.

[0154] In the image generation function 751, the image processing circuit 75 converts (scan conversion) the scanning method of the B-mode information into a scanning method suitable for display, thereby generating a B-mode image of the subject. Similarly, the image processing circuit 75 scan converts (scan conversion) the scanning method of the Doppler information into a scanning method suitable for display, thereby generating a Doppler image of the subject. Display images such as B-mode images and Doppler images are collectively called ultrasound images. The image processing circuit 75 also generates, together with the ultrasound image, information indicating the combination, juxtaposition, and display position of each image information, as well as various information to assist in the operation of the ultrasound diagnostic device 7 and additional information necessary for ultrasound diagnosis, such as patient information. The ultrasound image according to this embodiment may be either a two-dimensional image or a three-dimensional image.

[0155] In the reflector position identification function 752, the image processing circuit 75 calculates the position of a structure included in the ultrasound image. A structure has the property of reflecting ultrasound, so it can also be called a reflector.

[0156] In the artifact reduction function 753, the image processing circuit 75 uses a machine learning model trained by machine learning to reduce artifacts associated with the interaction of ultrasound with structures to generate an image in which artifacts have been reduced (hereinafter referred to as a reduced image) from the ultrasound image. This machine learning model will be referred to as an artifact reduction model.

[0157] In the display control function 754, the image processing circuit 75 outputs various information via the display 76. For example, the image processing circuit 75 displays the reduced image generated by the artifact reduction function 753 on the display 76.

[0158] The display 76 is a display device that converts display information from the image processing circuit 75 into visual video information and displays it in cooperation with the image processing circuit 75. For example, a CRT display, a liquid crystal display, an organic EL display, a plasma display, etc. may be used as the display 76. A projector may also be provided as the display 76.

[0159] The memory 77 is a storage device that stores various information, such as a ROM, RAM, HDD, SSD, integrated circuit storage device, etc. The memory 77 may also be a drive or the like that reads and writes various information from and to a portable storage medium, such as a CD-ROM drive, a DVD drive, or a flash memory.

[0160] The control circuit 78 is a processor that controls the overall processing of the ultrasound diagnostic apparatus 7. The control circuit 78 executes a program stored in the memory 77 to realize a function corresponding to the program. Specifically, the control circuit 78 controls the processing of the transmission / reception circuit 72, the B-mode processing circuit 73, the Doppler processing circuit 74, and the image processing circuit 75 based on various setting requests, various control programs, and various data input by the operator via the input interface 79.

[0161] The input interface 79 is a variety of user interfaces on a touch panel or an operation panel. The operator can input various operations and commands to the ultrasound diagnostic apparatus 7 through the input interface 79. The display 76 and the input interface 79 do not need to be separate, and may be mechanically integrated.

[0162] The transmission / reception circuitry 72, B-mode processing circuitry 73, Doppler processing circuitry 74, image processing circuitry 75, display 76, memory 77, control circuit 78, and input interface 79 are implemented in a single housing also referred to as the device main body, and the ultrasound probe 71 is detachably connected to the device main body via a cable. Note that the hardware configuration of the ultrasound diagnostic device 7 is not limited to this. For example, some or all of the functions of the transmission / reception circuitry 72, B-mode processing circuitry 73, Doppler processing circuitry 74, image processing circuitry 75, display 76, memory 77, control circuit 78, and input interface 79 may be implemented in the ultrasound probe 71. Some or all of the functions of the image processing circuit 75, display 76, and memory 77 may be implemented in a computer connected to the device main body via a network. Furthermore, the image processing circuit 75 and the control circuit 78 do not need to be implemented in separate hardware components and may be implemented in a single piece of hardware.

[0163] Next, an example of the operation of the ultrasonic diagnostic device 7 will be described.

[0164] First, the artifact reduction model used by the ultrasound diagnostic apparatus 7 will be described.

[0165] FIG. 16 is a diagram showing an example of input and output of an artifact reduction model according to the third embodiment. As shown in FIG. 16, the artifact reduction model is a machine learning model trained to input an ultrasound image and output a reduced image, which is an ultrasound image after artifact reduction. A deep neural network is used as the machine learning model. The input ultrasound image can typically be either a B-mode image or a Doppler image. For the specific description below, the ultrasound image is assumed to be a B-mode image. The artifact reduction model is generated by a learning device 5.

[0166] FIG. 17 is a diagram schematically illustrating an example of a process for generating an artifact reduction model by the learning device 5 according to the third embodiment. First, the processing circuitry 55 acquires an ultrasound image UI1 acquired in a previous ultrasound scan from a medical image database, such as an ultrasound diagnostic device, HIS, RIS, or PACS. The ultrasound image UI1 may be either a three-dimensional image or a two-dimensional image. The imaging region is not particularly limited. The subject to be imaged in the ultrasound image UI1 is assumed to contain neither strong nor weak reflectors. A strong reflector is a structure from which almost all ultrasound waves are reflected. Because ultrasound cannot penetrate a strong reflector, areas deeper than the strong reflector become anechoic regions, and low-intensity artifacts called acoustic shadows occur in image regions corresponding to these anechoic regions. A weak reflector is a structure from which almost all ultrasound waves are not reflected or attenuated. Areas deeper than the weak reflector become highly echoic regions, and high-intensity artifacts called backward echo enhancement occur in image regions corresponding to these highly echoic regions. In other words, strong reflectors and weak reflectors are sources of artifacts. Hereinafter, artifacts caused by strong reflectors and weak reflectors will be referred to as reflector artifacts.

[0167] Next, the processing circuitry 55 embeds a pseudo strong reflector region or weak reflector region (hereinafter referred to as a strong / weak reflector region) R1 in the ultrasound image UI1 (step SF1). In step SF1, the processing circuitry 55 embeds the strong / reflector region R1 at an arbitrary position in the ultrasound image UI1. A strong reflector region or a weak reflector region may be embedded as the strong / reflector region R1, or both a strong reflector region and a weak reflector region may be embedded. The embedding is performed by replacing the pixel value at the embedding target position in the ultrasound image UI1 with an arbitrary pixel value corresponding to a strong reflector and / or a weak reflector. The embedding position may be determined randomly.

[0168] After step SF1 is performed, the processing circuitry 55 performs a simulation on the ultrasound image UI2 in which the strong / weak reflector regions R1 are embedded to generate a simulation image UI3 (step SF2). The simulation simulates an ultrasound scan of a subject in which strong reflectors and / or weak reflectors corresponding to the strong / weak reflector regions R1 are embedded. The simulation image UI3 includes strong / weak reflector regions R2 based on the strong / weak reflector regions R1, and reflector artifacts corresponding to the strong / weak reflectors are generated. That is, when a strong reflector region is embedded, the simulation image UI3 includes acoustic shadows as a reflector artifact, and when a weak reflector region is embedded, the simulation image UI3 includes back echo enhancement as a reflector artifact.

[0169] The processing circuitry 55 executes the above steps SF1 and SF2 on ultrasound images of various subjects, and collects an ultrasound image UI2 that includes a strong / weak reflector region R1 but does not include reflector artifacts, and a simulation image UI3 that includes a strong / weak reflector region R2 and includes reflector artifacts, for various subjects. The ultrasound image UI2 and the simulation image UI3 are used as training samples for machine learning. The simulation image UI3 is used as an input training sample, and the ultrasound image UI2 is used as an output training sample. The embedding process of pseudo strong / weak reflector regions in step SF1 can be executed countless times for one ultrasound image UI1, making it easy to collect a large number of training samples.

[0170] The processing circuitry 55 generates an artifact reduction model based on supervised learning using the simulation image UI3 as an input training sample and the ultrasound image UI2 as an output training sample. More specifically, the processing circuitry 55 first applies a machine learning model to the input training sample to perform forward propagation processing to generate an estimated image. Next, the processing circuitry 55 applies a machine learning model that evaluates the difference (error) between the output training sample and the estimated image to perform backpropagation processing, thereby calculating a gradient vector, which is a differential coefficient of an error function that is a function of learning parameters such as weights and biases. Next, the processing circuitry 55 updates the learning parameters based on the gradient vector so as to minimize the error function. These forward propagation processing, backpropagation processing, and parameter update processing are repeated while changing the learning sample until a stopping condition is satisfied. The stopping condition is set when the number of updates reaches a predetermined number, the value of the error function falls below a threshold, or the like. The machine learning model to which the learning parameters assigned when the stopping condition is satisfied is the artifact reduction model. The artifact reduction model is supplied to the ultrasound diagnostic device 7.

[0171] This concludes the description of the artifact reduction model generation process. Note that various modifications of this generation process are possible.

[0172] 18 is a diagram showing an example of the flow of an ultrasound examination by the ultrasound diagnostic apparatus 7 according to the third embodiment. As shown in FIG. 18, the control circuit 78 starts an ultrasound scan (step SG1). In step SG1, the control circuit 78 controls the transmission / reception circuit 72 to perform an ultrasound scan on the subject. The ultrasound scan is assumed to be a two-dimensional B-mode scan.

[0173] When step SG1 is performed, the image processing circuit 75 generates an ultrasound image (B-mode image) of one scanning plane (step SG2) by implementing the image generation function 751. The generated ultrasound image includes strong reflectors and / or weak reflectors, and reflector artifacts resulting from the strong reflectors and / or weak reflectors are assumed to have occurred.

[0174] When step SG2 is performed, the image processing circuit 75 performs artifact reduction processing on the ultrasound image generated in step SG2 by implementing the artifact reduction function 753 to generate a reduced image (step SG3). In step SG3, the image processing circuit 75 applies the ultrasound image to the artifact reduction model to generate a reduced image in which reflector artifacts have been reduced.

[0175] When step SG3 is performed, image processing circuit 75 displays the reduced image generated in step SG3 by implementing display control function 754 (step SG4). In step SG4, image processing circuit 75 immediately displays the reduced image generated in step SG3 on display 76. This makes it possible to display a moving image of the reduced image.

[0176] After step SG4 is performed, the control circuit 78 determines whether or not to end the ultrasound scan (step SG5). If the user determines to end the ultrasound scan, he or she presses a scan end button provided on the ultrasound diagnostic device 7 or a scan end button displayed on the display 76. If the scan end button is pressed, the transmission / reception circuit 72 determines to end the volume CT fluoroscopy. If the imaging end button is not pressed, the control circuit 78 determines not to end the ultrasound scan. Steps SG2 to SG5 are repeated until the scan end button is pressed.

[0177] Then, if it is determined in step SG5 that the ultrasonic scan should be ended (step SG5: YES), the control circuit 78 ends the ultrasonic scan (step SG6).

[0178] This completes the ultrasound examination.

[0179] The third embodiment can be modified in various ways without departing from the spirit of the invention. For example, the following modifications are possible.

[0180] In the above embodiment, the imaging site of the learning sample used for machine learning is not particularly limited. In this case, the imaging site to which the artifact reduction model can be applied is not limited. However, the artifact reduction model may be trained using learning samples of only a specific imaging site. This makes it possible to train an artifact reduction model specialized for the specific imaging site. In this case, it is possible to reduce reflector artifacts in ultrasound images related to the specific imaging site with high accuracy.

[0181] In the above embodiment, the artifact reduction model is capable of processing both artifacts caused by strong reflectors and artifacts caused by weak reflectors. However, the artifact reduction model may be trained as a machine learning model specialized for either strong reflectors or weak reflectors.

[0182] In the above examples, the ultrasound image is a two-dimensional image. However, the ultrasound image according to this embodiment may be a three-dimensional image. In this case, the image processing circuit 75 performs image processing such as threshold processing on the three-dimensional image generated in step SG2 to detect strong / weak reflector regions and calculate the positions of the strong / weak reflector regions. The positions of the strong / weak reflector regions are defined, for example, in the tilt direction of the scanning plane. The image processing circuit 75 then extracts a two-dimensional image corresponding to the positions of the strong / weak reflector regions from the three-dimensional image and applies an artifact reduction model to the two-dimensional image to generate a reduced image. This allows reflector artifact reduction processing using the artifact reduction model to be performed even in three-dimensional scanning.

[0183] As described above, the ultrasound diagnostic apparatus 7 according to the third embodiment includes an image processing circuit 75. The image processing circuit 75 generates an ultrasound image based on echo signals collected via the ultrasound probe 71. The image processing circuit 75 generates a reduced image from the ultrasound image in which reflector artifacts have been reduced, using an artifact reduction model trained by machine learning to reduce reflector artifacts associated with interactions of ultrasound with structures. The image processing circuit 75 displays the reduced image.

[0184] According to the above configuration, it is possible to reduce reflector artifacts caused by strong and weak reflectors through image processing with low computational cost using machine learning. Because of the low computational cost, it is possible to instantly apply reflector artifact reduction processing to ultrasound images generated instantly during ultrasound scanning, and it is possible to instantly and dynamically display ultrasound images with reduced reflector artifacts (reduced images). Since the reduced images can be confirmed when meeting the subject (during ultrasound scanning), it is expected that the quality of ultrasound diagnosis will improve.

[0185] According to at least one of the embodiments described above, artifacts can be easily reduced.

[0186] The term "processor" used in the above description refers to a circuit such as a CPU, a GPU, an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA)). A processor realizes its functions by reading and executing a program stored in a memory circuit. Note that instead of storing a program in a memory circuit, a program may be directly embedded in the processor circuit. In this case, the processor realizes its functions by reading and executing the program embedded in the circuit. Alternatively, instead of executing a program, a function corresponding to the program may be realized by combining logic circuits. Note that each processor in this embodiment is not limited to being configured as a single circuit, but may be configured as a single processor by combining multiple independent circuits to realize its function. Furthermore, multiple components in FIGS. 1, 2, 11, and 15 may be integrated into a single processor to realize its function.

[0187] Although several embodiments have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations of embodiments can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention and its equivalents as defined in the claims. [Explanation of symbols]

[0188] 1,2 X-ray computed tomography equipment 5 Learning Device 7. Ultrasound diagnostic equipment 10 Mounting stand 11 X-ray tube 12 X-ray detector 13 Rotating Frame 14 X-ray high voltage device 15 Control device 16 Wedge 17 Collimator 18 Data Acquisition System (DAS) 30 berths 31 Foundation 32 Support frame 33 Top plate 34 Bed drive unit 40 Console 41 memory 42 Display 43 Input Interface 44 Processing circuit 51 memory 52 Display 53 Input Interface 54 Communication Interface 55 Processing circuit 71 Ultrasound Probe 72 Transmitting and receiving circuit 73 B-mode processing circuit 74 Doppler processing circuit 75 Image processing circuit 76 Display 77 memory 78 Control Circuit 79 Input Interface 441 Shooting control function 442 Sinogram acquisition function 443 Image generation function 444 Metal location identification function 445 Metal Artifact Estimation Function 446 DL-MAR function 447 MAR function 448 Display Control Function 449 Section restoration function 551 Acquisition Function 552 Reconfiguration function 554 Reverse reconstruction function 555 MAR function 556 Machine Learning Functions 557 Display Control Function 751 Image generation function 752 Reflector position identification function 753 Artifact Reduction Function 754 Display Control Function

Claims

1. a generating unit that generates a first reduced image in which metal artifacts have been reduced from a first sinogram acquired by three-dimensional positioning imaging of a subject or a positioning image based on the first sinogram, by utilizing a first machine learning model trained by machine learning to reduce metal artifacts; an estimation unit that generates a pseudo-main image based on the first sinogram or the positioning image, the pseudo-main image being an image that is estimated to be generated by the main photography and that includes a metal artifact that is estimated to be caused by the main photography; a display control unit that displays the first reduced image and the pseudo-actual captured image; Equipped with the estimation unit generates an image by applying a second machine learning model, which has been trained by machine learning to estimate a high-dose image from a low-dose image, to the positioning image, and uses the generated image as the pseudo-actual image. X-ray computed tomography equipment.

2. a metal detection unit that detects a metal region included in the positioning image; the generating unit generates the first reduced image from the first sinogram or the positioning image when the metal region is detected.

2. An X-ray computed tomography apparatus according to claim 1.

3. The metal detection unit identifies a position where the metal region exists, the generating unit generates the first reduced image related to the imaging cross section at the position from the first sinogram or the positioning image.

3. An X-ray computed tomography apparatus according to claim 2.

4. the display control unit displays, on a display screen including the pseudo-actual image and the first reduced image, a button for instructing execution of metal artifact reduction processing without using the first machine learning model; and When execution of the metal artifact reduction processing is instructed via the button, the generation unit executes the metal artifact reduction processing on a second sinogram acquired by the main imaging of the subject or on a main image based on the second sinogram, thereby generating a second reduced image.

2. An X-ray computed tomography apparatus according to claim 1.

5. The X-ray computed tomography apparatus according to claim 1 , wherein the display control unit displays the positioning image together with the pseudo actual image and the first reduced image.

6. 2. The X-ray computed tomography apparatus according to claim 1, wherein the generator generates the first reduced image from the first sinogram or the positioning image regardless of whether or not a metal region is included in the first sinogram or the positioning image.

7. The generation unit applying the first machine learning model to the first sinogram to generate a sinogram, and performing a reconstruction process on the sinogram to generate the first reduced image; applying the first machine learning model to the positioning image to generate the first reduced image; applying the first machine learning model to the first sinogram to generate the first reduced image; or applying the first machine learning model to the positioning image to generate a sinogram, and performing a reconstruction process on the sinogram to generate the first reduced image; 2. An X-ray computed tomography apparatus according to claim 1.

8. the first machine learning model is generated based on supervised learning using training samples including input training samples and output training samples; the input training sample is a third sinogram including a metal component; the output training sample is a fourth sinogram that does not include metal components; 2. An X-ray computed tomography apparatus according to claim 1.

9. the third sinogram is a sinogram generated by forward projecting a reconstructed image in which a pseudo-metal region is embedded onto a reconstructed image based on an original sinogram that does not contain a metal component; The fourth sinogram is a sinogram generated by performing a metal artifact reduction process on the third sinogram without using the first machine learning model.

9. An X-ray computed tomography apparatus according to claim 8.

10. the third sinogram is a sinogram generated by performing an inverse reconstruction process on the first reconstructed image in which the pseudo metal region is embedded, the fourth sinogram is a sinogram used to reconstruct a second reconstructed image that does not include a metal region, or a sinogram generated by performing an inverse reconstruction process on the second reconstructed image; 9. An X-ray computed tomography apparatus according to claim 8.

11. the first machine learning model is generated based on supervised learning using training samples including input training samples and output training samples; The input training samples are reconstructed images that include metal regions and metal artifacts, and the output training samples are reconstructed images that do not include metal regions and metal artifacts.

2. An X-ray computed tomography apparatus according to claim 1.

Images