Medical image processing apparatus, magnetic resonance imaging apparatus, medical image processing method, and medical image processing program

Abstract

To suppress an artifact caused by super-resolution processing and to generate an excellent super-resolution MR image.SOLUTION: A medical image processor includes an image generation part. Input a first magnetic resonance image and a second magnetic resonance image, the first magnetic resonance image being reconstructed by performing a super-resolution process on magnetic resonance data arranged in a k-space, the second magnetic resonance image being an image obtained by taking an image of the same subject as the subject of the first magnetic resonance image and having artifacts therein suppressed to a lower level than the first magnetic resonance image; And generating a third magnetic resonance image on the basis of the first magnetic resonance image and the second magnetic resonance image, by using a trained model configured to output the third magnetic resonance image having the same resolution as the first magnetic resonance image and having the artifacts suppressed therein.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The embodiments disclosed in this specification and the drawings relate to a medical image processing apparatus, a magnetic resonance imaging apparatus, a medical image processing method, and a medical image processing program. [Background technology]

[0002] Super-resolution processing for images has been known in the past. However, when super-resolution processing is performed on magnetic resonance (MR) images, the MR data that is the source of the MR images is data in k-space, so super-resolution processing for conventional images may not be suitable.

[0003] When super-resolution processing is performed on MR data in k-space, zero padding is performed in the high-frequency region outside the MR data in k-space. In this case, when a super-resolution MR image is reconstructed by reconstruction of the MR data with zero padding, Gibbs artifacts appear in the super-resolution MR image. Gibbs artifacts are caused by truncation of data at the edges of the MR data in k-space. In order to reduce Gibbs artifacts, a filter that reduces the truncation of the edge data is applied to the original MR data. However, since the edge data is reduced by applying the filter, the effect of super-resolution may be suppressed. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2020-201823 [Non-patent literature]

[0005] [Non-Patent Document 1] Robert D. Peters, Heide Harris, Steve Lawson, The clinical benefits of AIR Recon DL for MR image reconstruction, http: / / tinyurl.com / AIR-Recon-DL-whitepaper Summary of the Invention [Problem to be solved by the invention]

[0006] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to suppress artifacts that occur during super-resolution processing and generate good super-resolution MR images. 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]

[0007] A medical image processing apparatus according to an embodiment includes an image generation unit having a trained model that receives a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image obtained by imaging the same subject as the first magnetic resonance image but with high-frequency components suppressed, and outputs a third magnetic resonance image, which is a super-resolution image with fewer artifacts than the first magnetic resonance image. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram showing an example of medical image processing according to an embodiment. [Figure 2] FIG. 2 is a diagram showing an example of a magnetic resonance imaging apparatus according to an embodiment. [Figure 3] FIG. 3 is a schematic diagram showing an example of an outline of reconstructing a first MR image according to the embodiment. [Figure 4]FIG. 4 is a schematic diagram showing an example of an outline of reconstructing a second MR image according to the embodiment. [Figure 5] FIG. 5 is a schematic diagram showing an example of an outline of generation of a third MR image corresponding to a one-dimensional super-resolution image by an image generation function according to the embodiment. [Figure 6] FIG. 6 is a diagram showing an example of a super-resolution artifact suppression model according to the embodiment. [Figure 7] FIG. 7 is a flowchart showing an example of the procedure of an image generation process for generating a third MR image from a first MR image and a second MR image using a super-resolution artifact suppression model according to the embodiment. [Figure 8] FIG. 8 is a schematic diagram showing an example of an outline of reconstructing a second MR image according to a modified example of the embodiment. [Figure 9] FIG. 9 is a schematic diagram showing an example of an outline of generation of a third MR image corresponding to a one-dimensional super-resolution image by an image generation function according to a modified example of the embodiment. [Figure 10] FIG. 10 is a diagram showing an example of a super-resolution artifact suppression model according to a modified example of the embodiment. [Figure 11] FIG. 11 is a flowchart showing an example of the procedure of an image generation process for generating a third MR image from a first MR image and a second MR image using a super-resolution artifact suppression model according to a modified example of the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of a medical image processing device, a magnetic resonance imaging (MRI) device, a medical image processing method, and a medical image processing program will be described in detail with reference to the drawings.

[0010] FIG. 1 is a block diagram showing an example of a medical image processing device 1. The medical image processing device 1 is installed, for example, in various modalities capable of generating medical images, or in a server in a hospital. Note that various functions of the medical image processing device 1 may be installed in a server of a medical image management system (hereinafter referred to as a PACS (Picture Archiving and Communication Systems)) or a server of a hospital information system (hereinafter referred to as an HIS (Hospital Information System)). In this case, the medical image processing device 1 is connected via a network to various medical image capturing devices 2 capable of performing magnetic resonance imaging. In this case, the medical image capturing device 2 corresponds to a conventional modality.

[0011] Furthermore, the medical image processing device 1 equipped with various functions includes, for example, a magnetic resonance imaging (hereinafter referred to as MRI) device, a PET (Positron Emission Tomography)-MRI device, a SPECT (Single Photon Emission Computed Tomography)-MRI device, etc. For the sake of concrete explanation, it is assumed below that the medical image processing device 1 is equipped in an MRI device. In this case, the MRI device has various functions in the processing circuitry 15.

[0012] (Embodiment) 2 is a diagram showing an example of an MRI apparatus 100 according to this embodiment. As shown in FIG. 2, the medical image processing apparatus 1 in the MRI apparatus 100 further includes an input / output interface 17. Note that the medical image processing apparatus 1 may not include the input / output interface 17, as shown in FIG. 1. As shown in FIG. 2, the MRI apparatus 100 includes a static magnetic field magnet 101, a gradient magnetic field coil 103, a gradient magnetic field power supply 105, a bed 107, a bed control circuit (system control unit) 109, a transmission circuit 113, a transmission coil 115, a reception coil 117, a reception circuit 119, an imaging control circuit (acquisition unit) 121, a system control circuit (system control unit) 123, a storage device 125, and the medical image processing apparatus 1.

[0013] The static magnetic field magnet 101 is a magnet formed in a hollow, approximately cylindrical shape. The static magnetic field magnet 101 generates a substantially uniform static magnetic field in the internal space. For example, a superconducting magnet or the like is used as the static magnetic field magnet 101.

[0014] The gradient magnetic field coil 103 is a hollow, approximately cylindrical coil and is disposed on the inner surface of the cylindrical cooling vessel. The gradient magnetic field coil 103 receives current individually from a gradient magnetic field power supply 105 to generate gradient magnetic fields whose magnetic field strength varies along the mutually orthogonal X, Y, and Z axes. The gradient magnetic fields of the X, Y, and Z axes generated by the gradient magnetic field coil 103 form, for example, a slice selection gradient magnetic field, a phase encoding gradient magnetic field, and a frequency encoding gradient magnetic field (also referred to as a readout gradient magnetic field). The slice selection gradient magnetic field is used to arbitrarily determine an imaging cross section. The phase encoding gradient magnetic field is used to change the phase of a magnetic resonance signal (hereinafter referred to as an MR (Magnetic Resonance) signal) according to a spatial position. The frequency encoding gradient magnetic field is used to change the frequency of the MR signal according to a spatial position.

[0015] The gradient magnetic field power supply 105 is a power supply device that supplies current to the gradient magnetic field coil 103 under the control of the imaging control circuit 121 .

[0016] The bed 107 is a device equipped with a top plate 1071 on which the subject P is placed. The bed 107 inserts the top plate 1071 on which the subject P is placed into the bore 111 under the control of a bed control circuit 109.

[0017] The bed control circuit 109 is a circuit that controls the bed 107. The bed control circuit 109 drives the bed 107 in response to instructions from the operator via the input / output interface 17, thereby moving the tabletop 1071 in the longitudinal direction, the up-down direction, and in some cases the left-right direction.

[0018] The transmission circuit 113 supplies radio frequency pulses modulated at the Larmor frequency to the transmission coil 115 under the control of the imaging control circuit 121. For example, the transmission circuit 113 includes an oscillator, a phase selection unit, a frequency conversion unit, an amplitude modulation unit, an RF amplifier, and the like. The oscillator generates an RF pulse at a resonance frequency specific to the target atomic nucleus in a static magnetic field. The phase selection unit selects the phase of the RF pulse generated by the oscillator. The frequency conversion unit converts the frequency of the RF pulse output from the phase selection unit. The amplitude modulation unit modulates the amplitude of the RF pulse output from the frequency conversion unit according to, for example, a sinc function. The RF amplifier amplifies the RF pulse output from the amplitude modulation unit and supplies it to the transmission coil 115.

[0019] The transmission coil 115 is an RF (Radio Frequency) coil arranged inside the gradient magnetic field coil 103. In response to the output from the transmission circuit 113, the transmission coil 115 generates an RF pulse corresponding to a high frequency magnetic field.

[0020] The receiving coil 117 is an RF coil arranged inside the gradient magnetic field coil 103. The receiving coil 117 receives MR signals emitted from the subject P by a high frequency magnetic field. The receiving coil 117 outputs the received MR signals to a receiving circuit 119. The receiving coil 117 is, for example, a coil array having one or more, typically a plurality of coil elements. For the sake of concreteness, the receiving coil 117 will be described below as a coil array having a plurality of coil elements.

[0021] The receive coil 117 may be configured by a single coil element. Although the transmit coil 115 and receive coil 117 are shown as separate RF coils in Fig. 2, the transmit coil 115 and receive coil 117 may be implemented as an integrated transmit / receive coil. The transmit / receive coil corresponds to the imaging region of the subject P and is, for example, a local transmit / receive RF coil such as a head coil.

[0022] The receiving circuit 119 generates digital MR signals (hereinafter referred to as MR data) based on the MR signals output from the receiving coil 117 under the control of the imaging control circuit 121. Specifically, the receiving circuit 119 performs various signal processing on the MR signals output from the receiving coil 117, and then performs analog-to-digital (A / D) conversion on the data that has been subjected to various signal processing to generate MR data. The receiving circuit 119 outputs the generated MR data to the imaging control circuit 121. For example, MR data is generated for each coil element and output to the imaging control circuit 121 together with a tag that identifies the coil element.

[0023] The imaging control circuit 121 collects MR data by magnetic resonance imaging of the subject P. Specifically, the imaging control circuit 121 controls the gradient magnetic field power supply 105, the transmission circuit 113, the reception circuit 119, etc. in accordance with the imaging protocol output from the processing circuit 15 to perform imaging of the subject P. The imaging protocol has a pulse sequence according to the type of examination. The imaging protocol defines the magnitude of the current supplied to the gradient magnetic field coil 103 by the gradient magnetic field power supply 105, the timing at which the current is supplied to the gradient magnetic field coil 103 by the gradient magnetic field power supply 105, the magnitude and duration of the radio frequency pulse supplied to the transmission coil 115 by the transmission circuit 113, the timing at which the radio frequency pulse is supplied to the transmission coil 115 by the transmission circuit 113, the timing at which the MR signal is received by the reception coil 117, etc. The imaging control circuit 121 drives the gradient magnetic field power supply 105, the transmission circuit 113, the reception circuit 119, etc. to image the subject P, and then receives MR data from the reception circuit 119 and transfers the received MR data to the medical image processing device 1, etc. The imaging control circuit 121 corresponds to an imaging unit.

[0024] In the above description, an example has been described in which the "processor" reads out and executes a program corresponding to each function from memory 13, but the embodiment is not limited to this. The term "processor" refers to a circuit such as a CPU, a GPU (Graphics Processing Unit), 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)).

[0025] If the processor is a CPU, for example, the processor realizes its functions by reading and executing a program stored in memory 13. On the other hand, if the processor is an ASIC, instead of storing a program in memory 13, the function is directly incorporated into the processor circuit as a logic circuit. 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 functions. Also, although the description has been given assuming that a single storage circuit stores a program corresponding to each processing function, multiple storage circuits may be distributed and arranged, and the processing circuits may read corresponding programs from individual storage circuits.

[0026] The system control circuit 123 has hardware resources such as a processor, a read-only memory (ROM), a random access memory (RAM), etc. (not shown), and uses a system control function to control the MRI apparatus 100. Specifically, the system control circuit 123 reads a system control program stored in the storage device 125, loads it into the memory, and controls each circuit of the MRI apparatus 100 according to the loaded system control program.

[0027] For example, the system control circuit 123 reads out an imaging protocol from the storage device 125 based on imaging conditions input by the operator via the input / output interface 17. The system control circuit 123 transmits the imaging protocol to the imaging control circuit 121 and controls imaging of the subject P. The system control circuit 123 is realized by, for example, a processor. The system control circuit 123 may be incorporated into the processing circuitry 15. In this case, the system control function is executed by the processing circuitry 15, and the processing circuitry 15 functions as a substitute for the system control circuit 123.

[0028] The storage device 125 stores various programs executed by the system control circuit 123, various imaging protocols, imaging conditions including a plurality of imaging parameters that define the imaging protocols, etc. The storage device 125 is, for example, a semiconductor memory element such as a RAM or a flash memory, a hard disk drive (HDD), a solid state drive (SSD), an optical disk, etc. The storage device 125 may also be a drive that reads and writes various information from and to a portable storage medium such as a CD (Compact Disc)-ROM drive, a DVD (Digital Versatile Disc) drive, or a flash memory. Note that the data stored in the storage device 125 may also be stored in the memory 13. In this case, the memory 13 functions as a substitute for the storage device 125.

[0029] The medical image processing apparatus 1 has a communication interface 11, a memory 13, and a processing circuit 15. As shown in FIGS. 1 and 2, in the medical image processing apparatus 1, the communication interface 11, the memory 13, and the processing circuit 15 are electrically connected by a bus. As shown in FIGS. 1 and 2, the medical image processing apparatus 1 is connected to a network via the communication interface 11. The network is communicatively connected to, for example, various modalities and information processing systems within a medical institution, such as an HIS and a Radiology Information System (RIS). Note that the medical image processing apparatus 1 shown in FIG. 1 may also have an input / output interface 17, as shown in FIG. 2, which includes an input interface for inputting various instructions from an operator and a display (output interface) for displaying medical images generated by a generation function 157.

[0030] The communication interface 11 performs data communication with, for example, various modalities that capture images of the subject P during an examination of the subject P, HIS, PACS, etc. The standard for communication between the communication interface 11 and various modalities and hospital information systems may be any standard, and examples include HL7 (Health Level 7), DICOM (Digital Imaging and Communications in Medicine), or both.

[0031] The memory 13 is realized by a storage circuit that stores various types of information. For example, the memory 13 is a storage device such as an HDD, an SSD, or an integrated circuit storage device. The memory 13 corresponds to a storage unit. Note that, in addition to an HDD or an SSD, the memory 13 may also be a drive that reads and writes various types of information from a semiconductor memory element such as a random access memory (RAM) or a flash memory, an optical disc such as a compact disc (CD) or a digital versatile disc (DVD), a portable storage medium, or a semiconductor memory element such as a RAM.

[0032] The memory 13 stores the acquisition function 151, the reconstruction function 153, and the image generation function 155, which are realized by the processing circuitry 15, in the form of programs executable by a computer. The memory 13 stores various data received by the acquisition function 151 via the communication interface 11. Specifically, the memory 13 stores, for example, MR data acquired by the acquisition function 151 from the imaging control circuit 121 or the medical image imaging device 2. The memory 13 also stores magnetic resonance images (hereinafter referred to as MR images) generated by the reconstruction function 153 and the image generation function 155.

[0033] The memory 13 also stores a trained model used in the image generation function 155. The trained model is realized by a model using a neural network that has been trained in advance using, for example, a DNN (Deep Neural Network). Note that the trained model is not limited to a DNN and may be realized by other models. The trained model is a model (hereinafter referred to as a super-resolution artifact suppression model) using a neural network that has been trained to input a first MR image reconstructed by performing super-resolution processing on MR data arranged in k-space and a second MR image of the same imaging target as the first MR image but in which artifacts are suppressed more than in the first MR image, and output a third MR image having the same resolution as the first MR image but in which the artifacts are suppressed.

[0034] The artifact in the first MR image is a Gibbs artifact, which is also called a truncation artifact, a ringing artifact, etc. The first MR image, the second MR image, the third MR image, and a super-resolution artifact suppression model will be described later.

[0035] The processing circuitry 15 controls the entire medical image processing apparatus 1. The processing circuitry 15 is realized by the above-mentioned processor or the like. The processing circuitry 15 includes an acquisition function 151, a reconstruction function 153, an image generation function 155, and the like. The processing circuitry 15, which realizes the acquisition function 151, the reconstruction function 153, and the image generation function 155, respectively, corresponds to an acquisition unit, a reconstruction unit, and an image generation unit. Each function, such as the acquisition function 151, the reconstruction function 153, and the image generation function 155, is stored in the memory 13 in the form of a program executable by a computer. The processing circuitry 15 is a processor. For example, the processing circuitry 15 realizes a function corresponding to each program by reading and executing the program from the memory 13. In other words, the processing circuitry 15, after reading each program, has each function, such as the acquisition function 151, the reconstruction function 153, and the image generation function 155.

[0036] The processing circuitry 15 acquires MR data collected by imaging the subject P using an acquisition function 151. For example, as shown in FIG. 1 , when the processing circuitry 15 is installed in a standalone medical image processing device 1, the acquisition function 151 acquires MR data from a modality capable of performing MR imaging via a network and a communication interface 11. For example, as shown in FIG. 2 , when the processing circuitry 15 is installed in an MRI apparatus 100, the acquisition function 151 acquires MR data generated by a receiving circuitry 119 via an imaging control circuit 121. The acquisition function 151 stores the acquired MR data in the memory 13.

[0037] The processing circuitry 15 reconstructs a first MR image and a second MR image based on the MR data using the reconstruction function 153. Specifically, the reconstruction function 153 reconstructs the first MR image based on the MR data after super-resolution processing. More specifically, the reconstruction function 153 performs super-resolution processing by applying a first filter having a predetermined filter strength to the MR data, and reconstructs the first MR image based on the MR data on which super-resolution processing has been performed. The first filter corresponds to a weak filter that reduces the intensity of the edge of the high-frequency region in the MR data by a weaker filter strength than that of a second filter described below. Note that the first filter may also be a filter that passes the MR data. In this case, the first filter corresponds to a filter that essentially passes the MR data, i.e., a filterless filter.

[0038] The processing circuitry 15 uses the reconstruction function 153 to perform zero-filling in k-space on high-frequency regions of no signal in the MR data to which the first filter has been applied. Zero-filling in k-space is related to, for example, super-resolution in image space, i.e., super-resolution in the reconstructed image. The reconstruction function 153 performs a Fourier transform on the MR data to which the super-resolution processing involving the application of the first filter and zero-filling has been applied, and reconstructs a first MR image. The reconstruction function 153 stores the reconstructed first MR image in the memory 13.

[0039] FIG. 3 is a schematic diagram showing an example of an outline of reconstructing a first MR image. FIG. 3 illustrates, as an example, reconstruction involving super-resolution processing of one-dimensional MR data MD. As shown in FIG. 3, the reconstruction function 153 applies a first filter (weak filter or no filter) to the one-dimensional MR data MD, and then places the MR data to which the first filter has been applied in k-space KS with zero-filling ZF. The reconstruction function 153 then performs a Fourier transform FT on the MR data to which the super-resolution processing involving the application of the first filter and zero-filling has been applied, thereby reconstructing a first MR image MI1. The one-dimensional first MR image MI1 is a high-resolution image in which edge information is preserved. Furthermore, as shown in FIG. 3, a region GA corresponding to a Gibbs artifact appears in the one-dimensional first MR image MI1.

[0040] The reconstruction function 153 also reconstructs a second MR image based on the MR data so as to suppress Gibbs artifacts that occur during reconstruction involving super-resolution processing. As described above, the first and second MR images are images obtained by imaging the same object. That is, the same object does not mean the same subject or region, but means that the MR data underlying the first and second MR images are the same. Specifically, the reconstruction function 153 performs super-resolution processing by applying a second filter having a filter strength stronger than that of the first filter to the MR data, and reconstructs the second MR image based on the MR data on which the super-resolution processing has been performed. The second filter corresponds to a strong filter that reduces the intensity of the edge of the high-frequency region in the MR data more strongly than the filter strength of the first filter. That is, the second filter corresponds to a low-pass filter having a filter strength stronger than that of the first filter, and has the effect of smoothing the MR data. The first and second filters reduce the Gibbs artifacts by weakening the signal intensity at the boundary between the zero-filling and the MR data.

[0041] The reconstruction function 153 performs zero-filling on high-frequency regions of no signal in the k-space in the MR data to which the second filter has been applied. The reconstruction function 153 performs a Fourier transform on the MR data to which the super-resolution process involving the application of the second filter and zero-filling has been applied, and reconstructs a second MR image. As a result, the resolution of the first MR image and the second MR image becomes the same.

[0042] Furthermore, the resolution of the first MR image and the second MR image is higher than that of an MR image generated by reconstructing MR data without using super-resolution processing (hereinafter referred to as a non-super-resolution image), and corresponds to super-resolution of the non-super-resolution image. The reconstruction function 153 stores the reconstructed second MR image in the memory 13. The reconstruction of the first MR image and the second MR image is a reconstruction involving zero filling, and is also called Fine Reconstruction (FineRecon). FineRecon corresponds to a reconstruction that increases the resolution of the output.

[0043] FIG. 4 is a schematic diagram showing an example of an outline of reconstructing a second MR image. As an example, FIG. 4 shows reconstruction involving super-resolution processing of one-dimensional MR data MD. As shown in FIG. 4, the reconstruction function 153 applies a second filter (strong filter) to the one-dimensional MR data MD, and then arranges the MR data to which the second filter has been applied in k-space KS with zero filling ZF. The reconstruction function 153 performs a Fourier transform FT on the MR data to which the super-resolution processing involving the application of the second filter and zero filling has been applied, and reconstructs a second MR image MI2.

[0044] The one-dimensional second MR image MI2 shown in Fig. 4 is a blurred image in which, for example, high-frequency components are suppressed by a strong filter, compared to the first MR image MI1 shown in Fig. 3. However, unlike the first MR image MI1 shown in Fig. 3, the second MR image MI2 has suppressed Gibbs artifacts, as shown in Fig. 4. For convenience of explanation, one-dimensional data and images are shown in Figs. 3 and 4, but in reality, they correspond to two-dimensional data and images. Note that the first MR image and the second MR image are not limited to two dimensions, and if the MR data is three-dimensional data, both become three-dimensional images (volume data).

[0045] The processing circuit 15 generates a third MR image based on the first MR image and the second MR image using the image generation function 155 and a learned model (super-resolution artifact suppression model) stored in the memory 13. Specifically, the image generation function 155 inputs the first MR image and the second MR image into the super-resolution artifact suppression model and outputs the third MR image from the super-resolution artifact suppression model. The third MR image has the same resolution as the first MR image and corresponds to super-resolution of a non-super-resolution image. In addition, the third MR image is an image in which Gibbs artifacts are suppressed (reduced) to the same extent as the second MR image compared to the first MR image. That is, the third MR image has higher resolution than a non-super-resolution image and corresponds to an image in which Gibbs artifacts are suppressed (reduced).

[0046] 5 is a schematic diagram showing an example of an outline of generation of a third MR image MI3 corresponding to a one-dimensional super-resolution image by the image generation function 155. As shown in FIG. 5, the third MR image MI3 has Gibbs artifacts suppressed to the same extent as the second MR image MI2 compared to the first MR image MI1. In addition, the third MR image MI3 maintains edge information (rising pixel values) to the same extent as the first MR image MI1 compared to the second MR image MI2.

[0047] FIG. 6 is a diagram illustrating an example of a super-resolution artifact suppression model 90. The super-resolution artifact suppression model 90 has an input layer 91, an intermediate layer 93, and an output layer 95. The image generation function 155 inputs a first MR image and a second MR image to the input layer 91. As shown in FIG. 6, the components (pixel values) of the first MR image and the second MR image are input to the input layer 91 as a single input vector 92. Here, assuming that the first MR image and the second MR image each have q pixels, the input layer 91 is provided with 2q input units. The input layer 91 is divided into an input unit range 921 for the first MR image and an input unit range 922 for the second MR image. In this case, a pre-processing layer that performs denoising on the first MR image and the second MR image may be provided before the input layer 91. Furthermore, in the input layer 91, the range 921 of input units for the first MR image and the range 922 of input units for the second MR image do not need to be separated.

[0048] 6 is shown as one layer, but is not limited to this, and multiple intermediate layers may be provided between the input layer 91 and the output layer 95. Note that a known structure can be used as appropriate for the intermediate layer 93, and therefore a description thereof will be omitted.

[0049] The output layer 95 shown in FIG. 6 outputs a third MR image. The third MR image is output from the output layer 95 in the form of a single output vector 96. The output vector 96 includes multiple components y. Each component y is a pixel value of each pixel in the third MR image. The range 961 of the output unit of the output layer 95 is limited to the range for the third MR image. The number q of the components y is the same as the number q of pixels in the first MR image and the number q of pixels in the second MR image.

[0050] The image generation process executed by the MRI apparatus 100 of this embodiment configured as described above will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the procedure of the image generation process for generating a third MR image from the first MR image and the second MR image using the super-resolution artifact suppression model 90.

[0051] (Image generation processing) As a preliminary step to the image generation process in FIG. 5, the imaging control circuit 121 acquires MR data of the subject P by magnetic resonance imaging of the subject P.

[0052] (Step S701) The processing circuitry 15 acquires the MR data generated by the receiving circuitry 119 via the imaging control circuitry 121 using the acquisition function 151. The acquisition function 151 stores the acquired MR data in the memory 13. When the processing circuitry 15 is installed in a standalone medical image processing device 1, the acquisition function 151 acquires the MR data via the network and the communication interface 11 from a modality capable of performing MR imaging.

[0053] (Step S702) The processing circuitry 15 performs super-resolution processing by applying a first filter to the MR data using the reconstruction function 153 to reconstruct a first MR image. The reconstruction function 153 stores the reconstructed first MR image in the memory 13. The first MR image is a super-resolution image relative to a non-super-resolution image, and has Gibbs artifacts.

[0054] (Step S703) The processing circuitry 15 uses the reconstruction function 153 to perform super-resolution processing by applying a second filter to the MR data to reconstruct a second MR image. The reconstruction function 153 stores the reconstructed second MR image in the memory 13. The second MR image is a super-resolution image relative to a non-super-resolution image, and Gibbs artifacts are suppressed. That is, the second MR image is an MR image in which Gibbs artifacts are suppressed by the second filter in the super-resolution processing.

[0055] (Step S704) The processing circuitry 15 reads the super-resolution artifact suppression model 90, the first MR image, and the second MR image from the memory 13 using the image generation function 155. The image generation function 155 inputs the first MR image and the second MR image to the super-resolution artifact suppression model 90. The image generation function 155 generates a third MR image based on the output from the super-resolution artifact suppression model 90.

[0056] The MRI apparatus 100 and medical image processing device 1 according to the above-described embodiment input a first MR image reconstructed by performing super-resolution processing on MR data arranged in k-space, and a second MR image obtained by imaging the same subject as the first MR image and in which artifacts are suppressed more than in the first MR image, and generate a third MR image based on the first MR image and the second MR image using a trained model 90 that outputs a third MR image with the same resolution as the first MR image and in which artifacts are suppressed.

[0057] At this time, the MRI apparatus 100 and the medical image processing apparatus 1 according to the embodiment reconstruct a first MR image based on the MR data after the super-resolution processing, and reconstruct a second MR image based on the MR data so as to suppress artifacts caused by the reconstruction involving the super-resolution processing. Specifically, the MRI apparatus 100 and the medical image processing apparatus 1 according to the embodiment reconstruct the first MR image by performing super-resolution processing in which a first filter is applied to the MR data, and reconstruct the second MR image by performing super-resolution processing in which a second filter having a filter strength stronger than that of the first filter is applied to the MR data.

[0058] For these reasons, the MRI apparatus 100 and the medical image processing apparatus 1 according to the embodiment can generate a super-resolution third MR image in which Gibbs artifacts that occur due to super-resolution processing are suppressed, i.e., a high-definition MR image in which Gibbs artifacts are suppressed. For these reasons, the MRI apparatus 100 and the medical image processing apparatus 1 according to the embodiment can generate a super-resolution MR image in which edge information is preserved and Gibbs artifacts are effectively removed. As a result, the MRI apparatus 100 and the medical image processing apparatus 1 according to the embodiment can improve the image quality of the super-resolution image and improve the throughput of diagnoses regarding the subject P.

[0059] (Variation) The difference from the embodiment is that an MR image without FineRecon, i.e., a non-super-resolution image, is used as the second MR image. At this time, the processing circuitry 15 uses the reconstruction function 153 to perform super-resolution processing by applying a first filter to the MR data to reconstruct the first MR image. The first MR image is the same as that described in the embodiment, so its description is omitted. The reconstruction function 153 reconstructs the second MR image based on the MR data without using super-resolution processing. The second MR image in this modification has a lower resolution than the first MR image and corresponds to the non-super-resolution image in the embodiment.

[0060] Furthermore, Gibbs artifacts are more likely to occur due to zero-filling during super-resolution processing. Therefore, Gibbs artifacts in the second MR image in this modified example, in which super-resolution processing is not performed, are more suppressed than Gibbs artifacts in the first MR image. That is, the second MR image in this modified example has a lower resolution than the first MR image, and is an MR image in which Gibbs artifacts are inherently suppressed. In other words, the second MR image in this modified example has a lower resolution than the first MR image, and since super-resolution processing has not been performed, it is an MR image in which Gibbs artifacts are more suppressed than the first MR image.

[0061] FIG. 8 is a schematic diagram showing an example of an outline of reconstructing a second MR image according to this modification. FIG. 8 shows, as an example, reconstruction of one-dimensional MR data MD. As shown in FIG. 8, the reconstruction function 153 places the one-dimensional MR data MD in the k-space KS without performing super-resolution processing that applies a second filter and zero filling. The reconstruction function 153 performs a Fourier transform FT on the MR data MD to reconstruct a second MR image MI22.

[0062] The one-dimensional second MR image MI22 shown in FIG. 8 is a coarse image having a smaller number of pixels than the first MR image MI1 shown in FIG. 3. However, as shown in FIG. 8, unlike the first MR image MI1 shown in FIG. 3, the second MR image MI22 has suppressed Gibbs artifacts. For convenience of explanation, one-dimensional data and images are shown in FIG. 8, but in reality, they correspond to two-dimensional data and images. Note that the first MR image and the second MR image are not limited to two dimensions, and if the MR data is three-dimensional data, both become three-dimensional images (volume data).

[0063] The processing circuitry 15 generates a third MR image based on the first MR image and the second MR image using the super-resolution artifact suppression model stored in the memory 13 through the image generation function 155. The super-resolution artifact suppression model in this modification is trained separately from the embodiment because the second MR image input to the model is different from that in the embodiment.

[0064] 9 is a schematic diagram showing an example of the outline of generation of a third MR image MI3 corresponding to a one-dimensional super-resolution image by the image generation function 155. As shown in FIG. 9, the third MR image MI3 has Gibbs artifacts suppressed to the same extent as the second MR image MI22 compared to the first MR image MI1. In addition, the third MR image MI3 maintains edge information (rising pixel values) to the same extent as the first MR image MI1 compared to the second MR image MI22. In other words, the third MR image MI3 has Gibbs artifacts suppressed to the same extent as the second MR image compared to the first MR image MI1, and is a super-resolution image equivalent to the first MR image relative to the second MR image MI22.

[0065] The super-resolution artifact suppression model in this modification will be described below. Fig. 10 is a diagram showing an example of a super-resolution artifact suppression model 80 in this modification. The super-resolution artifact suppression model 80 has a pre-processing layer 81, an input layer 83, an intermediate layer 85, and an output layer 87. The pre-processing layer 81 has a denoising layer 811 and an up-sampling layer 813. A first MR image is input to the denoising layer 811. A second MR image is input to the up-sampling layer 813.

[0066] The denoising layer 811 removes noise from the first MR image. Note that, as an application example of this modification, the denoising layer 811 may be omitted. In this case, the first MR image is input to a range 921 of input units in the input layer 83.

[0067] The upsampling layer 813 upsamples the resolution of the second MR image to the resolution of the first MR image. Specifically, the upsampling layer 813 uses adjacent pixel values ​​in the second MR image to interpolate pixel values ​​between the adjacent pixel values ​​until the resolution matches that of the first MR image. A known technique can be used as appropriate for upsampling the resolution of the second MR image, i.e., interpolation, and therefore a description thereof will be omitted. Note that parameters such as weights in the preprocessing layer 81 are appropriately adjusted during training of the super-resolution artifact suppression model. For example, the preprocessing layer 81 is trained simultaneously with the artifact suppression model. Note that the training method for the super-resolution artifact suppression model is not limited to the above description. For example, in practice, the preprocessing layer 81 may be trained simultaneously with the intermediate layer 93 or separately. Furthermore, the preprocessing layer 81 may not include parameters.

[0068] The input layer 83, the intermediate layer 85, and the output layer 87 are the same as those in the embodiment, and therefore their explanation will be omitted. Note that the parameters such as weighting in the intermediate layer 85 and the like differ from the parameters such as weighting in the intermediate layer 93 and the like in the embodiment because the data input to the input layer 83 during learning differs from those in the embodiment.

[0069] The image generation function 155 inputs a first MR image and a second MR image to the pre-processing layer 81. Specifically, the image generation function 155 inputs the first MR image to the denoising layer 811 in the pre-processing layer 81. The first MR image denoised by the denoising layer 811 is input to a range 921 of input units in the input layer 83. The image generation function 155 inputs the second MR image to the up-sampling layer 813 in the pre-processing layer 81. The second MR image up-sampled to the same resolution as the first MR image by the up-sampling layer 813 is input to a range 922 of input units in the input layer 83. The other processing is the same as in the embodiment, and therefore description thereof will be omitted.

[0070] The procedure of the image generation process according to this modification will be described below with reference to Fig. 11. Fig. 11 is a flowchart showing an example of the procedure of the image generation process in this modification, in which a third MR image is generated from the first MR image and the second MR image using the super-resolution artifact suppression model 80.

[0071] (Image generation processing) 11, the imaging control circuit 121 acquires MR data of the subject P by magnetic resonance imaging of the subject P. The processes of steps S111 and S112 correspond to the processes of steps S701 and S702 in FIG. 7, and therefore a description thereof will be omitted.

[0072] (Step S113) The processing circuitry 15 reconstructs a second MR image based on the MR data using the reconstruction function 153 without performing super-resolution processing on the MR data. The reconstruction function 153 stores the reconstructed second MR image in the memory 13. Since the second MR image corresponds to a non-super-resolution image, Gibbs artifacts are suppressed.

[0073] (Step S114) The processing circuitry 15 inputs the first and second MR images to the super-resolution artifact suppression model 80 read from the memory 13 by the image generation function 155. The super-resolution artifact suppression model 80 performs upsampling on the second MR image and outputs a third MR image based on the upsampled second MR image and the denoised first MR image. As a result, the image generation function 155 generates the third MR image based on the first and second MR images using the super-resolution artifact suppression model 80.

[0074] According to the MRI apparatus 100 and medical image processing device 1 relating to the modification of the embodiment described above, a first MR image is reconstructed by performing super-resolution processing in which a first filter having a predetermined filter strength is applied to MR data, a second MR image is reconstructed based on the MR data without using super-resolution processing, and a third MR image is generated based on the first MR image and the second MR image using the super-resolution artifact suppression model 80. According to the MRI apparatus 100 and medical image processing device 1 relating to this modification, since super-resolution processing is not performed on the reconstruction of the second MR image, it is possible to shorten the processing time and improve the throughput of the diagnosis for the subject P. Other effects are similar to those of the embodiment, and therefore description thereof will be omitted.

[0075] To summarize the above-described embodiments and variations, the MRI apparatus 100 and medical image processing apparatus 1 according to the embodiments and variations generate a third MR image based on a first MR image and a second MR image that is an MR image of the same imaging subject as the first MR image but has a different resolution than the first MR image (the second MR image in the variations) or an MR image that has different frequency characteristics than the first MR image (the second MR image in the embodiments).

[0076] In the embodiment and the modified example, the number of MR images input to the super-resolution artifact suppression model corresponding to the trained model is not limited to two. For example, in an application example combining the embodiment and the modified example, the super-resolution artifact suppression model is trained so that the first MR image in the embodiment and the modified example, the second MR image described in the embodiment, and the second MR image described in the modified example are input, and a third MR image is output. In this case, the super-resolution artifact suppression model has a preprocessing layer 81, and the input layer 91 is provided with two ranges 922 of input units for the second MR image.

[0077] When the technical ideas of the embodiments and modifications are realized in a medical image processing method, the medical image processing method inputs a first MR image reconstructed by performing super-resolution processing on MR data arranged in k-space, and a second MR image of the same imaging subject as the first MR image but with more suppressed artifacts than the first MR image, and generates a third MR image based on the first MR image and the second MR image using a trained model that outputs a third MR image with the same resolution as the first MR image but with more suppressed artifacts. The procedure and effects of the image generation process for this medical image processing method are similar to those described in the embodiments and modifications, and therefore will not be described again.

[0078] When the technical ideas in the embodiments and variant examples are realized by a medical image processing program, the medical image processing program inputs to a computer a first MR image reconstructed by performing super-resolution processing on MR data arranged in k-space, and a second MR image of the same imaging subject as the first MR image but with more suppressed artifacts than the first MR image, and generates a third MR image based on the first MR image and the second MR image using a trained model that outputs a third MR image with the same resolution as the first MR image and with more suppressed artifacts.

[0079] For example, the image generation process can be realized by installing the medical image processing program in a computer in a modality such as the MRI apparatus 100 or a PACS server, and expanding the program in memory. In this case, the program that can cause the computer to execute the method can also be stored and distributed on a storage medium such as a magnetic disk (hard disk, etc.), an optical disk (CD-ROM, DVD, etc.), or a semiconductor memory. The procedure and effect of the image generation process using the medical image processing program are the same as those in the embodiment and modified examples, so a description thereof will be omitted.

[0080] According to at least one of the embodiments described above, it is possible to suppress artifacts that occur due to super-resolution processing and generate a good super-resolution MR image.

[0081] 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.

[0082] With respect to the above-described embodiments, the following supplementary notes are disclosed as one aspect and optional features of the invention. (Appendix 1) a first MR image; a second MR image of the same imaging target as the first MR image, which has a different resolution from the first MR image or a different frequency characteristic from the first MR image; and a medical image processing device comprising an image generation unit that generates a third MR image based on the above. [Explanation of symbols]

[0083] 1 Medical image processing equipment 2 Medical imaging equipment 11 Communication Interface 13. Memory 15 Processing circuit 17 Input / Output Interface 80 trained models (super-resolution artifact suppression model) 81 Pretreatment layer 83 Input Layer 85 Middle Class 87 Output layer 90 trained models (super-resolution artifact suppression model) 91 Input layer 93 Middle Class 95 Output Layer 96 output vectors 100 Magnetic resonance imaging device 101 Static Magnetic Field Magnet 103 Gradient magnetic field coil 105 Gradient magnetic field power supply 107 Sleeper 109 Bed control circuit 111 Bore 113 Transmitting circuit 115 Transmitting Coil 117 Receiving Coil 119 Receiving circuit 121 Imaging control circuit 123 System Control Circuit 125 Storage device 151 Acquisition Function 153 Reconfiguration function 155 Image generation function 811 Denoising Layer 813 upsampling layers 921 Range of input units for the first MR image 922 Range of input units for the second MR image 961 output unit range 1071 Top plate

Claims

1. an image generation unit having a trained model that receives as input a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image obtained by imaging the same subject as the first magnetic resonance image and in which high-frequency components have been suppressed, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image; A medical image processing device comprising:

2. the second magnetic resonance image is an image obtained by suppressing high frequency components of the first magnetic resonance image; The medical image processing device according to claim 1 .

3. the first magnetic resonance image includes artifacts resulting from the super-resolution processing; The medical image processing device according to claim 1 or 2.

4. The artifact is a Gibbs artifact. The medical image processing device according to claim 3 .

5. the super-resolution processing includes performing zero-filling on a high frequency region of the magnetic resonance data in k-space; The medical image processing device according to claim 1 .

6. The magnetic resonance data is two-dimensional or three-dimensional data. The medical image processing device according to claim 1 .

7. the first magnetic resonance image has been denoised; The medical image processing device according to claim 1 .

8. the second magnetic resonance image has been denoised; The medical image processing device according to claim 1 .

9. an image generating unit having a trained model that receives a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image reconstructed using the magnetic resonance data and has a resolution lower than that of the first magnetic resonance image, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image; A medical image processing device comprising:

10. The second magnetic resonance image is an image reconstructed without using super-resolution processing. The medical image processing device according to claim 9 .

11. the first magnetic resonance image has been denoised; The medical image processing device according to claim 9 or 10.

12. the second magnetic resonance image is upsampled. The medical image processing device according to any one of claims 9 to 11.

13. an imaging unit that acquires magnetic resonance data by magnetic resonance imaging of a subject; an image generation unit having a trained model that receives as input a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on the magnetic resonance data, and a second magnetic resonance image, which is an image obtained by imaging the same subject as the first magnetic resonance image and in which high-frequency components are suppressed, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image; A magnetic resonance imaging apparatus comprising:

14. an imaging unit that acquires magnetic resonance data by magnetic resonance imaging of a subject; an image generation unit having a trained model that receives as input a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on the magnetic resonance data, and a second magnetic resonance image, which is an image reconstructed using the magnetic resonance data and has a resolution lower than that of the first magnetic resonance image, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image; A magnetic resonance imaging apparatus comprising:

15. a trained model that receives as input a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image obtained by imaging the same subject as the first magnetic resonance image and in which high-frequency components have been suppressed, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image, and generates the third magnetic resonance image based on the first magnetic resonance image and the second magnetic resonance image; A medical image processing method comprising:

16. a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image reconstructed using the magnetic resonance data and has a resolution lower than that of the first magnetic resonance image, are input, and a trained model is used to output a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image, to generate the third magnetic resonance image based on the first magnetic resonance image and the second magnetic resonance image; A medical image processing method comprising:

17. On the computer, a trained model that receives as input a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image obtained by imaging the same subject as the first magnetic resonance image and in which high-frequency components have been suppressed, and outputs a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image, and generates the third magnetic resonance image based on the first magnetic resonance image and the second magnetic resonance image; A medical image processing program that makes this possible.

18. On the computer, a first magnetic resonance image, which is a super-resolution image reconstructed by performing super-resolution processing on magnetic resonance data, and a second magnetic resonance image, which is an image reconstructed using the magnetic resonance data and has a resolution lower than that of the first magnetic resonance image, are input, and a trained model is used to output a third magnetic resonance image, which is a super-resolution image in which artifacts are reduced more than in the first magnetic resonance image, to generate the third magnetic resonance image based on the first magnetic resonance image and the second magnetic resonance image; A medical image processing program that makes this possible.

Images