Spectrum generation device and magnetic resonance imaging apparatus

Abstract

To improve accuracy of an artificial spectrum.SOLUTION: A spectrum generation device includes an acquisition unit, an estimation unit, and a generation unit. The processing circuitry acquires, for each of a plurality of voxels included in a volume of interest, a parameter value of at least one shape correlation parameter correlated with a shape in the volume of interest. The estimation unit estimates, for each of the plurality of voxels, a parameter value of at least one type of spectrum generation parameter based on the parameter value of the shape correlation parameter. The generation unit generates a plurality of first artificial spectra corresponding to the plurality of voxels by applying the parameter values of the one or more types of spectrum generation parameters of each of the plurality of voxels to the base spectrum, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The embodiments disclosed in this specification and the drawings relate to a spectrum generating device and a magnetic resonance imaging device. [Background technology]

[0002] Magnetic resonance spectroscopy (MRS) collects an average spectrum within a volume of interest. On the other hand, there is a technique to artificially generate an MRS spectrum within a volume of interest using simulation. However, it is difficult for the artificial spectrum to perfectly reproduce the measured spectrum, and there is a discrepancy between the artificial spectrum and the measured spectrum. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] N. Schmid et al., “Deconvolution of 1D NMR spectra: A deep learning-based approach”, Journal of Magnetic Resonance 347(2023)107357 [Non-patent document 2] William T. Clarke et al., “FSL-MRS: An end-to-end spectroscopy analysis package”, Magn Reson Med. 2021; 85: 2950-2964 Summary of the Invention [Problem to be solved by the invention]

[0004] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to improve the accuracy of the artificial spectrum. 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]

[0005] A spectrum generating device according to an embodiment includes an acquisition unit, an estimation unit, and a generation unit. The acquisition unit acquires, for each of a plurality of voxels included in a volume of interest, parameter values ​​of one or more morphological correlation parameters correlated with a morphology within the volume of interest. The estimation unit estimates, for each of the plurality of voxels, parameter values ​​of one or more spectrum generation parameters based on the parameter values ​​of the morphological correlation parameters. The generation unit applies the parameter values ​​of the one or more spectrum generation parameters for each of the plurality of voxels to a basis spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra. [Brief explanation of the drawings]

[0006] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a magnetic resonance imaging apparatus according to this embodiment. [Figure 2] FIG. 2 is a diagram showing an example of an artificial spectrum generation process performed by the magnetic resonance imaging apparatus shown in FIG. [Figure 3] FIG. 3 is a diagram schematically illustrating the artificial spectrum generation process shown in FIG. [Figure 4] FIG. 4 is a diagram showing a mathematical expression of the spectral signal model Y(ν). [Figure 5] FIG. 5 shows a first display screen of the second artificial spectrum. [Figure 6] FIG. 6 shows a second display screen of the second artificial spectrum. [Figure 7] FIG. 7 is a diagram showing the input / output relationship of the parameter estimation model. [Figure 8] FIG. 8 is a diagram illustrating a process for generating a parameter estimation model. [Figure 9] FIG. 9 is a diagram illustrating the input / output relationship of the parameter estimation model according to the first specific example. [Figure 10] FIG. 10 is a diagram illustrating the input / output relationship of the parameter estimation model according to the second specific example. [Figure 11] FIG. 11 is a diagram illustrating the input / output relationship of the parameter estimation model according to the third specific example. [Figure 12] FIG. 12 is a diagram illustrating the input / output relationship of the parameter estimation model according to the fourth specific example. [Figure 13] FIG. 13 is a diagram illustrating the input / output relationship of a trained model according to the fifth specific example. [Figure 14] FIG. 14 is a diagram schematically illustrating the first stage of the training process for trained model #1 and trained model #2 shown in FIG. [Figure 15] FIG. 15 is a diagram schematically illustrating the second stage of the training process for trained model #1 and trained model #2 shown in FIG. [Figure 16] FIG. 16 is a diagram illustrating the input / output relationship of a trained model according to the sixth specific example. [Figure 17] FIG. 17 is a diagram illustrating an example of the configuration of a spectrum generating device according to the first modification. DETAILED DESCRIPTION OF THE INVENTION

[0007] Hereinafter, the spectrum generating device and magnetic resonance imaging device according to this embodiment will be described in detail with reference to the drawings.

[0008] The spectrum generating device according to this embodiment is a computer that artificially generates various spectra that can be collected by a magnetic resonance imaging apparatus. The spectrum according to this embodiment refers to digital data that represents the frequency distribution of the signal intensity values ​​of a magnetic resonance signal. The spectrum generating device may be a computer incorporated in the magnetic resonance imaging apparatus, or may be a computer separate from the magnetic resonance imaging apparatus. In the following embodiments, the spectrum generating device is assumed to be incorporated in the magnetic resonance imaging apparatus.

[0009] Fig. 1 is a diagram showing an example of the configuration of a magnetic resonance imaging apparatus 1 according to the first embodiment. As shown in Fig. 1, the magnetic resonance imaging apparatus 1 includes a gantry 11, a bed 13, a gradient magnetic field power supply 21, a transmission circuit 23, a reception circuit 25, a bed driving device 27, a sequence control circuit 29, and a spectrum generating device (host computer) 50.

[0010] The gantry 11 has a static magnetic field magnet 41 and a gradient magnetic field coil 43. The static magnetic field magnet 41 and the gradient magnetic field coil 43 are housed in a housing of the gantry 11. A hollow bore is formed in the housing of the gantry 11. A transmitting coil 45 and a receiving coil 47 are arranged in the bore of the gantry 11.

[0011] The static magnetic field magnet 41 has a hollow, approximately cylindrical shape and generates a static magnetic field inside the approximately cylinder. For example, a permanent magnet, a superconducting magnet, or a normal-conducting magnet may be used as the static magnetic field magnet 41. Here, the central axis of the static magnetic field magnet 41 is defined as the Z axis, the axis perpendicular to the Z axis is defined as the Y axis, and the axis horizontally perpendicular to the Z axis is defined as the X axis. The X axis, Y axis, and Z axis form an orthogonal three-dimensional coordinate system.

[0012] The gradient coil 43 is a hollow, approximately cylindrical coil unit attached to the inside of the static magnetic field magnet 41. The gradient coil 43 generates a gradient magnetic field by receiving a current from the gradient power supply 21. More specifically, the gradient coil 43 has three coils corresponding to the X-axis, Y-axis, and Z-axis, which are orthogonal to each other. The three coils form gradient magnetic fields whose field strength varies along each of the X-axis, Y-axis, and Z-axis. The gradient magnetic fields along the X-axis, Y-axis, and Z-axis are combined to form a slice selection gradient magnetic field Gs, a phase encoding gradient magnetic field Gp, and a frequency encoding gradient magnetic field Gr, which are orthogonal to each other, in desired directions. The slice selection gradient magnetic field Gs is used to arbitrarily determine an imaging plane (slice). The phase encoding gradient magnetic field Gp is ​​used to change the phase of a magnetic resonance signal (hereinafter referred to as an MR signal) according to a spatial position. The frequency encoding gradient magnetic field Gr is used to change the frequency of the MR signal according to a spatial position. In the following description, the gradient direction of the slice selection gradient magnetic field Gs is the Z axis, the gradient direction of the phase encoding gradient magnetic field Gp is ​​the Y axis, and the gradient direction of the frequency encoding gradient magnetic field Gr is the X axis.

[0013] The gradient magnetic field power supply 21 supplies a current to the gradient magnetic field coil 43 in accordance with a sequence control signal from the sequence control circuit 29. The gradient magnetic field power supply 21 supplies a current to the gradient magnetic field coil 43, thereby causing the gradient magnetic field coil 43 to generate gradient magnetic fields along the X-axis, Y-axis, and Z-axis. The gradient magnetic fields are superimposed on the static magnetic field formed by the static magnetic field magnet 41 and applied to the subject P.

[0014] The transmission coil 45 is disposed, for example, inside the gradient magnetic field coil 43, and receives a current from the transmission circuit 23 to generate a radio frequency pulse (hereinafter referred to as an RF pulse).

[0015] The transmission circuitry 23 supplies a current to the transmission coil 45 to apply an RF pulse to the subject P via the transmission coil 45, for exciting target protons such as hydrogen nuclei present in the subject P. The RF pulse oscillates at a resonance frequency specific to the target protons, exciting the target protons. An MR signal is generated from the excited target protons and detected by the reception coil 47. The transmission coil 45 is, for example, a whole-body coil (WB coil). The whole-body coil may be used as a transmission / reception coil.

[0016] The receive coil 47 receives MR signals emitted from target protons present in the subject P in response to the action of RF pulses. The receive coil 47 has multiple receive coil elements capable of receiving MR signals. The received MR signals are supplied to the receiver circuit 25 via wire or wirelessly. Although not shown in FIG. 1 , the receive coil 47 has multiple receive channels implemented in parallel. Each receive channel has receive coil elements that receive MR signals and amplifiers that amplify the MR signals. MR signals are output for each receive channel. The total number of receive channels and the total number of receive coil elements may be the same, or the total number of receive channels may be greater or less than the total number of receive coil elements.

[0017] The receiver circuitry 25 receives MR signals generated from excited target protons via the receiver coil 47. The receiver circuitry 25 processes the received MR signals to generate digital MR signals. The digital MR signals can be expressed in k-space, which is defined by spatial frequencies. Therefore, hereinafter, the digital MR signals will be referred to as k-space data. The k-space data is supplied to a host computer 50 via a wired or wireless connection.

[0018] The above-described transmitting coil 45 and receiving coil 47 are merely examples. A transmitting / receiving coil having both transmitting and receiving functions may be used instead of the transmitting coil 45 and receiving coil 47. Furthermore, the transmitting coil 45, receiving coil 47, and transmitting / receiving coil may be combined.

[0019] A bed 13 is installed adjacent to the gantry 11. The bed 13 has a top plate 131 and a base 133. A subject P is placed on the top plate 131. The base 133 supports the top plate 131 so that it can slide along the X-axis, Y-axis, and Z-axis. A bed driving device 27 is housed in the base 133. The bed driving device 27 moves the top plate 131 under the control of a sequence control circuit 29. The bed driving device 27 may include any motor, such as a servo motor or a stepping motor.

[0020] The sequence control circuit 29 has, as hardware resources, a processor such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) and memories such as a ROM (Read Only Memory) and a RAM (Random Access Memory). The sequence control circuit 29 synchronously controls the gradient magnetic field power supply 21, the transmission circuit 23, and the reception circuit 25 based on data acquisition conditions set by the processing circuit 51, and performs data acquisition on the subject P according to the data acquisition conditions to acquire k-space data regarding the subject P.

[0021] The sequence control circuit 29 according to this embodiment is also capable of performing MRS imaging, which is a type of spectrum acquisition. MRS imaging is an imaging method that measures chemical shifts, which are minute differences in the resonance frequencies of target protons that occur depending on differences in chemical environment. MRS imaging includes a single-voxel method that acquires data for a single voxel and a multi-voxel method that acquires data for multiple voxels, and this embodiment is applicable to either method. The multi-voxel method is also called chemical shift imaging (CSI) or MRS imaging (MRSI: Magnetic Resonance Spectroscopic Imaging). The spatial region of the measurement target is called a volume of interest. A volume of interest refers to a spatial region formed by multiple voxels.

[0022] When the sequence control circuit 29 executes MRS imaging, a free induction decay (FID) signal or a spin echo signal is generated from a volume of interest set in the subject P. The receiving circuit 25 receives the FID signal or the spin echo signal via the receiving coil 47 and performs signal processing on the received FID signal or the spin echo signal to acquire k-space data. The acquired k-space data is assumed to be digital data that represents signal intensity values ​​emitted from the volume of interest as a time function. The pulse sequence of MRS imaging is repeated the number of excitations (NEX), and k-space data for the number of excitations is acquired.

[0023] As shown in FIG. 1, the spectrum generating device 50 is a computer having a processing circuit 51 , a memory 53 , a display 55 , an input interface 57 and a communication interface 59 .

[0024] The processing circuitry 51 has a processor such as a CPU as a hardware resource. The processing circuitry 51 functions as the core of the magnetic resonance imaging apparatus 1. For example, the processing circuitry 51 executes various programs to realize a data collection control function 511, an acquisition function 512, an estimation function 513, an artificial spectrum generation function 514, a training function 515, and a display control function 516.

[0025] Using the data acquisition control function 511, the processing circuitry 51 controls the sequence control circuitry 29 to acquire various types of data on the subject P and acquire k-space data via the receiving circuitry 25. Possible types of data acquisition include actual imaging, calibration imaging, and spectrum acquisition. The actual imaging acquires T1-weighted images, T2-weighted images, and other MRI images. The calibration imaging is performed before the actual imaging and acquires a shimming map that represents the spatial distribution of magnetic field inhomogeneity. For spectrum acquisition, MRS imaging is used to acquire MRS spectra. The processing circuitry 51 is also capable of generating various images and spectra based on the acquired k-space data.

[0026] The acquisition function 512 causes the processing circuitry 51 to acquire, for each of multiple voxels included in the volume of interest, parameter values ​​of one or more morphological correlation parameters correlated with the morphology within the volume of interest. The morphological correlation parameters include morphological images representing the morphology within the volume of interest and / or labels of anatomical locations corresponding to the morphology within the volume of interest. The morphological images include acquired images acquired by the sequence control circuitry 29, segmentation images obtained by performing segmentation processing on the acquired images, and / or quantitative value maps generated based on the acquired images. The acquired images include MRI images acquired by actual imaging and / or shimming data acquired by calibration imaging. Shimming data may include a shimming map representing the spatial distribution of magnetic field inhomogeneity and a B0 map representing the spatial distribution of static magnetic field strength. As an example, the processing circuitry 51 may acquire measured spectra of metabolites within the volume of interest acquired by spectral imaging.

[0027] The processing circuitry 51 may generate the parameter values ​​of the above-mentioned morphological correlation parameters, or may receive them from another computer, etc. For example, the processing circuitry 51 can generate MRI images, shimming maps, and measured spectra based on k-space data acquired by the data acquisition control function 511. As another example, the processing circuitry 51 can generate segmentation images and quantitative value maps based on the MRI images generated as described above.

[0028] The processing circuitry 51 uses the estimation function 513 to estimate a plurality of parameter sets corresponding to a plurality of voxels based on the parameter values ​​of the morphological correlation parameters acquired by the acquisition function 512. Each of the plurality of parameter sets includes parameter values ​​of one or more types of spectrum generation parameters.

[0029] The processing circuitry 51 applies the plurality of parameter sets estimated by the estimation function 513 to the basis spectrum using the artificial spectrum generation function 514 to generate a plurality of first artificial spectra corresponding to a plurality of voxels included in the volume of interest. Then, the processing circuitry 51 generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra.

[0030] The training function 515 causes the processing circuitry 51 to train an untrained machine learning model based on a number of training samples to generate a trained model for use in the estimation function 513.

[0031] The display control function 516 causes the processing circuitry 51 to display various information on the display 55. For example, the processing circuitry 51 displays the second artificial spectrum generated by the artificial spectrum generation function 514 on the display 55.

[0032] The memory 53 is a storage device that stores various information, such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), an integrated circuit storage device, etc. The memory 53 may also be a drive device 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.

[0033] The display 55 displays various information using a display control function 516. As the display 55, for example, a CRT display, a liquid crystal display, an organic EL display, an LED display, a plasma display, or any other display known in the art can be appropriately used.

[0034] The input interface 57 includes input devices that accept various commands from the user. Examples of input devices that can be used include a keyboard, a mouse, various switches, a touch screen, and a touch pad. Note that the input devices are not limited to those equipped with physical operating parts such as a mouse and a keyboard. For example, an example of the input interface 57 also includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the magnetic resonance imaging apparatus 1 and outputs the received electrical signal to various circuits. The input interface 57 may also be a voice recognition device that converts a voice signal collected by a microphone into an instruction signal.

[0035] The communication interface 59 is an interface that connects the magnetic resonance imaging apparatus 1 to a workstation, a PACS (Picture Archiving and Communication System), an HIS (Hospital Information System), a RIS (Radiology Information System), etc. via a LAN (Local Area Network), etc. The network IF transmits and receives various types of information to and from the connected workstation, PACS, HIS, and RIS.

[0036] The artificial spectrum generation process performed by the magnetic resonance imaging apparatus 1 will be described in detail below.

[0037] Fig. 2 is a diagram showing an example of an artificial spectrum generation process performed by the magnetic resonance imaging apparatus 1. Fig. 3 is a diagram showing a schematic diagram of the artificial spectrum generation process of Fig. 2.

[0038] First, in step S1, the processing circuitry 51 sets a volume of interest VOI in a morphological image I1 by implementing the acquisition function 512 (step S1). The morphological image I1 is an MRI image representing the internal morphology of the subject P, such as a T1-weighted image, a T2-weighted image, or a FLAIR image, collected by performing MR imaging on the subject P. In other words, the morphological image I1 is a spatial distribution of parameter values ​​of morphological correlation parameters, and each pixel of the morphological image I1 is assigned a parameter value of the morphological correlation parameter. It is assumed that the morphological image I1 has been generated in advance by the processing circuitry 51 prior to step S1.

[0039] The morphological image I1 is displayed on the display 55 by the display control function 516 of the processing circuitry 51, and a volume of interest VOI is set to a desired spatial region indicated by the user via the input interface 57. The volume of interest VOI refers to a spatial region in which the second artificial spectrum is to be generated. The volume of interest VOI is preferably set to be larger than the area of ​​one pixel of the morphological image I1.

[0040] 3 is a head image showing the head of the subject P, and the volume of interest VOI is set in the brain region of the head, but this embodiment is not limited to this, and the volume of interest VOI may be set in any anatomical region such as the heart, liver, breast, prostate, muscle fibers, etc. Furthermore, the shape of the volume of interest VOI is not limited to a square, but may be a rectangle such as a rectangle, or may be a shape combining rectangles of any shape. In the following description, the shape of the volume of interest VOI is assumed to be square.

[0041] When step S1 is performed, the processing circuit 51, by implementing the acquisition function 512, divides the volume of interest VOI set in step S1 into a plurality of voxels xn (n is a subscript representing the voxel number; 1≦n≦N, N is the number of voxels included in the volume of interest VOI) (step S2). The voxels xn have a size according to the spatial resolution of the anatomical image I1 and are typically smaller than the size of the volume of interest VOI. In FIG. 3, as an example, the volume of interest VOI is divided into nine voxels xn (N=9). In other words, the anatomical image I1 is an image collected at a spatial resolution corresponding to N voxels, which is higher than the spatial resolution corresponding to the volume of interest VOI.

[0042] After step S2 is performed, the processing circuitry 51, by implementing the acquisition function 512, acquires a parameter value Pm(xn) of the morphological correlation parameter Pm for each of the multiple voxels xn divided in step S2 (step S3). Specifically, the processing circuitry 51 reads out the parameter value Pm(xn) of the morphological correlation parameter Pm from each voxel xn of the volume of interest VOI from the morphological image I1. Hereinafter, the parameter value of the morphological correlation parameter will be referred to as the morphological correlation parameter value.

[0043] After step S3 is performed, the processing circuitry 51, by implementing the estimation function 513, estimates a parameter value Ps(xn) of the spectrum generation parameter Ps for each of the multiple voxels xn based on the morphological correlation parameter value Pm(xn) acquired in step S3 (step S4). The spectrum generation parameter Ps is a collective term for one or more parameters constituting a spectrum signal model. Here, the spectrum signal model refers to a mathematical model representing a base spectrum with the spectrum generation parameter Ps. Specifically, the spectrum generation parameter Ps includes a baseline relative to the base spectrum, a phase shift relative to the base spectrum, the concentration of metabolites represented by the base spectrum, a half-width of the base spectrum, and / or a frequency shift relative to the base spectrum. In step S4, the processing circuitry 51 applies the parameter value Pm(xn) to a trained model or a random number generator to estimate the parameter value Ps(xn). The trained model and the random number generator are collectively referred to as a spectrum generation parameter estimator. Hereinafter, the parameter value of the spectrum generation parameter will be referred to as a spectrum generation parameter value.

[0044] After step S4 is performed, the processing circuit 51, by implementing the artificial spectrum generation function 514, applies the spectrum generation parameter value Ps(xn) estimated in step S4 to the base spectrum for each of the multiple voxels xn to generate a first artificial spectrum S1(xn) (step S5). The first artificial spectrum S1(xn) is mathematically expressed by a spectral signal model. As described above, the spectral signal model is a mathematical model that represents the base spectrum with the spectral generation parameter Ps.

[0045] FIG. 4 is a diagram showing a mathematical expression of the spectral signal model Y(ν). As shown in FIG. 4, the spectral signal model Y(ν) is expressed as the sum of a first term and a second term. The first term represents the baseline B(ν) of the first artificial spectrum. The second term represents the sum spectrum to which the phase shift exp[i(φ0+νφ1)] is applied. The sum spectrum represents the sum of the artificial spectra for each metabolite l and metabolite group g. Each metabolite l belongs to one group g. In the following embodiment, it is assumed that all metabolites l are aggregated into one group g, i.e., g=1. For example, all metabolites such as NAA and Cho are assigned to one group. Metabolite l may represent a metabolite or a class of metabolites with similar structures. The artificial spectrum for each metabolite l and group g is calculated based on the concentration C of metabolite l and group g. l,g and the numerator term M l,g The numerator term M l,g is the half-width (γ g +σ g 2 ) and frequency shift ε g The basis spectrum m l,g is expressed as the inverse Fourier transform of

[0046] The spectral signal model Y(ν) shown in Fig. 4 has the spectral generation parameters Ps, which are the phase shift exp[i(φ0+νφ1)], the concentration C l,g , the basis spectrum m l,g , half-width (γ g +σ g 2 ) and frequency shift ε gThe half-width is assumed to be the full width at half maximum (FWHM), but is not limited to this and may be any index that represents the degree of width of a spectral peak, such as half width at half maximum (HWHM). ν represents frequency. Note that each metabolite l may be assigned to multiple metabolite groups g. For example, Glx, Glu, and GABA may be assigned to a first group, and other metabolites may be assigned to a second group. In this case, g=2. By assigning metabolite l to multiple groups g, it becomes possible to more flexibly estimate the spectrum generation parameter value Ps(xn).

[0047] An example of a method for generating a first artificial spectrum using a spectral signal model Y(ν) will be described. Human tissue has a combination of metabolites according to the type of tissue, and the combination of spectrum generation parameter values ​​Ps(xn) varies depending on the type of metabolite. The memory 53 stores a first LUT (look up table) that associates the type of human tissue with the metabolite group contained in the human tissue, and a second LUT that associates a combination of spectrum generation parameter values ​​Ps(xn) for each metabolite group. The second LUT stores parameter values ​​Ps(xn) for each of multiple metabolites belonging to each group. The first LUT and the second LUT are assumed to have been generated in advance based on data obtained by collecting actual spectra, data obtained by simulation, etc. The memory 53 also stores a basis spectrum m for each combination of metabolites and metabolite groups. l,g The basis spectrum m l,g is obtained by simulation or by collecting measured spectra on a phantom.

[0048] The processing circuitry 51 identifies the type of human tissue for which the volume of interest VOI is set. The type of human tissue may be identified by image processing using registration of each corresponding reference point between the morphological image and the human body atlas, or may be manually designated via the input interface 57. The processing circuitry 51 inputs the identified type of human tissue into a first LUT to identify the type of metabolite group associated with the type, inputs the identified type into a second LUT, and identifies a combination of parameter values ​​Ps(xn) of the spectrum generation parameter Ps associated with the type. The processing circuitry 51 also calculates the basis spectrum m for each metabolite included in the identified type. l,g is read from the memory 53.

[0049] The processing circuit 51 calculates a combination of the specified parameter values ​​Ps(xn) and the basis spectrum m l,g The first artificial spectrum S1(xn) is generated by applying the above to the spectral signal model Y(ν) shown in Figure 4, adding up the artificial spectra (second term) for all metabolites contained in all voxels xn, and adding the baseline B(ν) to the addition result. This generates a first artificial spectrum as the sum of the artificial spectra corresponding to all metabolites contained in each voxel xn. For example, as shown in Figure 3, when the number of voxels xn is nine (N=9), nine first artificial spectra S1(x1) to S1(x9) are generated.

[0050] The first artificial spectrum S1(xn) may be not only the sum of the artificial spectra of all metabolites contained in each voxel xn, but also an artificial spectrum of each metabolite. Hereinafter, when distinguishing between the two, the first artificial spectrum obtained by summing all the artificial spectra corresponding to all metabolites will be referred to as the first total artificial spectrum, and the first artificial spectrum corresponding to each metabolite will be referred to as the first individual artificial spectrum.

[0051] After step S5 is performed, the processing circuit 51, by implementing the artificial spectrum generation function 514, adds up the multiple first artificial spectra S1(xn) corresponding to the multiple voxels xn generated in step S5 to generate a second artificial spectrum S2(VOI) corresponding to the volume of interest VOI (step S6). For example, as shown in FIG. 3, when the number of voxels xn is nine (N=9), nine first artificial spectra S1(x1) to S1(x9) are generated. As described above, multiple first artificial spectra S1(xn) corresponding to the multiple voxels xn are generated, and then these multiple first artificial spectra S1(xn) are added up to generate one second artificial spectrum S2(VOI) corresponding to the volume of interest VOI. In actual spectrum collection, spectra are collected in units of volumes of interest VOI rather than in units of voxels xn. Therefore, the second artificial spectrum S2(VOI) is expected to be closer in accuracy to the actually collected spectrum than the first artificial spectrum S1.

[0052] In addition to the second total artificial spectrum, a second individual artificial spectrum may be generated as the second artificial spectrum S2(VOI). The second total artificial spectrum can be generated by adding up a plurality of first total artificial spectra corresponding to a plurality of voxels xn included in the volume of interest. The second individual artificial spectrum can be generated by adding up a plurality of first individual artificial spectra corresponding to a plurality of voxels xn included in the volume of interest.

[0053] When step S6 is performed, the processing circuitry 51 displays the second artificial spectrum S2(VOI) generated in step S6 by implementing the display control function 516 (step S7). As an example, the processing circuitry 51 displays the second artificial spectrum S2(VOI) on the display 55 in a predetermined layout.

[0054] FIG. 5 is a diagram showing a first display screen I2 of the second artificial spectrum S2 (VOI). The display screen I2 is displayed on the display 55 by the processing circuit 51. As shown in FIG. 5, a morphological image I21 is displayed on the display screen I2, and a mark I22 representing the volume of interest set in step S1 is superimposed on the morphological image I21. The display screen I2 also displays a second integrated artificial spectrum I23, which was generated in step S6 and corresponds to the volume of interest indicated by the mark I22. Each peak of the second integrated artificial spectrum I23 may be assigned a character string representing the name or symbol of the corresponding metabolite. For example, as shown in FIG. 5, if the volume of interest is set to cerebral gray matter, the volume of interest includes metabolites such as Lac, Cho, Cr, and NAA. Therefore, it is preferable to display the character strings "Lac," "Cho," "Cr," or "NAA" next to each peak of the second integrated artificial spectrum I23.

[0055] The second synthetic artificial spectrum I23 is displayed to allow the user to see the spectrum that may be collected in the volume of interest before the actual spectral collection is performed.

[0056] FIG. 6 is a diagram showing a second display screen I3 of the second artificial spectrum S2 (VOI). The display screen I3 is displayed on the display 55 by the processing circuit 51. As shown in FIG. 6, a morphological image I31 is displayed on the display screen I3, and a mark I32 representing the volume of interest set in step S1 is superimposed on the morphological image I31. The display screen I3 also displays a second individual artificial spectrum I33m (m is the index of the second individual artificial spectrum, 1≦m≦M, M is the number of second individual artificial spectra) generated in step S6 and corresponding to the volume of interest indicated by the mark I32. More specifically, as the second individual artificial spectrum I33m, M second individual artificial spectra I33m corresponding to M metabolites that may be included in the volume of interest are displayed. The second individual artificial spectra I33m are arranged around the morphological image I31.

[0057] For example, as shown in Figure 6, when the volume of interest is set to the gray matter, a spectrum I331 corresponding to Lac, a spectrum I332 corresponding to Cho, a spectrum I333 corresponding to Cr, and a spectrum I334 corresponding to NAA are arranged to surround the morphological image I31. The display of the second individual artificial spectrum I33m allows the user to check metabolites that may be included in the volume of interest and the individual spectra of the metabolites.

[0058] The processing circuitry 51 may switch between displaying the second total artificial spectrum shown in FIG. 5 and displaying the second individual artificial spectrum shown in FIG. 6 in accordance with a switching instruction from the user via the input interface 57.

[0059] When step S7 is performed, the artificial spectrum generation process by the magnetic resonance imaging apparatus 1 ends.

[0060] As described above, the artificial spectrum generation process shown in FIG. 2 makes it possible to accurately generate a second artificial spectrum expected to be collected from a volume of interest before actually performing spectrum collection for the volume of interest. This allows the user to accurately narrow down the size and / or position of the volume of interest from which spectrum collection will be performed in advance. According to this embodiment, it is possible to reduce the number of spectrum collections required to collect a desired actual spectrum, compared to a comparative example in which the artificial spectrum is not confirmed. Furthermore, the processing circuitry 51 generates and displays the second artificial spectrum when the user sets a volume of interest in a morphological image, allowing the user to easily confirm the second artificial spectrum with high accuracy.

[0061] Next, we will explain the trained model used to estimate the spectrum generation parameter values ​​in step S4. Hereinafter, this trained model will be referred to as a parameter estimation model.

[0062] FIG. 7 is a diagram showing the input / output relationship of the parameter estimation model. As shown in FIG. 7, the parameter estimation model is a machine learning model that inputs morphological correlation parameter values ​​and outputs spectrum generation parameter values. The parameter estimation model is assumed to be a neural network configured by combining a fully connected layer, a convolutional layer, a pooling layer, a normalization layer, and / or any other network layers. The morphological correlation parameter values ​​may be input as a vector whose elements include multiple morphological correlation parameter values ​​corresponding to multiple voxels included in the volume of interest, or may be input as image data of the volume of interest. The spectrum generation parameter values ​​are output as a vector whose elements include multiple spectrum generation parameter values ​​corresponding to multiple voxels included in the volume of interest. As described above, the spectrum generation parameter values ​​include the baseline, phase shift, half-width, and / or concentration as elements.

[0063] Next, the process of generating a parameter estimation model by the training function 515 will be described. FIG. 8 is a diagram schematically illustrating the process of generating a parameter estimation model. The neural network shown in FIG. 8 refers to a parameter estimation model to be machine-learned. The processing circuit 51 trains the neural network based on supervised machine learning using multiple training samples. Each training sample includes input parameter values ​​and a reference spectrum corresponding to the input parameter values. The reference spectrum is an actually measured spectrum for a target volume and is used as ground truth data. The NEX of the reference spectrum is not particularly limited as long as it is 1 or more, but NEX=1 is assumed. The input parameter values ​​are parameter values ​​of morphological correlation parameters for the target volume and are used as input data. By performing machine learning based on such training samples, the neural network can learn the correlation between the input parameter values ​​and the reference spectrum.

[0064] The processing circuit 51 inputs input parameter values ​​into the neural network and performs forward propagation processing based on the input data to calculate predicted parameter values. The processing circuit 51 generates a first artificial spectrum from the calculated predicted parameter values ​​using a spectral signal model. The processing circuit 51 calculates an L1 loss, which evaluates the difference between the generated first artificial spectrum and a reference spectrum, and updates the network parameters of the neural network so that the calculated L1 loss is reduced. The network parameters include weight coefficients and biases between network layers. The processing circuit 51 repeatedly updates the network parameters while changing the training sample until a predetermined termination condition is met. The network parameter update process may use stochastic gradient descent, Adam, or any other optimization method. When the termination condition is met, the network parameters at that update count are stored as trained network parameters. The neural network to which the trained network parameters are assigned is used as a parameter estimation model. Using the parameter estimation model makes it possible to obtain spectrum generation parameter values ​​correlated with morphological correlation parameter values.

[0065] The learning method for the parameter estimation model is not limited to the above-mentioned supervised learning, and may be performed by any method, such as unsupervised learning, semi-supervised learning, or self-supervised learning, as long as it is possible to generate a machine learning model that inputs morphological correlation parameter values ​​and outputs spectrum generation parameter values.

[0066] The processing circuit 51 may estimate the spectrum generation parameter value by applying the morphological correlation parameter value to a random number generator. The random number generator refers to an algorithm that converts the morphological correlation parameter value into a random number. The range of the random number is preferably limited to the range of the spectrum generation parameter value. As an example, the processing circuit 51 sets the morphological correlation parameter value as a seed value and inputs the seed value to the random number generator to generate a random number. The generated random number is set as the spectrum generation parameter value. When a random number generator is used, it is possible to obtain the spectrum generation parameter value more easily than when a parameter estimation model is used.

[0067] Next, a specific example of input and output of the parameter estimation model will be described.

[0068] <Example 1> Fig. 9 is a diagram showing the input / output relationship of the parameter estimation model according to specific example 1. As shown in Fig. 9, one type of morphological correlation parameter value is input to the parameter estimation model according to specific example 1, and a spectrum generation parameter value is output. The pixel value of the morphological image is used as the morphological correlation parameter value.

[0069] The morphological image may be an MRI image such as a T1-weighted image or a T2-weighted image acquired by the magnetic resonance imaging apparatus 1, or a segmentation image generated by applying segmentation processing to the MRI image. For example, in the case of an acquired image showing the morphology of a subject's brain, the acquired image is divided into a gray matter image, a white matter image, a CSF (cerebrospinal fluid) image, and other partial images by segmentation processing. Furthermore, instead of an MRI image, a quantitative value map generated by analyzing the MRI image may be used as the morphological image.

[0070] Since the parameter estimation model only needs to process pixel values ​​within a volume of interest in a morphological image, for example, the parameter estimation model may receive pixel values ​​of the volume of interest in the morphological image. Alternatively, the parameter estimation model may receive pixel values ​​of the entire morphological image. In this case, the parameter estimation model may include a network layer that extracts pixel values ​​within the volume of interest from the pixel values ​​of the morphological image.

[0071] <Example 2> Fig. 10 is a diagram showing the input / output relationship of the parameter estimation model according to specific example 2. As shown in Fig. 10, two types of morphological correlation parameter values ​​are input to the parameter estimation model according to specific example 2, and a spectrum generation parameter value is output. The morphological correlation parameter values ​​used are pixel values ​​of an MRI image and pixel values ​​of a shimming map. Because the shimming map represents the spatial distribution of magnetic field inhomogeneity, determining the spectrum generation parameter value by taking into account the pixel values ​​of the shimming map in addition to the pixel values ​​of the MRI image is expected to improve the accuracy of the spectrum generation parameter value.

[0072] As with the MRI image, the pixel values ​​of an image region of the shimming map corresponding to the volume of interest may be input to the parameter estimation model, or the pixel values ​​of the entire shimming map may be input. Here, to match the resolution of the shimming map with the resolution of the MRI image, the processing circuitry 51 may upsample the shimming map and input the upsampled shimming map to the parameter estimation model. As another example, a network layer that upsamples the shimming map may be provided in the parameter estimation model. Furthermore, a quantitative value map may be used instead of an MRI image.

[0073] <Example 3> FIG. 11 is a diagram showing the input / output relationship of a parameter estimation model according to specific example 3. As shown in FIG. 11, one type of morphological correlation parameter value and an MRS spectrum are input to the parameter estimation model according to specific example 3, and a spectrum generation parameter value is output. The pixel values ​​of a shimming map are used as the morphological correlation parameter value. The MRS spectrum is an example of a measured spectrum. A spectrum actually collected by MRS imaging (measured spectrum) is used as the MRS spectrum. The measured spectrum is collected by performing MRS imaging in advance at a position identical to or close to the volume of interest set in the morphological image. Because the MRS spectrum reflects the composition within the volume of interest, improving the accuracy of the spectrum generation parameter value is expected by determining the spectrum generation parameter value by taking into account the MRS spectrum in addition to the pixel values ​​of the morphological image.

[0074] The method of inputting pixel values ​​of the shimming map to the parameter estimation model is the same as in Specific Example 2. Also, like in Specific Example 2, upsampling may be performed at any timing. In Specific Example 3, a collected image or a quantitative value map may be used instead of the shimming map. Also, the number of morphological correlation parameter values ​​input to the parameter estimation model is not limited to one type, and two or more types may be used as in Specific Example 1.

[0075] <Example 4> FIG. 12 is a diagram showing the input / output relationship of a parameter estimation model according to specific example 4. As shown in FIG. 12, two types of morphological correlation parameter values ​​are input to the parameter estimation model according to specific example 4, and a spectrum generation parameter value is output. The morphological correlation parameter values ​​are made up of pixel values ​​of a shimming map and labels of anatomical regions. The labels of anatomical regions refer to text data such as the names and symbols of anatomical regions in which a volume of interest is set. Since the labels of anatomical regions have values ​​that correlate with the morphology within the volume of interest, determining the spectrum generation parameter values ​​by taking into account the labels of anatomical regions in addition to the pixel values ​​of the shimming map is expected to improve the accuracy of the spectrum generation parameter values.

[0076] The method of inputting pixel values ​​of the shimming map to the parameter estimation model is the same as in Specific Example 2. Also, like in Specific Example 2, upsampling may be performed at any timing. In Specific Example 4, a collected image or a quantitative value map may be used instead of the shimming map. Also, the number of morphological correlation parameter values ​​input to the parameter estimation model is not limited to one type, and two or more types may be used as in Specific Example 1.

[0077] <Example 5> A spectrum generating device 50 according to a fifth specific example uses a trained model to generate an artificial spectrum of a volume of interest directly from a morphological correlation parameter value for each voxel, without going through a spectrum generation parameter value for each pixel value, thereby reducing the computational load for generating the artificial spectrum and improving the accuracy of the artificial spectrum.

[0078] Fig. 13 is a diagram showing the input / output relationship of a trained model according to specific example 5. In Fig. 13, the volume of interest VOI is, for example, a rectangular region composed of nine pixels x1 to x9. The volume of interest VOI according to specific example 5 can be applied to any size and shape.

[0079] As shown in Fig. 13, the processing circuitry 51 acquires, by the acquisition function 512, an anatomical image and a shimming map relating to an imaging region including a volume of interest of a subject. As parameter values ​​for the anatomical image and the shimming map, it is sufficient to acquire pixel values ​​of nine pixels x1 to x9 constituting the volume of interest. Furthermore, by the acquisition function 512, the processing circuitry 51 acquires concentrations of metabolites that may be contained in the volume of interest (hereinafter referred to as metabolite concentrations). Data on metabolite concentrations can be obtained from various documents such as papers, textbooks, experimental data, and electronic medical records.

[0080] Using the artificial spectrum generation function 514, the processing circuit 51 applies pixel values ​​of a shimming map for the volume of interest and concentrations of metabolites that may be included in the volume of interest to the trained model #1 to generate multiple artificial spectra #1 corresponding to multiple voxels included in the volume of interest. Next, the processing circuit 51 generates a second artificial spectrum #2 corresponding to the volume of interest based on the generated multiple first artificial spectra #1. The artificial spectrum #1 contains concentration value (signal intensity value) components, molecular components, and phase shift components related to metabolites, but does not contain a baseline component. The processing circuit 51 also applies pixel values ​​of the shimming map for the volume of interest and pixel values ​​of the morphological image to the trained model #2 to generate multiple artificial baselines #1 corresponding to multiple voxels. Next, the processing circuit 51 generates a second artificial baseline #2 corresponding to the volume of interest based on the generated multiple first artificial baselines #1.

[0081] The processing circuitry 51 then adds the artificial spectrum #2 and the artificial baseline #2 to generate an artificial spectrum #3 related to the volume of interest. The artificial spectrum #3 includes a baseline component as well as concentration value (signal intensity value) components, molecular components, and phase shift components related to metabolites, as shown in FIG. 4. Specifically, the processing circuitry 51 performs a weighted addition of the artificial spectrum #2 and the artificial baseline #2 to generate the artificial spectrum #3. As an example, the processing circuitry 51 multiplies the artificial spectrum #2 by a weighting coefficient α, multiplies the artificial baseline #2 by a weighting coefficient β, and adds the artificial spectrum #2 with the weighting coefficient α to the artificial baseline #2 with the weighting coefficient β to generate the artificial spectrum #3.

[0082] Trained model #1 and trained model #2 are trained in advance and stored in memory 53 or the like. Trained model #1 and trained model #2 may be managed as individual machine learning models, or trained model #1, trained model #2, and the adder may be collectively managed as a single trained model #0. The adder refers to a network layer that adds artificial spectrum #2 and artificial baseline #2.

[0083] When expressed as trained model #0, processing circuit 51 applies the parameter values ​​(pixel values) of the shimming map for the volume of interest, the parameter values ​​(pixel values) of the morphological image for the volume of interest, and the metabolite concentrations that may be contained in the volume of interest to trained model #0 to generate artificial spectrum #3 for the volume of interest.

[0084] Next, the training process of trained model #1 and trained model #2 will be described with reference to Figures 14 and 15. The training process is executed by the training function 515 of the processing circuit 51. The training process is composed of a first stage in which trained model #1 and trained model #2 are trained individually, and a second stage in which trained model #1 and trained model #2 are fine-tuned together.

[0085] Trained model #1 and trained model #2 are generated by machine learning #3, which minimizes the difference between estimated artificial spectrum #3, which is obtained by adding estimated artificial spectrum #2 corresponding to a volume of interest based on multiple estimated artificial spectra #1 corresponding to multiple voxels output by trained neural network #1 and estimated artificial baseline #2 corresponding to a volume of interest based on multiple artificial baselines #1 corresponding to multiple voxels output by trained neural network #2, and correct spectrum #3. Here, weighting coefficient α for artificial spectrum #2 and weighting coefficient β for artificial baseline #2 are trained in machine learning #3. Trained neural network #1 is generated by machine learning #1, which minimizes the difference between estimated artificial spectrum #2 corresponding to a volume of interest based on multiple estimated artificial spectra #1 corresponding to multiple voxels output by untrained neural network #1, and correct spectrum #2 calculated by a mathematical model. The trained neural network #2 is generated by machine learning #2, which reduces the difference between an estimated artificial baseline #2 corresponding to a volume of interest based on multiple estimated artificial baselines #1 corresponding to multiple voxels output by the untrained neural network #2 and the correct baseline obtained by subtracting the correct spectrum #2 from the correct spectrum #3.

[0086] 14 is a diagram showing a schematic diagram of the first stage of the training process for trained model #1 and trained model #2. Untrained neural network #1 is a neural network corresponding to trained model #1 before two stages of machine learning are performed, and untrained neural network #2 is a neural network corresponding to trained model #2 before two stages of machine learning are performed.

[0087] The processing circuitry 51 acquires multiple training samples from different subjects and / or under different acquisition conditions. Each training sample is a combination of pixel values ​​of a shimming map for a volume of interest, pixel values ​​of a morphological image for the volume of interest, and metabolite concentrations that may be included in the volume of interest. The processing circuitry 51 also acquires a correct spectrum #2, a correct spectrum #3 (shown in FIG. 15), and a correct baseline #2 for each training sample. The correct spectrum #2 is correct data of an artificial spectrum that does not include a baseline component and is calculated using a mathematical model based on the metabolite concentrations in the training sample. For example, the processing circuitry 51 calculates the correct spectrum #2 as the sum of the concentration component and the molecular component of the spectral signal model Y(ν) shown in FIG. 4. The correct spectrum #3 is an observed spectrum obtained by performing spectral collection, such as MRS or NMR, on the subject of the training sample. The correct baseline is correct data of the baseline. The correct baseline is generated by subtracting the correct spectrum #2 from the correct spectrum #3.

[0088] First, machine learning #1 will be described. For each training sample, the processing circuit 51 inputs pixel values ​​of a shimming map for a volume of interest and metabolite concentrations that may be contained in the volume of interest into the untrained neural network #1, and estimates an artificial spectrum #1 corresponding to each of the multiple voxels contained in the volume of interest and that does not contain a baseline component (hereinafter referred to as the estimated artificial spectrum #1). Next, the processing circuit 51 averages the multiple estimated artificial spectra #1 corresponding to the multiple voxels, respectively, to generate an artificial spectrum #2 corresponding to the volume of interest and that does not contain a baseline component (hereinafter referred to as the estimated artificial spectrum #2). The processing circuit 51 calculates the loss between the estimated artificial spectrum #2 and the correct spectrum #2. The L1 loss or L2 loss can be used as the loss. The processing circuit 51 updates the network parameters of the untrained neural network #1 so that the calculated loss becomes smaller. The processing circuit 51 repeatedly updates the network parameters while changing the training sample until a predetermined termination condition is met. When the termination condition is met, the network parameters for that update are stored as the network parameters at the completion of machine learning #1. The neural network to which the network parameters are assigned is used as neural network #1 during training (shown in FIG. 15).

[0089] Next, machine learning #2 will be described. For each training sample, the processing circuit 51 inputs pixel values ​​of a shimming map for the volume of interest and pixel values ​​of a morphological image for the volume of interest into the untrained neural network #2, and estimates an artificial baseline #1 corresponding to each of the multiple voxels included in the volume of interest (hereinafter, referred to as the estimated artificial baseline #1). Next, the processing circuit 51 averages the multiple estimated artificial baselines #1 corresponding to the multiple voxels, respectively, to generate an artificial baseline #2 corresponding to the volume of interest (hereinafter, referred to as the estimated artificial baseline #2). The processing circuit 51 calculates the loss between the estimated artificial baseline #2 and the correct artificial baseline #2. The loss may be an L1 loss or an L2 loss. The processing circuit 51 updates the network parameters of the untrained neural network #2 so that the calculated loss is reduced. The processing circuit 51 repeatedly updates the network parameters while changing the training sample until a predetermined termination condition is met. When the termination condition is met, the network parameters for that update are stored as the network parameters at the completion of machine learning #2. The neural network to which the network parameters are assigned is used as neural network #2 during training (shown in FIG. 15).

[0090] FIG. 15 is a diagram schematically illustrating the second stage of the training process for trained model #1 and trained model #2. As shown in FIG. 15, for each training sample, the processing circuit 51 inputs pixel values ​​of a shimming map for a volume of interest and metabolite concentrations that may be contained in the volume of interest into the in-training neural network #1, and estimates multiple estimated artificial spectra #1 corresponding to multiple voxels. Next, the processing circuit 51 averages the multiple estimated artificial spectra #1 to generate an estimated artificial spectrum #2 corresponding to the volume of interest. Furthermore, for each training sample, the processing circuit 51 inputs pixel values ​​of a shimming map for a volume of interest and pixel values ​​of a morphological image into the in-training neural network #2, and estimates multiple estimated artificial baselines #1 corresponding to multiple voxels. Next, the processing circuit 51 averages the multiple estimated artificial baselines #1 to generate an estimated artificial baseline #2 corresponding to the volume of interest.

[0091] Next, the processing circuit 51 multiplies the estimated artificial spectrum #2 by a weighting factor α, multiplies the estimated artificial baseline #2 by a weighting factor β, and adds the estimated artificial spectrum #2 with the weighting factor α to the estimated artificial baseline #2 with the weighting factor β to generate an estimated artificial spectrum #3 for the volume of interest. The processing circuit 51 calculates the loss between the estimated artificial spectrum #3 and the ground truth spectrum #3. L1 loss or L2 loss can be used as the loss. The processing circuit 51 updates the network parameters #1 of the neural network #1 in the midst of training, the network parameters #2 of the neural network #2 in the midst of training, the weighting factor α, and / or the weighting factor β so as to reduce the calculated loss. The processing circuit 51 repeats the above update process while changing the training sample until a predetermined termination condition is met. When the termination condition is met, the network parameters #1, network parameters #2, weighting factor α, and / or weighting factor β for that update are stored as the network parameters #1, network parameters #2, weighting factor α, and / or weighting factor β at the time of completion of machine learning #3. The neural network to which the network parameter #1 is assigned is used as trained model #1, and the neural network to which the network parameter #2 is assigned is used as trained model #2.

[0092] This completes the training process for trained model #1 and trained model #2.

[0093] As described above, in Example 5, by using a neural network, it is possible to estimate an artificial spectrum for a volume of interest from a combination of pixel values ​​of a shimming map for the volume of interest, pixel values ​​of a morphological image, and metabolite concentrations, thereby making it possible to obtain an artificial spectrum for the volume of interest with a low computational load and high accuracy.

[0094] <Example 6> Specific Example 6 is a modified example of Specific Example 5. A spectrum generation device 50 according to Specific Example 6 generates an artificial spectrum of a volume of interest from a morphological correlation parameter value for each pixel value via a spectrum generation parameter value for each voxel while using a trained model. As a result, since the spectrum generation parameter value for each voxel is used, it is expected that the accuracy of the artificial spectrum will be improved compared to Specific Example 5.

[0095] Fig. 16 is a diagram showing the input / output relationship of a trained model according to specific example 6. In Fig. 16, the volume of interest VOI is, for example, a rectangular region composed of nine pixels x1 to x9. The volume of interest VOI according to specific example 6 can be applied to any size and shape.

[0096] 16, the processing circuitry 51 acquires a morphological image and a shimming map relating to an imaging region including a volume of interest of a subject using the acquisition function 512. The processing circuitry 51 also acquires concentrations of metabolites (metabolite concentrations) that may be included in the volume of interest using the acquisition function 512. The morphological image, shimming map, and metabolite concentrations are the same as those in Example 5.

[0097] Using the estimation function 513, the processing circuitry 51 estimates the parameter value of the spectrum generation parameter (spectrum generation parameter value) for each of the multiple voxels included in the volume of interest by applying the parameter value of the shimming map and the metabolite concentration that may be included in the volume of interest to the trained model #5. The trained model #5 is a neural network trained to input the parameter value of the shimming map for each voxel and the metabolite concentration that may be included in the volume of interest, and output the spectrum generation parameter value for that voxel.

[0098] The artificial spectrum generation function 514 causes the processing circuitry 51 to generate an artificial spectrum #1 for the volume of interest based on the spectrum generation parameter values ​​for each of the plurality of voxels. For example, as shown in FIG. 3, the processing circuitry 51 applies the spectrum generation parameter values ​​to a spectral signal model for each of the plurality of voxels to generate an artificial spectrum that does not include a baseline component, and then generates an artificial spectrum #2 for the volume of interest that does not include a baseline component by summing the artificial spectra across the plurality of voxels that make up the volume of interest.

[0099] Meanwhile, the processing circuitry 51 generates multiple artificial baselines #1 corresponding to multiple voxels by applying the parameter values ​​of the shimming map related to the volume of interest and the parameter values ​​of the morphological image to the trained model #2 using the artificial spectrum generation function 514, and generates an artificial baseline #2 related to the volume of interest by averaging the multiple artificial baselines #1. This generation process is the same as the generation process of the artificial baseline according to the fifth specific example.

[0100] Then, the processing circuit 51 generates artificial spectrum #3 by adding the artificial spectrum #2 and the artificial baseline #2 using the artificial spectrum generation function 514. Specifically, the processing circuit 51 generates artificial spectrum #2 by performing weighted addition of the artificial spectrum #2 and the artificial baseline #2. As an example, the processing circuit 51 multiplies the artificial spectrum #2 by a weighting coefficient α, multiplies the artificial baseline #2 by a weighting coefficient β, and adds the artificial spectrum #2 with the weighting coefficient α to the artificial baseline #2 with the weighting coefficient β to generate artificial spectrum #3.

[0101] The trained model #5 and the trained model #2 are trained in advance and stored in memory 53 or the like. The trained model #5 and the trained model #2 may be managed as individual machine learning models, or the trained model #5, the trained model #2, and the adder may be collectively managed as a single trained model #4. The adder refers to a network layer that adds the artificial spectrum #2 and the artificial baseline #2.

[0102] (Variation 1) The spectrum generating device according to the above embodiment is incorporated into the magnetic resonance imaging apparatus 1. However, the spectrum generating device according to this embodiment does not necessarily need to be incorporated into the magnetic resonance imaging apparatus 1, as long as it can generate an artificial spectrum that simulates a spectrum that can be collected by the magnetic resonance imaging apparatus 1 from morphological correlation parameter values. Below, a spectrum generating device according to Modification 1 will be described. In the following description, components that have substantially the same functions as those in the above embodiment will be assigned the same reference numerals and will be described repeatedly only when necessary.

[0103] Fig. 17 is a diagram showing an example of the configuration of a spectrum generating device 50 according to Modification 1. The spectrum generating device 50 shown in Fig. 17 is a computer separate from the magnetic resonance imaging apparatus 1 shown in Fig. 1. As shown in Fig. 17, an acquisition function 512, an estimation function 513, an artificial spectrum generating function 514, a training function 515, and a display control function 516 are realized by executing various programs.

[0104] The processing circuitry 51 acquires, via the acquisition function 512, parameter values ​​of one or more morphological correlation parameters correlated with the morphology within the volume of interest for each of multiple voxels included in the volume of interest. The morphological correlation parameters include morphological images representing the morphology within the volume of interest and / or labels of anatomical sites corresponding to the morphology within the volume of interest. The morphological images include acquired images acquired by a medical image diagnostic device or an optical camera, segmentation images obtained by performing segmentation processing on the acquired images, and / or quantitative value maps generated based on the acquired images. The acquired images are not limited to MRI images acquired by a magnetic resonance imaging device, as long as they represent the morphology within the volume of interest. They may also be X-ray CT images acquired by an X-ray computed tomography device, ultrasound images acquired by an ultrasound diagnostic device, or nuclear medicine images acquired by a nuclear medicine diagnostic device.

[0105] The processing circuitry 51 uses the estimation function 513 to estimate a plurality of parameter sets corresponding to a plurality of voxels based on the parameter values ​​of the morphological correlation parameters acquired by the acquisition function 512. Each of the plurality of parameter sets includes parameter values ​​of one or more types of spectrum generation parameters.

[0106] The processing circuitry 51 applies the parameter sets estimated by the estimation function 513 to the basis spectrum using the artificial spectrum generation function 514 to generate a plurality of first artificial spectra corresponding to a plurality of voxels included in the volume of interest. The processing circuitry 51 then generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra. The second artificial spectrum can be used as a training sample for any machine learning. According to this embodiment, there is no need to actually perform spectrum collection on a subject, so it is possible to generate a large number of second artificial spectra as training samples inexpensively. Note that, in the first modification, as in the above embodiment, a second artificial spectrum may be generated to confirm a spectrum expected to be collected prior to spectrum collection by the magnetic resonance imaging apparatus.

[0107] Training function 515 causes processing circuitry 51 to train an untrained neural network based on a number of training samples to generate a parameter estimation model for use in estimation function 513 .

[0108] The display control function 516 causes the processing circuitry 51 to display various information on the display 55. For example, the processing circuitry 51 displays the second artificial spectrum generated by the artificial spectrum generation function 514 on the display 55.

[0109] (Variation 2) In some of the above embodiments, the spectrum is assumed to be an MRS spectrum. However, this embodiment is not limited to this. A difference spectrum acquired by edited MRS imaging is used as the spectrum according to Modification 2. Edited MRS imaging is a spectrum acquisition method in which two pulse sequences using frequency-selective pulses with different frequency specifications are executed and a difference spectrum between the two obtained spectra is acquired. Known pulse sequences for edited MRS imaging include MEGA-PRESS. Similarly to this embodiment, the first artificial spectrum and the second artificial spectrum of the difference spectrum can also be generated and displayed according to the processing example shown in FIG. 2 . Therefore, according to Modification 2, it is possible to generate an artificial difference spectrum with high accuracy.

[0110] (Variation 3) In some of the above embodiments, the spectrum is assumed to be an MRS spectrum. However, this embodiment is not limited to this. A CEST (chemical exchange saturation transfer) spectrum acquired by CEST imaging is used as the spectrum according to Modification 3. Similarly to this embodiment, the first artificial spectrum and the second artificial spectrum of the CEST spectrum can also be generated and displayed according to the processing example shown in FIG. 2 . Therefore, according to Modification 3, it is possible to generate an artificial CEST spectrum with high accuracy.

[0111] According to some of the above-described embodiments, the spectrum generating device 50 of this embodiment includes a processing circuit 51. The processing circuit 51 acquires, for each of a plurality of voxels included in the volume of interest, parameter values ​​of one or more morphology-correlated parameters correlated with the morphology within the volume of interest. The processing circuit 51 estimates, for each of the plurality of voxels, parameter values ​​of one or more spectrum generating parameters based on the parameter values ​​of the morphology-correlated parameters. The processing circuit 51 applies the parameter values ​​of the one or more spectrum generating parameters for each of the plurality of voxels to a basis spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra.

[0112] According to the above configuration, the processing circuitry 51 generates the second artificial spectrum corresponding to the volume of interest by approximately reproducing the process of generating a measured spectrum by MRS imaging, in which a measured spectrum is generated by performing frequency analysis on an MR signal obtained by summing up MR signal components emitted from various metabolites contained in the volume of interest. In this way, the processing circuitry 51 can generate an average spectrum within the volume of interest as the second artificial spectrum, and therefore can generate an artificial spectrum closer to a spectrum that is actually collected, compared to a comparative example in which an artificial spectrum is generated by a complex simulation for each voxel.

[0113] According to at least one of the embodiments described above, the accuracy of the artificial spectrum can be improved.

[0114] 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 function by reading and executing a program stored in a memory circuit. Note that instead of storing a program in a memory circuit, the program may be directly embedded in the processor circuit. In this case, the processor realizes its function by reading and executing the program embedded in the circuit. On the other hand, if the processor is, for example, an ASIC, the function is directly embedded in the processor circuit as a logic circuit instead of storing the program in a memory circuit. Note that each processor in this embodiment is not limited to being configured as a single circuit for each processor, but may be configured as a single processor by combining multiple independent circuits to realize its function. Furthermore, multiple components in FIGS. 1 and 17 may be integrated into a single processor to realize its function.

[0115] 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.

[0116] With respect to the above embodiment, the following supplementary notes are disclosed as one aspect and optional features of the invention. (Appendix 1) an acquisition unit that acquires, for each of a plurality of voxels included in a volume of interest, parameter values ​​of one or more types of morphology-correlated parameters that correlate with a morphology within the volume of interest; an estimation unit that estimates parameter values ​​of one or more spectrum generation parameters for each of the plurality of voxels based on the parameter values ​​of the morphological correlation parameters; a generation unit that applies parameter values ​​of the one or more types of spectrum generation parameters for each of the plurality of voxels to a base spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A spectrum generating device comprising: (Appendix 2) A spectrum generating device as described in Appendix 1, wherein the morphological correlation parameters include a morphological image representing a morphology within the volume of interest and / or a label of an anatomical location corresponding to a morphology within the volume of interest. (Appendix 3) The spectrum generating device of Appendix 2, wherein the morphological images include acquired images acquired by a medical image diagnostic device or an optical camera, segmentation images obtained by subjecting the acquired images to segmentation processing, and / or quantitative value maps generated based on the acquired images. (Appendix 4) 4. The spectrum generating device according to claim 3, wherein the collected images include MRI images collected by actual imaging with a magnetic resonance imaging device and / or shimming data collected by calibration imaging with a magnetic resonance imaging device. (Appendix 5) A spectrum generating device as described in Appendix 3, wherein the morphological image is an image collected at a spatial resolution corresponding to the plurality of voxels that is higher than the spatial resolution corresponding to the volume of interest. (Appendix 6) The acquisition unit further acquires a measured spectrum relating to the volume of interest collected by a magnetic resonance imaging apparatus, the estimation unit estimates parameter values ​​of the one or more spectrum generation parameters based on the parameter values ​​of the morphological correlation parameters and the measured spectrum; 2. The spectral generating device of claim 1. (Appendix 7) 7. The spectrum generating device according to claim 6, wherein the measured spectrum includes an MRS spectrum, a difference spectrum, and / or a CEST spectrum. (Appendix 8) 2. The spectrum generating device according to claim 1, wherein the estimation unit estimates parameter values ​​of the one or more spectrum generation parameters by applying the parameter values ​​of the morphological correlation parameters to a trained model or a random number generator. (Appendix 9) the generation unit applies parameter values ​​of the spectrum generation parameters for each of the plurality of voxels to a spectral signal model to generate the plurality of first artificial spectra; the spectral signal model is a mathematical model representing the basis spectrum with the spectral generation parameters; 2. The spectral generating device of claim 1. (Appendix 10) 10. The spectrum generating device of claim 9, wherein the spectrum generating parameters include a baseline for the base spectrum, a phase shift for the base spectrum, a concentration of a metabolite represented by the base spectrum, a half-width of the base spectrum, and / or a frequency shift for the base spectrum. (Appendix 11) 2. The spectrum generating device according to claim 1, wherein the generating unit generates the second artificial spectrum for each of a plurality of metabolites included in the volume of interest. (Appendix 12) 12. The spectrum generating device according to claim 11, wherein the generating unit generates a sum spectrum by adding the second artificial spectrum for each of the plurality of metabolites to the second artificial spectrum for each of the plurality of metabolites. (Appendix 13) a display control unit that displays a morphological image relating to the volume of interest and the second artificial spectrum for each of the plurality of metabolites; the display control unit superimposes a mark representing the volume of interest on the anatomical image, and arranges the second artificial spectrum for each of the plurality of metabolites around the anatomical image. 13. The spectrum generating device of claim 12. (Appendix 14) 2. The spectrum generating device according to claim 1, further comprising a display control unit that displays the second artificial spectrum on a display device. (Appendix 15) 15. The spectrum generating device according to claim 14, wherein the display control unit displays the second artificial spectrum alongside a morphological image relating to the volume of interest. (Appendix 16) the morphological correlation parameters include a shimming map and a morphological image; The generation unit applying the parameter values ​​of the shimming map corresponding to the volume of interest and the parameter values ​​of the one or more spectrum generation parameters to a first trained model to generate a third artificial spectrum for the volume of interest; applying the parameter values ​​of the shimming map and the parameter values ​​of the morphological image for the volume of interest to a second trained model to generate an artificial baseline for the volume of interest; adding the third artificial spectrum and the artificial baseline to generate the second artificial spectrum; 2. The spectral generating device of claim 1. (Appendix 17) an acquisition unit that performs MR imaging on a volume of interest set in a subject and acquires parameter values ​​of one or more types of morphology-correlated parameters that correlate with a morphology within the volume of interest for each of a plurality of voxels included in the volume of interest; an estimation unit that estimates parameter values ​​of one or more spectrum generation parameters for each of the plurality of voxels based on the parameter values ​​of the morphological correlation parameters; a generation unit that applies parameter values ​​of the one or more types of spectrum generation parameters for each of the plurality of voxels to a base spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A magnetic resonance imaging apparatus comprising: (Appendix 18) The computer obtaining, for each of a plurality of voxels included in the volume of interest, parameter values ​​of one or more morphology-correlated parameters that correlate with a morphology within the volume of interest; estimating, for each of the plurality of voxels, parameter values ​​for one or more spectrum generating parameters based on the parameter values ​​of the morphological correlation parameters; applying parameter values ​​of the one or more spectrum generation parameters for each of the plurality of voxels to a basis spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generating a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A spectrum generation method comprising: (Appendix 19) an acquisition unit that acquires a morphological image and a shimming map relating to an imaging region including a volume of interest of a subject; an acquisition unit that acquires concentrations of metabolites that may be included in the volume of interest; a generation unit that applies parameter values ​​of the shimming map for the volume of interest and concentrations of the metabolites that may be included in the volume of interest to a trained model #1 to generate multiple artificial spectra #1 corresponding to multiple voxels included in the volume of interest, generates a second artificial spectrum #2 corresponding to the volume of interest based on the multiple first artificial spectra #1, applies parameter values ​​of the shimming map for the volume of interest and parameter values ​​of the morphological image to a trained model #2 to generate multiple artificial baselines #1 corresponding to the multiple voxels, generates a second artificial baseline #2 corresponding to the volume of interest based on the multiple first artificial baselines #1, and generates an artificial spectrum #3 by adding the artificial spectrum #2 and the artificial baseline #2 together; A spectrum generating device comprising: (Appendix 20) an acquisition unit that acquires a morphological image and a shimming map relating to an imaging region including a volume of interest of a subject; an acquisition unit that acquires concentrations of metabolites that may be included in the volume of interest; For each of a plurality of voxels included in the volume of interest, apply the parameter values ​​of the shimming map and the concentrations of the metabolites that may be included in the volume of interest to a trained model #5 to estimate parameter values ​​of spectrum generation parameters; generating an artificial spectrum #1 for each of the plurality of voxels based on parameter values ​​of the spectrum generation parameters for each of the plurality of voxels, and generating an artificial spectrum #2 for the volume of interest based on the artificial spectrum #1 for each of the plurality of voxels; Applying the parameter values ​​of the shimming map and the parameter values ​​of the morphological image for the volume of interest to a trained model #2 to generate a plurality of artificial baselines #1 corresponding to the plurality of voxels, and generating an artificial baseline #2 for the volume of interest based on the plurality of baselines #1; a generation unit that generates an artificial spectrum #3 by adding the artificial spectrum #2 and the artificial baseline #2; A spectrum generating device comprising: (Appendix 21) 21. The spectrum generating device according to claim 20, wherein the generating unit generates the artificial spectrum #3 by performing a weighted addition of the artificial spectrum #2 and the artificial baseline #2. (Appendix 22) The trained model #1 and the trained model #2 are generated by machine learning #3 that reduces the difference between an estimated artificial spectrum #3 obtained by adding an estimated artificial spectrum #2 based on a plurality of estimated artificial spectra #1 output by a neural network #1 in the middle of training and an estimated artificial baseline #2 based on a plurality of estimated artificial baselines #1 output by a neural network #2 in the middle of training, and a correct spectrum #3. (Appendix 23) The spectrum generating device described in Appendix 22, wherein the trained neural network #1 is generated by machine learning #1 that reduces the difference between an estimated artificial spectrum #2 based on a plurality of estimated artificial spectra #1 output by an untrained neural network #1 and a correct spectrum #2 calculated by a mathematical model. (Appendix 24) The spectrum generating device of Appendix 22, wherein the trained neural network #2 is generated by machine learning #2 that reduces the difference between an estimated artificial baseline #2 based on multiple estimated artificial baselines #1 output by an untrained neural network #2 and a correct baseline #2 obtained by subtracting the correct spectrum #2 from the correct spectrum #3. (Appendix 25) 22. The spectrum generating device of claim 21, wherein a weighting factor α for the artificial spectrum #2 and a weighting factor β for the artificial baseline #2 are trained in machine learning #3. (Appendix 26) an acquisition unit that acquires a morphological image and a shimming map relating to an imaging region including a volume of interest of a subject; an acquisition unit that acquires concentrations of metabolites that may be included in the volume of interest; a generation unit that applies a trained model #0 to parameter values ​​of the shimming map related to the volume of interest, parameter values ​​of the morphological image related to the volume of interest, and concentrations of the metabolites that may be included in the volume of interest, to generate an artificial spectrum #2 related to the volume of interest; A spectrum generating device comprising: [Explanation of symbols]

[0117] 1. Magnetic resonance imaging device 11 Mounting stand 13 berths 21 Gradient magnetic field power supply 23 Transmitting circuit 25 Receiving circuit 27 Bed drive unit 29 Sequence control circuit 41 Static magnetic field magnet 43 Gradient magnetic field coil 45 Transmitting coil 47 receiving coil 50 Spectrum Generator (Host Computer) 51 Processing circuit 53 Memory 55 Display 57 Input Interface 59 Communication Interface 131 Top plate 133 Foundation 511 Data Collection Control Function 512 Acquisition Function 513 Estimation Function 514 Artificial Spectrum Generation Function 515 Training Function 516 Display Control Function

Claims

1. an acquisition unit that acquires, for each of a plurality of voxels included in a volume of interest, parameter values ​​of one or more types of morphology-correlated parameters that correlate with a morphology within the volume of interest; an estimation unit that estimates, for each of the plurality of voxels, parameter values ​​of one or more spectrum generation parameters based on the parameter values ​​of the morphological correlation parameters; a generation unit that applies parameter values ​​of the one or more types of spectrum generation parameters for each of the plurality of voxels to a base spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A spectrum generating device comprising:

2. The spectrum generating device according to claim 1 , wherein the morphological correlation parameters include a morphological image representing a morphology within the volume of interest and / or a label of an anatomical location corresponding to the morphology within the volume of interest.

3. The spectrum generating device of claim 2, wherein the morphological images include acquired images acquired by a medical image diagnostic device or an optical camera, segmentation images obtained by subjecting the acquired images to segmentation processing, and / or quantitative value maps generated based on the acquired images.

4. 4. The spectrum generating device according to claim 3, wherein the acquired images include MRI images acquired by actual imaging with a magnetic resonance imaging device and / or shimming data acquired by calibration imaging with the magnetic resonance imaging device.

5. The spectrum generating device according to claim 3 , wherein the morphological image is an image acquired at a spatial resolution corresponding to the plurality of voxels that is higher than a spatial resolution corresponding to the volume of interest.

6. The acquisition unit further acquires a measured spectrum relating to the volume of interest collected by a magnetic resonance imaging apparatus, the estimation unit estimates parameter values ​​of the one or more spectrum generation parameters based on the parameter value of the morphological correlation parameter and the measured spectrum; 2. The spectrum generating device according to claim 1.

7. The spectrum generating device according to claim 6 , wherein the measured spectrum includes an MRS spectrum, a difference spectrum, and / or a CEST spectrum.

8. The spectrum generating device according to claim 1 , wherein the estimation unit estimates the parameter values ​​of the one or more spectrum generating parameters by applying the parameter values ​​of the morphological correlation parameters to a trained model or a random number generator.

9. the generation unit applies parameter values ​​of the spectrum generation parameters for each of the plurality of voxels to a spectral signal model to generate the plurality of first artificial spectra; the spectral signal model is a mathematical model representing the basis spectrum with the spectral generation parameters; 2. The spectrum generating device according to claim 1.

10. The spectrum generating device of claim 9, wherein the spectrum generating parameters include a baseline for the basis spectrum, a phase shift for the basis spectrum, a concentration of a metabolite represented by the basis spectrum, a half-width of the basis spectrum, and / or a frequency shift for the basis spectrum.

11. The spectrum generating device according to claim 1 , wherein the generating unit generates the second artificial spectrum for each of a plurality of metabolites included in the volume of interest.

12. The spectrum generating device according to claim 11 , wherein the generating unit generates a sum spectrum by adding the second artificial spectrum for each of the plurality of metabolites to the second artificial spectrum for each of the plurality of metabolites.

13. a display control unit that displays a morphological image relating to the volume of interest and the second artificial spectrum for each of the plurality of metabolites; the display control unit superimposes a mark representing the volume of interest on the anatomical image, and arranges the second artificial spectrum for each of the plurality of metabolites around the anatomical image.

13. The spectrum generating device according to claim 12.

14. The spectrum generating device according to claim 1 , further comprising a display control unit that displays the second artificial spectrum on a display device.

15. The spectrum generating device according to claim 14 , wherein the display control unit displays the second artificial spectrum alongside a morphological image relating to the volume of interest.

16. the morphological correlation parameters include a shimming map and a morphological image; The generation unit applying the parameter values ​​of the shimming map corresponding to the volume of interest and the parameter values ​​of the one or more spectrum generation parameters to a first trained model to generate a third artificial spectrum for the volume of interest; applying the parameter values ​​of the shimming map and the parameter values ​​of the morphological image for the volume of interest to a second trained model to generate an artificial baseline for the volume of interest; adding the third artificial spectrum and the artificial baseline to generate the second artificial spectrum; 2. The spectrum generating device according to claim 1.

17. an acquisition unit that performs MR imaging on a volume of interest set in a subject and acquires parameter values ​​of one or more types of morphology-correlated parameters that correlate with a morphology within the volume of interest for each of a plurality of voxels included in the volume of interest; an estimation unit that estimates, for each of the plurality of voxels, parameter values ​​of one or more spectrum generation parameters based on the parameter values ​​of the morphological correlation parameters; a generation unit that applies parameter values ​​of the one or more types of spectrum generation parameters for each of the plurality of voxels to a base spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generates a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A magnetic resonance imaging apparatus comprising:

18. The computer obtaining, for each of a plurality of voxels included in the volume of interest, parameter values ​​of one or more morphology-correlated parameters that correlate with a morphology within the volume of interest; estimating, for each of the plurality of voxels, parameter values ​​of one or more spectrum generating parameters based on the parameter values ​​of the morphological correlation parameters; applying parameter values ​​of the one or more spectrum generation parameters for each of the plurality of voxels to a basis spectrum to generate a plurality of first artificial spectra corresponding to the plurality of voxels, and generating a second artificial spectrum corresponding to the volume of interest based on the plurality of first artificial spectra; A spectrum generation method comprising:

Images