Reconstruction device, method, and magnetic resonance imaging device

Abstract

To improve image quality.SOLUTION: A reconstruction device includes a reconstruction unit, a generation unit, an execution unit, and a synthesis unit. The reconstruction unit performs image reconstruction for k space data collected by under-sampling in each of a plurality of coils, and generates one or more sparse reconstruction images. The generation unit generates a first sensitivity map corresponding to each of the plurality of coils using a first learned model. The execution unit executes data consistency processing for improving the coincidence of data on the one or more sparse reconstruction images using the first sensitivity map. The synthesis unit executes coil synthesis processing using a second sensitivity map on the one or more sparse reconstruction images subjected to the data consistency processing, and generates a full reconstruction image.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The embodiments disclosed in this specification and the drawings relate to a reconstruction apparatus, method, and magnetic resonance imaging apparatus.

Background Art

[0002] As an image reconstruction method for generating MR (Magnetic Resonance) images, a method of combining regularization processing by a machine learning model, data consistency processing, and coil synthesis processing is known. Usually, the same sensitivity map is used when performing data consistency processing and coil synthesis processing. Also, as a positioning process for obtaining a sensitivity map, there is a method of estimating a sensitivity map using a machine learning model.

[0003] When the above-described methods are combined, the processing result of the coil synthesis processing greatly affects the training, and the training of the model proceeds so as to be similar to the sensitivity map used as the correct data. Therefore, there is a problem that the training of the machine learning model for estimating the sensitivity map is not effectively performed, and the image quality of the reconstructed image does not become as high as expected.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] One of the problems to be solved by the embodiments disclosed in this specification and the drawings is to improve the image quality. However, the problems to be solved by the embodiments disclosed in this specification and the drawings are not limited to the above problems. It is also possible to position the problems corresponding to the respective effects of the respective configurations shown in the embodiments described later as other problems.

Means for Solving the Problems

[0006] The reconstruction device according to the embodiment includes a reconstruction unit, a generation unit, an execution unit, and a synthesis unit. The reconstruction unit reconstructs k-space data collected by undersampling in each of a plurality of coils, and generates a plurality of sparse reconstruction images. The generation unit generates a first sensitivity map corresponding to each of the plurality of coils using a first trained model. The execution unit executes data consistency processing for improving the degree of coincidence of data related to the sparse reconstruction image using the first sensitivity map. The synthesis unit executes coil synthesis processing using a second sensitivity map for each sparse reconstruction image on which the data consistency processing has been executed, and generates a full reconstruction image.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Embodiments for Carrying Out the Invention

[0008] Hereinafter, embodiments of a reconstruction apparatus, method, program, and magnetic resonance imaging apparatus will be described in detail with reference to the drawings. In the following embodiments, parts denoted by the same reference numerals perform the same operations, and overlapping descriptions will be omitted as appropriate. Hereinafter, one embodiment will be described with reference to the drawings.

[0009] (First Embodiment) The reconstruction apparatus according to the first embodiment will be described with reference to the block diagram of FIG. 1. The reconstruction apparatus 1 shown in FIG. 1 is a computer having a processing circuit 10, a memory 11, an input interface 12, a communication interface 13, and a display 14.

[0010] The processing circuit 10 has a processor such as a CPU as a hardware resource. For example, the processing circuit 10 realizes an acquisition function 101, a reconstruction function 102, a map generation function 103, a regularization function 104, a DC (data consistency) execution function 105, a determination function 106, and a synthesis function 107 by executing various programs.

[0011] By the acquisition function 101, the processing circuit 10 acquires k-space data collected by undersampling with each of a plurality of coils of the magnetic resonance imaging apparatus. The k-space data collected by undersampling is not limited to k-space data collected by regularly decimating like the parallel imaging method, but may be k-space data randomly sampled like compressed sensing, or k-space data collected by the half-Fourier method, or k-space data collected only for low-frequency components without collecting high-frequency parts for super-resolution.

[0012] The processing circuit 10, by means of the reconstruction function 102, reconstructs the acquired k-space data to generate one or more sparse reconstruction images. The sparse reconstruction images are reconstruction images based on the undersampled k-space data corresponding to each coil. Since the number of samples is insufficient due to undersampling, the sparse reconstruction images are images with aliasing.

[0013] The processing circuit 10, by means of the map generation function 103, generates first sensitivity maps respectively corresponding to a plurality of coils using a first pre-trained model.

[0014] The processing circuit 10, by means of the regularization function 104, performs regularization processing on one or more sparse reconstruction images using a second pre-trained model. The regularization processing is processing for removing the aliasing of the sparse reconstruction images.

[0015] The processing circuit 10, by means of the DC execution function 105, performs data consistency processing using the first sensitivity map. The data consistency processing is processing for improving the degree of consistency of data regarding one or more sparse reconstruction images.

[0016] The processing circuit 10, by means of the determination function 106, determines whether the regularization processing and the data consistency processing have been repeatedly executed the required number of times.

[0017] The processing circuit 10, by means of the synthesis function 107, performs coil synthesis processing using the second sensitivity map on one or more sparse reconstruction images on which the data consistency processing has been performed. As a result of the coil synthesis processing, a full reconstruction image, which is an image equivalent to the reconstruction image generated from the fully sampled k-space data, is generated.

[0018] The memory 11 is a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an integrated circuit memory device that stores various information. Further, the memory 11 may be a driving device that reads and writes various information to and from portable storage media such as a CD-ROM drive, a DVD drive, or a flash memory. For example, the memory 11 stores medical data including reconstructed images collected in the past, control programs, and learned models.

[0019] The input interface 12 includes input devices that receive various commands from the user. As input devices, a keyboard, a mouse, various switches, a touch screen, a touch pad, etc. can be used. Note that the input device is not limited to those having physical operation parts such as a mouse and a keyboard. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the reconstruction device and outputs the received electrical signal to various circuits is also included in the example of the input interface 12. Further, the input interface 12 may be a voice recognition device that converts a voice signal collected by a microphone into an instruction signal.

[0020] The communication interface 13 is an interface that connects the reconstruction device to a workstation, a PACS (Picture Archiving and Communication System), a HIS (Hospital Information System), a RIS (Radiology Information System), etc. via a LAN (Local Area Network) or the like. The communication interface 13 transmits and receives various information to and from the connected workstation, PACS, HIS, and RIS.

[0021] The display 14 displays various information. As the display 14, 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. Note that although FIG. 1 shows a configuration in which the reconstruction device 1 includes the display 14, the display may use a display device such as an external monitor, and the reconstruction device 1 may not include the display 14.

[0022] Next, the reconstruction process (inference process) of the reconstruction device 1 according to the first embodiment will be described with reference to the flowchart of FIG. 2.

[0023] In step SA1, the acquisition function 101 causes the processing circuit 10 to acquire k-space data collected by undersampling with each of a plurality of coils of the magnetic resonance imaging device. For example, k-space data based on the subsampled MR signals obtained by the parallel imaging method may be acquired.

[0024] In step SA2, the reconstruction function 102 causes the processing circuit 10 to perform image reconstruction on the k-space data for each coil and generate one or more sparse reconstruction images corresponding to each coil. In step SA3, the map generation function 103 causes the processing circuit 10 to generate a first sensitivity map corresponding to each of the plurality of coils using the first learned model.

[0025] In step SA4, the regularization function 104 causes the processing circuit 10 to perform regularization processing on one or more sparse reconstruction images using the second learned model. In step SA5, the DC execution function 105 causes the processing circuit 10 to perform data consistency processing on the one or more sparse reconstruction images that have been regularized using the first sensitivity map. The data consistency processing may be, for example, processing that reduces the error between the input data, which is the k-space data, and the regularized sparse reconstruction image obtained in step SA4. For example, optimization may be performed such that the mean square error between the k-space data obtained by Fourier-transforming the regularized sparse reconstruction image and the k-space data in step SA1 is minimized.

[0026] In step SA6, the processing circuit 10 determines, by the determination function 106, whether the regularization process and the data consistency process are repeatedly executed as a set of processes the required number of times. The required number of times may be determined, for example, by setting the number of repetitions in advance and determining that the set of processes has been repeatedly executed the required number of times when the number of repetitions of the set of processes reaches a predetermined number of times. Alternatively, when the mean squared error in the data consistency process is equal to or less than a threshold value, it may be determined that the set of processes has been repeatedly executed the required number of times. When the regularization process and the data consistency process have been repeatedly executed the required number of times, the process proceeds to step SA7. When they have not been repeatedly executed the required number of times, the process returns to step SA4 and repeats the processes of steps SA4 and SA5.

[0027] In step SA7, the processing circuit 10, by the synthesis function 107, executes coil synthesis processing using the second sensitivity map for one or more sparse reconstruction images on which the data consistency process has been executed, and generates a full reconstruction image. The second sensitivity map may be a sensitivity map (e.g., ESPIRiT) obtained using the k-space data acquired in step SA1, or a sensitivity map of a coil acquired during a prescan such as a positioning scan, or a sensitivity map based on k-space data related to this scan collected at a different timing from the scan in which the k-space data was acquired. Alternatively, the second sensitivity map may be generated using a learned model. For example, the second sensitivity map may be generated by inputting k-space data to a learned model trained to input k-space data and output a sensitivity map.

[0028] Next, a first example of the inference process of the reconstruction apparatus according to the first embodiment will be described with reference to the conceptual diagram of FIG. 3. In FIG. 3, image reconstruction processing 302 is performed on k-space data 301 obtained by undersampling with a plurality of coils, and a plurality of sparse reconstruction images are obtained. On the other hand, for each coil from which the k-space data 301 is obtained, first sensitivity map generation processing 303 is performed using a first learned model, and a first sensitivity map is generated. The first learned model may use, for example, a so-called SME (Sensitivity Map Estimation) module. The processing of the SME module is, for example, masking data other than the data in the central part of the k-space with respect to the k-space data 301, that is, if it is the GRAPPA (Generalized Auto calibrating Partially Parallel Acquisition) method, masking data other than the ACS (Autocalibration Signal) data, performing inverse Fourier transform, applying a convolutional neural network to each image, and executing a process of normalizing the sensitivity map for each coil.

[0029] A second learned model is applied to the sparse reconstruction image after the image reconstruction processing 302, and regularization processing 304 is executed. Thereafter, data consistency processing 305 (also referred to as DC processing 305) is executed on the plurality of sparse reconstruction images after the regularization processing 304. The regularization processing and the DC processing are regarded as one processing set 350, and the processing set 350 is repeatedly executed N times (N is a natural number of 1 or more).

[0030] Thereafter, using the second sensitivity map 306, coil synthesis 307 is executed on the plurality of sparse reconstruction images for which the processing by the processing set 350 has ended, and a full reconstruction image 310 is generated.

[0031] Next, a first example of the inference processing of the reconstruction apparatus according to the first embodiment will be described with reference to the conceptual diagram of FIG. 4. In FIG. 4, it is assumed that there is one sparse reconstruction image after the image reconstruction process, rather than a plurality of sparse reconstruction images. That is, image reconstruction processing 302 is performed on the k-space data 301 acquired by undersampling with a plurality of coils, and one sparse reconstruction image is obtained. For example, in the case of CG-SENSE (Conjugate Gradient - Sensitivity Encoding) type DC processing, since processing for one image is performed, it is only necessary to process one sparse reconstruction image. The difference from FIG. 3 is that after the DC processing 305, for example, the processing circuit 10 performs coil splitting 401 by the synthesis function 107. The coil splitting 401 generates a plurality of sparse reconstruction images from one sparse reconstruction image using a plurality of first sensitivity maps corresponding to a plurality of coils. Then, coil synthesis 307 is performed, and a full reconstruction image 310 is generated in the same manner as in the case of FIG. 3.

[0032] In the following, the case where a plurality of sparse reconstruction images are generated after the image reconstruction process will be described as an example, but the same processing can be performed even when one sparse reconstruction image is generated after the image reconstruction process. That is, as shown in FIG. 4, the processing related to coil splitting may be executed before coil synthesis.

[0033] According to the first embodiment described above, the k-space data collected by undersampling in each of the plurality of coils is image-reconstructed to generate a plurality of sparse reconstruction images. Using the first learned model, a first sensitivity map corresponding to each of the plurality of coils is generated, and using the first sensitivity map, data consistency processing regarding the sparse reconstruction image is executed. For each sparse reconstruction image on which the data consistency processing has been executed, coil synthesis processing using the second sensitivity map is executed to generate a full reconstruction image. Thereby, a sensitivity map specialized for data consistency processing can be utilized, and the image quality of the full reconstruction image, that is, the finally obtained reconstruction image, can be improved.

[0034] (Second Embodiment) In the first embodiment, it is assumed that the machine learning model included in the reconstruction device 1 is a pre-trained learned model. However, in the second embodiment, the reconstruction device 1 is different from the first embodiment in that it has a training function for training the model.

[0035] The reconstruction device 1 according to the second embodiment will be described with reference to the block diagram of FIG. 5. The reconstruction device 1 according to the second embodiment includes a processing circuit 10, a memory 11, an input interface 12, a communication interface 13, and a display 14. In addition to the configuration according to the first embodiment, the processing circuit 10 includes a training function 108. The training function 108 trains the machine learning model used in the regularization function 104 and the machine learning model used in the DC execution function 105 to generate a learned model.

[0036] Next, the training process of the reconstruction device 1 according to the second embodiment will be described with reference to the flowchart of FIG. 6.

[0037] In step SB1, the processing circuit 10 collects reference k-space data, which is full-sampling k-space data serving as correct data, by the acquisition function 101. In step SB2, the processing circuit 10 performs image reconstruction processing on the reference k-space data by the reconstruction function 102 to generate a coil reference reconstruction image, which is a reconstruction image for each coil.

[0038] In step SB3, the processing circuit 10 performs coil synthesis processing on the coil reference reconstruction image using the sensitivity map for each coil generated from the reference k-space data by the synthesis function 107 to generate a reference full reconstruction image. In step SB4, the acquisition function 101 causes the processing circuit 10 to undersample the reference k-space data used in step SB1, and acquire the undersampled k-space data for each coil as the input data of the training data. For example, for the reference k-space data, the undersampled k-space data may be generated by thinning the data so as to sample one column every four columns along the ky direction or the kx direction. When a method other than the parallel imaging method is applied as the undersampling, the reference k-space data may be processed so as to be the data undersampled from the reference k-space data respectively. For example, if the half-Fourier method is applied as the undersampling, the k-space data related to the region of half of the k-space + α may be acquired.

[0039] In step SB5, the reconstruction function 102 causes the processing circuit 10 to perform image reconstruction processing on the undersampled k-space data, and generate a sparse reconstruction image for each coil. In step SB6, the processing circuit 10 uses the first model for generating the sensitivity map to be trained, and generates the first sensitivity map corresponding to each of the plurality of coils by the map generation function 103. As the first model, a general machine learning model such as a convolutional neural network or an autoencoder may be used.

[0040] In step SB7, the processing circuit 10 performs a regularization process on the sparse reconstruction image using the second model for performing the regularization process to be trained by the regularization function 104. Similar to the first model, a general machine learning model may be used for the second model. In step SB8, the processing circuit 10 performs a data consistency process on the regularized sparse reconstruction image using the first sensitivity map generated in step SB6 by the DC execution function 105.

[0041] In step SB9, the processing circuit 10 determines, by the determination function 106, whether the regularization process and the data consistency process are repeated the required number of times as a set. The repetition determination may use the same method as in step SA6. If the required number of repetitions is completed, the process proceeds to step SB10. If the required number of repetitions is not completed, the process returns to step SB7 and the same process is repeated. In step SB10, the processing circuit 10 performs coil synthesis processing using the sensitivity map for each sparse reconstruction image on which the data consistency process has been performed by the synthesis function 107, and generates an estimated full reconstruction image. The sensitivity map used in step SB3 may be used.

[0042] In step SB11, the processing circuit 10 calculates the loss value between the reference full reconstruction image and the estimated full reconstruction image using the loss function by the training function 108. As the loss function, a general loss function used in machine learning such as MSE (Mean Squared Error) or binary cross entropy may be used.

[0043] In step SB12, the processing circuit 10 determines, by the training function 108, whether the training is completed based on the loss value calculated in step SB10. Specifically, for example, if the loss value is less than or equal to the threshold value and the loss value has converged, it may be determined that the training is completed. Alternatively, it may be determined that the training is completed when the training is repeated a predetermined number of epochs. That is, a general training end determination in machine learning may be used. If the training is not completed, the process proceeds to step SB13. On the other hand, if the training is completed, the first trained model and the second trained model are determined and the process ends.

[0044] In step SB13, the parameters (weights and biases) of each of the first model and the second model are updated, the process returns to step SB1, and the same process is repeated for the new reference space k data.

[0045] In the example of FIG. 6, an example is shown in which undersampled k-space data is acquired after generating a reference full reconstruction image. However, the present invention is not limited to this, and the generation processes of the reference full reconstruction image and the estimated full reconstruction image may be performed in parallel. That is, the processes according to steps SB2-SB3 and the processes according to steps SB4-SB10 may be executed in parallel.

[0046] Next, a conceptual diagram of the training process of the reconstruction apparatus according to the second embodiment is shown in FIG. 7. From the reference k-space data 601, a second sensitivity map 602 for each coil is generated. Here, four reference k-space data 601 are used to generate the second sensitivity maps 602, respectively. By performing an image reconstruction process 603 on the reference k-space data 601 and a coil synthesis process 604 using the second sensitivity map 602, a reference full reconstruction image 605 serving as correct data is generated.

[0047] On the other hand, the reference k-space data 601 is subsampled or masked along columns in the kx direction or the ky direction to generate undersampled k-space data 610. A first sensitivity map generation process 611 is performed on the k-space data 610. Specifically, the k-space data 610 is input to the first model, and a first sensitivity map is output. Also, the k-space data 610 is subjected to an image reconstruction process 603 to generate a sparse reconstruction image. The sparse reconstruction image is input to the second model to execute a regularization process 612, and then a DC process 613 is executed. A process set 614 including the regularization process 612 by the second model and the DC process 613 is repeatedly executed (N times). Thereafter, a coil synthesis process 604 is performed using the second sensitivity map 602 to generate an estimated full reconstruction image 615.

[0048] The reference full reconstruction image 605 and the estimated full reconstruction image 615 are compared, and the training status is determined based on the loss value calculated from the loss function. The parameters of the first model and the second model are updated until it is determined that the training is completed.

[0049] Next, FIG. 8 shows a conceptual diagram of another example of the training process of the reconstruction apparatus 1 according to the second embodiment. In the example of FIG. 7, the second sensitivity map 602 is shown as being generated from the reference k-space data 601 or the undersampled k-space data 610. In FIG. 8, an example using the second sensitivity map 701 obtained from a different scan is shown. The second sensitivity map 701 may be generated, for example, from a reference scan in a parallel imaging method, or may be generated using a neural network. As long as a sensitivity map corresponding to the k-space data 610 can be generated, it may be generated using any method.

[0050] Note that the second sensitivity map does not have to be the same sensitivity map during the training of the first model 611 and the second model 612 and during inference as shown in the first embodiment, and different sensitivity maps may be used. For example, during the training of the first model 611 and the second model 612 according to the second embodiment, the second sensitivity map 602 (for example, ESPIRiT) may be used, and during inference according to the first embodiment, the second sensitivity map 701 (for example, a sensitivity map generated from a reference scan in a parallel imaging method) may be used.

[0051] According to the second embodiment described above, the first model for estimating the sensitivity map and the second model for performing the regularization process are trained simultaneously. In the coil synthesis process, the first model and the second model are trained using a second sensitivity map different from the first sensitivity map output from the first model. As a result, compared to the case where the first sensitivity map is also used in the coil synthesis process, training can proceed without making an estimation close to the sensitivity map of the correct data used as a reference in the training of the first model, and a learned model with improved image quality of the reconstructed image can be generated.

[0052] (Third Embodiment) In the third embodiment, it is different from the above-described embodiments in that it is applied to the reconstruction process in multi-echo (multi-contrast) imaging, such as a water-fat separation technique such as the DIXON method.

[0053] A conceptual diagram of the inference process of the reconstruction device 1 according to the third embodiment is shown in FIG. 9. FIG. 9 shows the processing of the reconstruction device 1 for the first echo 801 and the subsequent processing of the reconstruction device 1 for the second echo 802. Here, the second sensitivity map 306 used for the coil synthesis process 307 uses a common sensitivity map for the first echo 801 and the second echo 802.

[0054] Next, a conceptual diagram of another example of the inference process of the reconstruction device 1 according to the third embodiment is shown in FIG. 10. FIG. 10 uses the first sensitivity map, which is the output from the first learned model obtained in another echo, as the second sensitivity map in the echo to be processed. Specifically, in the example of FIG. 10, the first sensitivity map output from the first learned model in the first sensitivity map generation process 303 for the first echo 801 may be used as the second sensitivity map in the processing for the second echo 802 in the coil synthesis process 307.

[0055] Note that it is not limited to using the previous first sensitivity map in time series. The first sensitivity map obtained in the processing for the second echo 802 may also be used as the second sensitivity map in the processing for the first echo 801 in the coil synthesis process 307.

[0056] Also, here, the two echoes of the first echo 801 and the second echo 802 have been described. Similarly, in the case of three or more echoes, the first sensitivity map in another echo can be applied as the second sensitivity map of the echo to be processed.

[0057] According to the third embodiment shown above, in multi-echo imaging, a common second sensitivity map is used. Or, the first sensitivity map generated in the processing related to another echo is used as the second sensitivity map in the echo to be processed. Thereby, the image quality of each full reconstruction image generated in multi-echo imaging can be improved.

[0058] (Fourth Embodiment) In the fourth embodiment, a magnetic resonance imaging apparatus equipped with a reconstruction device is assumed. FIG. 11 is a block diagram showing a configuration example of the magnetic resonance imaging apparatus 2 according to the present embodiment. As shown in FIG. 11, the magnetic resonance imaging apparatus 2 includes a gantry 40, a bed 90, a gradient magnetic field power supply 21, a transmission circuit 23, a reception circuit 25, a bed drive device 27, a sequence control circuit 29, and a host computer 50.

[0059] The gantry 40 includes 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 the housing of the gantry 40. A bore having a hollow shape is formed in the housing of the gantry 40. A transmission coil 45 and a reception coil 47 are disposed in the bore of the gantry 40.

[0060] The static magnetic field magnet 41 has a substantially hollow cylindrical shape and generates a static magnetic field inside the substantially cylinder. As the static magnetic field magnet 41, for example, a permanent magnet, a superconducting magnet, or a normal conducting magnet is used. Here, the central axis of the static magnetic field magnet 41 is defined as the Z axis, the axis perpendicular to the Z axis vertically is defined as the Y axis, and the axis perpendicular to the Z axis horizontally is defined as the X axis. The X axis, the Y axis, and the Z axis constitute an orthogonal three-dimensional coordinate system.

[0061] The gradient magnetic field coil 43 is attached inside the static magnetic field magnet 41 and is a coil unit formed in a hollow substantially cylindrical shape. The gradient magnetic field coil 43 generates a gradient magnetic field upon receiving the supply of current from the gradient magnetic field power supply 21. More specifically, the gradient magnetic field coil 43 has three coils corresponding to the X-axis, Y-axis, and Z-axis that are orthogonal to each other. The three coils form a gradient magnetic field in which the magnetic 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 synthesized to form a frequency-encoding gradient magnetic field Gr, a phase-encoding gradient magnetic field Gp, and a slice-selection gradient magnetic field Gs that are orthogonal to each other in a desired direction. The frequency-encoding gradient magnetic field Gr is used to change the frequency of the magnetic resonance signal (MR signal) according to the spatial position. The phase-encoding gradient magnetic field Gp is used to change the phase of the MR signal according to the spatial position. The slice-selection gradient magnetic field Gs is used to arbitrarily determine the imaging section (slice). In the following description, it is assumed that the gradient direction of the frequency-encoding gradient magnetic field Gr is the X-axis, the gradient direction of the phase-encoding gradient magnetic field Gp is the Y-axis, and the gradient direction of the slice-selection gradient magnetic field Gs is the Z-axis.

[0062] The gradient magnetic field power supply 21 supplies current to the gradient magnetic field coil 43 in accordance with the sequence control signal from the sequence control circuit 29. By supplying current to the gradient magnetic field coil 43, the gradient magnetic field power supply 21 causes the gradient magnetic field coil 43 to generate a gradient magnetic field along each of the X-axis, Y-axis, and Z-axis. The gradient magnetic field is superimposed on the static magnetic field formed by the static magnetic field magnet 41 and applied to the subject P.

[0063] The transmission coil 45 is disposed, for example, inside the gradient magnetic field coil 43 and generates a high-frequency pulse (hereinafter referred to as an RF pulse) upon receiving the supply of current from the transmission circuit 23.

[0064] The transmission circuit 23 supplies a current to the transmission coil 45 in order to apply an RF pulse for exciting target protons existing in the subject P to the subject P via the transmission coil 45. The RF pulse oscillates at the resonance frequency peculiar to the target protons and excites 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 transmit-receive coil.

[0065] The reception coil 47 receives an MR signal emitted from target protons existing in the subject P under the action of the RF pulse. The reception coil 47 has a plurality of reception coil elements capable of receiving the MR signal. The received MR signal is supplied to the reception circuit 25 via wired or wireless means. Although not shown in FIG. 11, the reception coil 47 has a plurality of reception channels mounted in parallel. The reception channel has a reception coil element for receiving the MR signal, an amplifier for amplifying the MR signal, and the like. The MR signal is output for each reception channel. The total number of reception channels may be the same as the total number of reception coil elements, or may be larger or smaller than the total number of reception coil elements.

[0066] The reception circuit 25 receives the MR signal generated from the excited target protons via the reception coil 47. The reception circuit 25 processes the received MR signal to generate a digital MR signal. The digital MR signal can be represented in the k-space defined by the spatial frequency. Therefore, hereinafter, the digital MR signal will be referred to as k-space data.

[0067] Note that the above transmission coil 45 and reception coil 47 are merely examples. Instead of the transmission coil 45 and the reception coil 47, a transmit-receive coil having a transmission function and a reception function may be used. Also, the transmission coil 45, the reception coil 47, and the transmit-receive coil may be combined.

[0068] A bed 90 is installed adjacent to the gantry 40. The bed 90 has a top plate 901 and a base 903. The subject P is placed on the top plate 901. The base 903 supports the top plate 901 so as to be slidable along each of the X-axis, Y-axis, and Z-axis. A bed driving device 27 is housed in the base 903. The bed driving device 27 moves the top plate 901 under the control from the sequence control circuit 29. The bed driving device 27 may include any motor such as a servo motor or a stepping motor, for example.

[0069] 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 the data collection conditions, subjects the subject P to data collection according to the data collection conditions, and collects k-space data regarding the subject P. The data collection conditions define, for example, a pulse sequence, the magnitude of the current supplied to the gradient magnetic field coil 43 by the gradient magnetic field power supply 21, the timing at which the current is supplied to the gradient magnetic field coil 43 by the gradient magnetic field power supply 21, the magnitude of the RF pulse supplied to the transmission coil 45 by the transmission circuit 23, the timing at which the RF pulse is supplied to the transmission coil 45 by the transmission circuit 23, the timing at which the MR signal is received by the reception coil 47, and the like.

[0070] As shown in FIG. 11, the host computer 50 is a computer having a processing circuit 10, a memory 53, a display 55, an input interface 57, and a communication interface 59.

[0071] The processing circuit 10 is the same as that in FIG. 1. That is, the processing circuit 10 controls the sequence control circuit 29, the transmission coil 45, the gradient magnetic field coil 43, the reception coil 47, etc., and collects k-space data undersampled in each of the plurality of coils. The memory 53 is a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or an integrated circuit memory device that stores various information. Further, the memory 53 may be a driving device that reads and writes various information to and from portable storage media such as a CD-ROM drive, a DVD drive, or a flash memory. For example, the memory 53 stores k-space data, a sparse reconstruction image, a full reconstruction image, a learned model, a control program, and the like.

[0072] The display 55 displays various information. 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.

[0073] The input interface 57 includes an input device that receives various commands from the user. As the input device, a keyboard, a mouse, various switches, a touch screen, a touch pad, and the like can be used. Note that the input device is not limited to those having physical operation parts such as a mouse and a keyboard. For example, an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the magnetic resonance imaging apparatus 2 and outputs the received electrical signal to various circuits is also included in the example of the input interface 57. Further, the input interface 57 may be a voice recognition device that converts a voice signal collected by a microphone into an instruction signal.

[0074] The communication interface 59 is an interface that connects the magnetic resonance imaging device 2 via a LAN (Local Area Network) or the like to workstations, PACS (Picture Archiving and Communication System), HIS (Hospital Information System), RIS (Radiology Information System), and the like. The communication interface 59 transmits and receives various types of information between the connected workstations, PACS, HIS, and RIS. According to the fourth embodiment described above, similar to the first embodiment, the image quality of the MR image can be improved.

[0075] The term "processor" used in the above description means, for example, a CPU, a GPU, or a circuit such as an application specific integrated circuit (ASIC), a programmable logic device (for example, a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA)). The processor realizes its functions by reading and executing a program stored in a storage circuit. Instead of storing the program in the storage circuit, the program may be directly incorporated into the circuit of the processor. In this case, the processor realizes its functions by reading and executing the program incorporated in the circuit. Also, instead of executing the program, the functions corresponding to the program may be realized by a combination of logic circuits. Each processor of the present embodiment is not limited to being configured as a single circuit for each processor, and may be configured as one processor by combining a plurality of independent circuits to realize its functions. Further, a plurality of components may be integrated into one processor to realize its functions.

[0076] According to at least one embodiment described above, the image quality can be improved.

[0077] Although several embodiments have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, changes, and combinations of embodiments can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, as well as in the invention described in the claims and the equivalent scope thereof.

Description of Reference Numerals

[0078] 1 Reconfiguration device 2 Magnetic resonance imaging device 10 Processing circuit 11, 53 Memory 12, 57 Input interface 13, 59 Communication interface 14, 55 Display 21 Gradient magnetic field power supply 23 Transmission circuit 25 Reception circuit 27 Bed driving device 29 Sequence control circuit 40 Gantry 41 Static magnetic field magnet 43 Gradient magnetic field coil 45 Transmission coil 47 Reception coil 50 Host computer 90 Bed 101 Acquisition function 102 Reconfiguration function 103 Map generation function 104 Regularization function 105 DC execution function 106 Judgment function 107 Synthesis function 108 Training function 301, 610 k-space data 302,603 Image reconstruction processing 303,611 First sensitivity map generation processing 304,612 Regularization processing 305,613 DC processing 306,602,701 Second sensitivity map 307,604 Coil synthesis processing 310 Full reconstruction image 350,614 Processing set 601 Reference k-space data 605 Reference full reconstruction image 615 Estimated full reconstruction image 801 First echo 802 Second echo 901 Ceiling 903 Base

Claims

1. A reconstruction unit that reconstructs k-space data collected by undersampling in each of a plurality of coils to generate one or more sparse reconstruction images; A generation unit that generates a first sensitivity map corresponding to each of the plurality of coils using a first trained model; An execution unit that executes data consistency processing for improving the degree of agreement of data regarding the one or more sparse reconstruction images using the first sensitivity map; A synthesis unit that executes coil synthesis processing using a second sensitivity map for one or more sparse reconstruction images on which the data consistency processing has been executed to generate a full reconstruction image; A reconstruction apparatus comprising the above.

2. The second sensitivity map is generated based on the k-space data, pre-scan data when the k-space data was acquired, or k-space data related to this scan collected at a timing different from the k-space data. The reconstruction apparatus according to Claim 1.

3. The second sensitivity map is generated by inputting the k-space data used by the generation unit into a trained model trained to input k-space data and output a sensitivity map. The reconstruction apparatus according to Claim 1.

4. The k-space data collected by the undersampling is any one of k-space data collected by the half-Fourier method, randomly sampled k-space data, regularly decimated k-space data, and k-space data that has collected low-frequency components. The reconstruction apparatus according to Claim 1.

5. When there is one sparse reconstruction image, the synthesis unit generates a plurality of sparse reconstruction images corresponding to each coil using the first sensitivity map, and executes the coil synthesis processing on the generated plurality of sparse reconstruction images. The reconstruction apparatus according to Claim 1.

6. The synthesis unit uses a second sensitivity map common to the coil synthesis processing in processing of data of a first echo and processing of data of a second echo different from the first echo in multi-echo imaging. The reconstruction apparatus according to Claim 1.

7. The synthesis unit uses the first sensitivity map generated from processing of data of a first echo in multi-echo imaging in the coil synthesis processing for processing of data of a second echo different from the first echo. The reconstruction apparatus according to Claim 1. Claim 8 In each of a plurality of coils, k-space data collected by undersampling is image reconstructed to generate one or more sparse reconstructed images, using a first pre-trained model, a first sensitivity map corresponding to each of the plurality of coils is generated, using the first sensitivity map, data consistency processing is performed to improve the degree of data agreement regarding the one or more sparse reconstructed images, for the one or more sparse reconstructed images on which the data consistency processing has been performed, coil combination processing using a second sensitivity map is performed to generate a full reconstructed image, A reconstruction method. Claim 9 A collection unit that collects, for each coil, k-space data undersampled in each of a plurality of coils, a reconstruction unit that image reconstructs the k-space data to generate one or more sparse reconstructed images, a generation unit that uses a first pre-trained model to generate a first sensitivity map corresponding to each of the plurality of coils, an execution unit that performs data consistency processing to improve the degree of data agreement regarding the one or more sparse reconstructed images using the first sensitivity map, a combination unit that performs coil combination processing using a second sensitivity map for the one or more sparse reconstructed images on which the data consistency processing has been performed to generate a full reconstructed image, A magnetic resonance imaging apparatus comprising the same.

Images