Magnetic resonance imaging system and method, computer readable storage medium

Abstract

The application provides a kind of magnetic resonance imaging system and method, computer readable storage medium.The method includes: the object is executed medical scanning and obtains the first medical image with first noise interference artifact, and the object is executed additional scanning to obtain second medical image, wherein, the second medical image has second noise interference artifact, and the position mapping of the second noise interference artifact in the first medical image is symmetrical relative to the position of the first noise interference artifact with the pixel center of the first medical image;And, the first medical image and second medical image are processed in relation to synthesis, to obtain the post-processing image of reduced noise interference artifact.

Description

Technical Field

[0001] This invention relates to the field of medical imaging, and in particular to a magnetic resonance imaging (MRI) system and method, and a computer-readable storage medium. Background Technology

[0002] A magnetic resonance imaging system can excite the imaging tissue to generate magnetic resonance signals when scanning the object being detected. These magnetic resonance signals are then acquired and reconstructed into magnetic resonance images.

[0003] Sometimes, artifacts of a specific shape appear in magnetic resonance images due to radio frequency interference noise. In the existing technology, it is necessary to first identify the source of the artifact in order to remove the interference in a targeted manner. Such processing methods are relatively complex and time-consuming. Summary of the Invention

[0004] One aspect of the present invention provides a magnetic resonance imaging method, comprising:

[0005] A medical scan is performed on the object to acquire a first medical image with a first noise interference artifact, and an additional scan is performed on the object to acquire a second medical image, wherein the second medical image has a second noise interference artifact, and the position of the second noise interference artifact in the first medical image is mapped to be symmetrical about the position of the first noise interference artifact with respect to the pixel center of the first medical image; and,

[0006] The first and second medical images are subjected to synthesis-related processing to obtain post-processed images with reduced noise artifacts.

[0007] In another aspect, the present invention provides a computer-readable storage medium including a stored computer program, wherein the above-described method is executed when the computer program is run.

[0008] In another aspect, the present invention provides a magnetic resonance imaging system, comprising:

[0009] Scanner;

[0010] A controller for controlling the scanner to perform a magnetic resonance imaging (MRI) scan, the MRI scan comprising: performing a medical scan on an object and acquiring a first medical image, the first medical image having a first noise interference artifact; and performing an additional scan on the object to acquire a second medical image, wherein the second medical image has a second noise interference artifact, the position of the second noise interference artifact in the first medical image being mapped symmetrically with respect to the pixel center of the first medical image to the position of the first noise interference artifact; and...

[0011] An image processor for performing synthesis-related processing on a first medical image and a second medical image to obtain a post-processed image with reduced noise artifacts.

[0012] It should be understood that the brief description provided above is intended to introduce some concepts further described in the detailed embodiments in a simplified form. This is not intended to identify key or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims following the detailed description. Furthermore, the claimed subject matter is not limited to the implementation of any shortcomings mentioned above or in any paragraph of this disclosure. Attached Figure Description

[0013] The invention will be better understood by referring to the accompanying drawings and by reading the following description of non-limiting embodiments, in which:

[0014] Figure 1 A schematic diagram of the structure of a magnetic resonance imaging system according to some embodiments of the present invention is shown;

[0015] Figure 2 A flowchart of a magnetic resonance imaging method according to some embodiments of the present invention is shown;

[0016] Figure 3 An example of a first medical image, a second medical image, and a post-processed image is shown respectively;

[0017] Figure 4 An example of K-space encoding used in medical scanning is shown;

[0018] Figure 5 An example of K-space encoding used in an additional scan is shown;

[0019] Figure 6 A block diagram of a first deep learning network according to an embodiment of the present invention is shown;

[0020] Figure 7 An example is shown for a clinical image without interference noise artifacts, a background noise image, a noise interference image, a first image, a third frequency domain image, a fourth frequency domain image, and a second image, respectively.

[0021] Figure 8 A flowchart of a medical imaging method according to another embodiment of the present invention is shown;

[0022] Figure 9 A block diagram of a magnetic resonance imaging system according to an embodiment of the present invention is shown;

[0023] Figure 10 A block diagram of a magnetic resonance imaging system according to another embodiment of the present invention is shown;

[0024] Figure 11 The images shown are a gold standard image of a body part, an original image with noise artifacts, a post-processed image obtained based on an embodiment of the present invention, and a post-processed image obtained based solely on a first medical image. Detailed Implementation

[0025] The following describes specific embodiments of the present invention. It should be noted that, in order to provide a concise description, this specification cannot exhaustively describe all features of the actual embodiments. It should be understood that, in the actual implementation of any embodiment, just as in any engineering or design project, various specific decisions are often made to achieve the developer's specific goals and to meet system-related or business-related constraints, and this can change from one embodiment to another. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this invention, some design, manufacturing, or production modifications based on the technical content disclosed herein are merely conventional technical means and should not be construed as insufficient content of this disclosure.

[0026] Unless otherwise defined, the technical or scientific terms used in the claims and description shall have the ordinary meaning understood by one of ordinary skill in the art. The terms "first," "second," and similar terms used in this specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0027] Figure 1 A schematic diagram of a magnetic resonance imaging (MRI) system according to some embodiments of the present invention is shown. The MRI system 100 includes a scanner 110. The scanner 110 is used to perform magnetic resonance scanning on an object (e.g., a human body) 16 to generate image data of a region of interest of the object 16, which may be a predetermined anatomical location or anatomical tissue.

[0028] The magnetic resonance imaging system 100 may include a controller 130 coupled to the scanner 110 for controlling the scanner 110 to perform the magnetic resonance scanning process described above.

[0029] In one example, scanner 110 may include main magnet assembly 111, bed 112, radio frequency generator 113, radio frequency power amplifier 119, radio frequency transmitting coil 114, surface coil 118, gradient coil driver 115, gradient coil assembly 116, and data acquisition unit 117.

[0030] The main magnet assembly 111 typically includes a toroidal superconducting magnet defined within a housing, which is mounted within a toroidal vacuum container. The toroidal superconducting magnet and its housing define a cylindrical space surrounding the object 16, such as... Figure 1 The scanning cavity 120 is shown. The main magnet assembly 111 generates a constant magnetic field, or B0 field, along the Z direction of the scanning cavity 120. Typically, a more uniform portion of the B0 field is formed in the central region of the main magnet.

[0031] The bed 112 is used to carry the object 16 and responds to the control of the controller 130 to move along the Z direction to enter and exit the scanning cavity 120. For example, in one embodiment, the imaging volume of the object 16 can be positioned in the central region of the scanning cavity where the magnetic field strength is relatively uniform, so as to facilitate the scanning imaging of the imaging volume of the object 16.

[0032] The Z direction mentioned above is typically the direction that extends from head to foot (or from foot to head) when the object 16 is positioned on the bed 112. For example, the selected layer can be a slice at any location in the Z direction.

[0033] The magnetic resonance imaging system 100 uses the generated B0 field to emit a static magnetic field to the object 16 located in the scanning cavity, thereby ordering the precession of protons in the resonance region within the object 16 and generating a longitudinal magnetization vector.

[0034] The radio frequency (RF) generator 113 is used to generate RF pulses, such as RF excitation pulses, in response to control signals from the controller 130. The RF power amplifier 119 is used to amplify the low-power RF signal generated by the RF generator 113 to generate a high-power RF signal capable of exciting human tissue. This high-power RF signal can be input to the RF transmitting coil 114 via an RF transmission line, causing the RF transmitting coil 114 to emit an RF field B1 orthogonal to the B0 field towards the object 16 to excite the atomic nuclei in the aforementioned resonant region, generating a transverse magnetization vector.

[0035] The radio frequency transmitting coil 114 may include, for example, a body coil disposed along the periphery of the main magnet, or a local coil dedicated to local imaging.

[0036] After the radio frequency excitation pulse ends, the proton group loses phase, and the macroscopic transverse magnetization vector in the tissue gradually decays. During the process of the transverse magnetization vector of object 16 gradually recovering to zero, a free induction decay signal is generated, which is the magnetic resonance signal that can be acquired.

[0037] The gradient coil driver 115 is used to provide appropriate current / power to the gradient coil assembly 116 in response to the gradient pulse control signal or the shimming control signal issued by the controller 130.

[0038] The gradient coil assembly 116 forms a changing magnetic field in the imaging space to provide three-dimensional position information for the aforementioned magnetic resonance signal, and can also be used to generate a compensation magnetic field for the B0 field to homogenize the B0 field.

[0039] The gradient coil assembly 116 may include three sets of gradient coils, which are used to generate magnetic field gradients tilted to three mutually perpendicular spatial axes (e.g., the X-axis, Y-axis, and Z-axis). More specifically, the gradient coil assembly 116 applies a magnetic field gradient in the slice selection direction (e.g., the Z-axis) to change the field strength in the region, causing different precession frequencies of protons in different layers of the imaging tissue in that region, thereby achieving slice selection. Those skilled in the art will understand that the slice can be any one of multiple two-dimensional slices distributed along the Z-axis in the three-dimensional imaging volume. When the imaging region is scanned, the radio frequency transmitting coil 114 responds to the aforementioned radio frequency excitation signal, and the slice having a precession frequency corresponding to the radio frequency excitation signal is excited. The gradient coil assembly 116 applies magnetic field gradients in the phase encoding direction (e.g., the Y-axis) and the frequency encoding direction (e.g., the X-axis), respectively, so that the magnetic resonance signals of the excited slices have different phases and frequencies, achieving phase encoding and frequency encoding.

[0040] The aforementioned radio frequency transmitting coil 114 can be connected to a transmit / receive (T / R) switch. By controlling the transmit / receive switch, the body coil can be switched between transmit and receive modes. In receive mode, the radio frequency transmitting coil can be used to receive magnetic resonance signals from the object 16.

[0041] The surface coil 118 is typically positioned close to the scanning area (region of interest) of the object 16 (e.g., covering or laying on the body surface of the object 16), and the surface coil 118 is also used to receive magnetic resonance signals from the object 16.

[0042] The data acquisition unit 117 is used to acquire the magnetic resonance signal (e.g., received by a body coil or surface coil) in response to the data acquisition control signal of the controller 130. In one embodiment, the data acquisition unit 117 may include, for example, a radio frequency preamplifier, a phase detector, and an analog-to-digital converter, wherein the radio frequency preamplifier is used to amplify the magnetic resonance signal, the phase detector is used to perform phase detection on the amplified magnetic resonance signal, and the analog-to-digital converter is used to convert the phase-detected magnetic resonance signal from an analog signal to a digital signal.

[0043] The data acquisition unit 117 is further configured to respond to the data storage control signal of the controller 130 to store the digitized magnetic resonance signal (or echo) in the K-space, which is the filling space for the original data of the magnetic resonance signal with spatial positioning encoding information. The data acquisition unit 117 fills signals with different phase and frequency information into the corresponding positions in the K-space according to a preset data filling method. In one example, the two-dimensional K-space may include frequency encoding lines Kx and phase encoding lines Ky. Data acquisition at each level may include multiple signal acquisition cycles (or repetition time TR), each signal acquisition cycle corresponding to a change in the magnetic field gradient (incrementally or decrementally) in the phase encoding direction (i.e., a signal acquisition is performed each time a phase encoding gradient is applied). The magnetic resonance signal acquired in each signal acquisition cycle is filled into a frequency encoding line Kx. Through multiple signal acquisition cycles, multiple frequency encoding lines with different phase information can be filled, and each acquired magnetic resonance signal has multiple decomposed frequencies.

[0044] Those skilled in the art will understand that when imaging scans are performed on object 16, controller 130 can send sequence control signals to the aforementioned components of scanner 110 (e.g., radio frequency generator 113, gradient coil driver 115, etc.) via sequence generator (not shown in the figure) so that scanner 110 executes a preset scanning sequence.

[0045] Those skilled in the art will understand that the aforementioned "scanning sequence" refers to a combination of pulses with specific amplitude, width, direction, and timing applied during magnetic resonance imaging (MRI) scanning. These pulses typically include, for example, radio frequency (RF) pulses and gradient pulses. The RF pulses may include, for example, RF transmit pulses, RF refocusing pulses, inversion recovery pulses, etc. The gradient pulses may include, for example, the aforementioned gradient pulses for slice selection, gradient pulses for phase encoding, gradient pulses for frequency encoding, etc. Generally, multiple scanning sequences can be pre-set in the MRI system to allow selection of sequences appropriate to clinical testing needs, which may include, for example, imaging sites, imaging functions, and imaging effects.

[0046] Performing a magnetic resonance imaging (MRI) scan on subject 16 may include a localization scan (three-panel scan) and a formal scan, during which one or more scan sequences may be executed. During the localization scan, at least one coronal, sagittal, and transverse localization image of the subject may be acquired, and the scanning parameters for the formal scan, such as the scanning range, are determined based on this localization image. Typically, a formal scan is performed on the subject to acquire medical images suitable for clinical diagnosis.

[0047] In a single MRI examination of a subject 16, the localization image obtained by the localization scan can be a prior scan image obtained before the formal scan. The controller 130 can also increase / decrease / adjust the scan sequence in response to different workflow settings selected by the user based on different clinical applications.

[0048] Before performing one or more scan sequences for positioning or formal scanning, a pre-scan can be performed. During the pre-scan, system presets can be completed. For example, frequency adjustment can be performed to determine the Larmor frequency of proton resonance (or the center frequency of the MRI scan) based on magnetic resonance signal feedback at different frequencies; radio frequency emission intensity adjustment can be performed to determine the radio frequency emission power based on magnetic resonance signal feedback at different radio frequency emission intensities; and a pre-scan image can be obtained based on the pre-scan of the object, and the homogeneity of the main magnetic field can be analyzed and adjusted based on the pre-scan image to achieve shimming. The pre-scan image can be a prior scan image obtained before performing the aforementioned positioning or formal scanning.

[0049] The magnetic resonance imaging system 100 may include an image reconstructor 140, which performs an inverse Fourier transform on data stored in K-space to reconstruct a three-dimensional image of the imaging volume of the object 16 or a series of two-dimensional slice images. Specifically, the image reconstructor 140 may perform the above-described image reconstruction in communication with the controller 130.

[0050] The magnetic resonance imaging system 100 may include an image processor 150, which can perform any necessary post-processing on any image from the aforementioned three-dimensional image or image sequence. This post-processing may involve improvements or adaptive adjustments to the image in terms of contrast, uniformity, sharpness, brightness, artifacts, etc. In one example, the image processor 150 may include a trained deep learning network or a unit communicating with a trained neural network. The image processor 150 may also perform image processing based on communication with the controller 130.

[0051] In one embodiment, the controller 130, image reconstructor 140, and image processor 150 may individually or jointly include a computer processor and a storage medium on which a program for predetermined data processing to be executed by the computer processor is recorded. For example, the storage medium may store programs for performing scanning processing, image reconstruction, image processing, etc., such as programs for implementing the magnetic resonance imaging method of the embodiments of the present invention. The storage medium may include, for example, a ROM, floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, or non-volatile memory card.

[0052] The magnetic resonance imaging system 100 may include a display unit 160, which can be used to display the operation interface and various data, images or parameters generated during data acquisition and processing.

[0053] The magnetic resonance imaging system 100 includes an operation console 170, which may include user input devices such as a keyboard and mouse. The controller 130 can respond to control commands generated by the user based on the operation console 170 or the operation panel / buttons set on the main magnet housing, and communicate with the scanner 110, image reconstructor 140, image processor 150, display unit 160, etc.

[0054] Radio frequency interference signals from outside the magnetic resonance system may be acquired during the acquisition of magnetic resonance signals, thus forming a noisy interference image in the reconstructed image. Such noisy interference images may include, for example, chain-like artifacts.

[0055] Radio frequency shielding devices can be used to isolate external environmental interference with image signals, thereby reducing the aforementioned artifacts. However, the causes of these artifacts can be varied, such as insufficient shielding in the scanning room, gaps in the equipment's shielding layer, or the door to the MRI room not being closed tightly, leading to the introduction of external radio frequency interference.

[0056] To reduce such artifacts, complex analysis may be required to identify the sources of radio frequency interference (RF interference) and remove artifacts from the image based on the characteristics of the RF interference sources. When there are many or new types of RF interference sources, the complexity of image processing is increased, costs are increased, and the efficiency and / or effectiveness of image processing are reduced.

[0057] Figure 2 A flowchart 200 illustrating a magnetic resonance imaging method according to some embodiments of the present invention is shown, which can be applied to... Figure 1The method is implemented using the system shown or any variation thereof. Specifically, in step S203, a medical scan is performed on the object 16 to obtain a first medical image with a first noise interference artifact, and an additional scan is performed on the object 16 to obtain a second medical image, wherein the second medical image has a second noise interference artifact. In step S205, the first and second medical images are processed to obtain a post-processed image with reduced noise interference artifacts.

[0058] In one embodiment, the above processing is a synthesis-related process, which may include various image processing techniques to generate a post-processed image from a first medical image and a second medical image through a specific processing method. For example, data fusion of the image data of the first medical image and the second medical image, deep learning of the first medical image and the second medical image, or data fusion and deep learning of the first medical image may be performed to obtain a post-processed image that has the same region of interest as the first medical image and the second medical image and reduces radio frequency noise interference artifacts.

[0059] The aforementioned medical scan may include a formal scan performed on object 16. Images from formal scans are typically used for clinical diagnosis. Therefore, eliminating artifacts in these images is often a key aspect of patient diagnostic assessment. However, for images generated during scans at any other stage or process (e.g., localization images), where it is necessary to eliminate artifacts relevant to the embodiments of the present invention, an additional scan may be performed to obtain a second medical image (in which the noise interference artifacts are located in a different position than those in the first medical image). The aforementioned artifacts in the images may then be removed based on subsequent processing of the first and second medical images.

[0060] Figure 3 An example of a first medical image 31 is shown, which may be an image obtained in any scan procedure of a magnetic resonance imaging examination, in which chain-like noise interference artifacts appear. Since radio frequency interference noise has a fixed frequency and is typically acquired in multiple (continuous or spaced) signal acquisition cycles, chain-like artifacts are formed in the image domain along the phase encoding direction (perpendicular to the frequency encoding direction). Figure 3 An example of a second medical image 32 is also shown, which, as previously described, is obtained in an additional scan that may have the same scan site and substantially the same scan parameter settings as the aforementioned medical scan, such that the data used to describe the tissue structure of object 16 in the second medical image 32 and the first medical image 31 are substantially the same.

[0061] The difference lies in the location of the noise artifact in the second medical image 32, which differs from its location in the first medical image 31. This can be achieved by changing a certain scanning parameter, as will be described in detail below. Specifically, the location mapping of the second noise artifact in the first medical image (e.g., location L2) is symmetrical to the location of the first noise artifact (L1) with respect to the pixel center of the first medical image. This pixel center can be the intersection of the zero-phase line and the center frequency line. As mentioned earlier, noise artifacts from the same source (or the same frequency) are distributed along the phase encoding direction. Therefore, the location mapping of the second noise artifact (L2) in the first medical image is symmetrical to the location L1 of the first noise with respect to the center frequency line. In other words, the mapping of the second noise artifact on the first medical image 31 is symmetrical or mirrored to the first noise artifact with respect to the center frequency line or the image center. For example, the noise artifacts of the first and second medical images are located on opposite sides of the center frequency line and are equidistant from it.

[0062] In one embodiment, the frequency encoding directions used in the additional scan and the medical scan are opposite, thereby making the first noise interference artifact and the second noise interference artifact mirror images or symmetrical. Furthermore, the additional scan can be performed either before or after the medical scan, and other scanning procedures can also be performed between the additional scan and the medical scan.

[0063] Figure 4 An example of K-space encoding used in medical scanning is shown. Figure 5 An example of K-space encoding used in an additional scan is shown, in which, Figure 5 Frequency coding direction and Figure 4 In the gradient coil driver 115, the frequency encoding direction Kx is opposite, while the phase encoding direction Ky is the same. For example, in a medical scan, the gradient field applied in the frequency encoding direction gradually increases, while in an additional scan, the gradient field applied in the frequency encoding direction gradually decreases; or, in a medical scan, the gradient field applied in the frequency encoding direction gradually decreases, while in an additional scan, the gradient field applied in the frequency encoding direction gradually increases. This can be achieved by controlling the direction and / or magnitude of the drive current output by the gradient coil driver 115.

[0064] As mentioned earlier, one frequency coding line is filled in each signal acquisition cycle (or within one repetition of the scan sequence). The more frequency coding lines acquired, the more image details can be presented in the reconstructed image. Since additional scans require extra examination time, in embodiments of the present invention, the time of additional scans can be shortened by downsampling, i.e., less data is acquired compared to medical scans. For example, only the frequency coding lines of the central region A1 in K-space are filled, with the center frequency line of region A1 as the center, for example, centered on the center frequency coding line. Data outside this central region is filled with 0. The second medical image 32 obtained in this way has a lower resolution, but can be acquired in a shorter time.

[0065] Figure 3 An example of post-processed image 33 is also shown. Since the location of the noise artifact in the second medical image 32 is different from that in the first medical image, and the first noise artifact does not exist at the location mapping (L3) of the second medical image, a major objective of processing the first and second medical images in step S205 is to use clean data from the two images, respectively, to synthesize / fuse data to obtain a post-processed image with reduced artifacts. For example, the first noise artifact in the first medical image 31 is replaced with image data without noise artifacts at the location mapping L3 of the second medical image 32, so as to reflect the anatomical structure at that location as realistically as possible while removing the noise artifact.

[0066] In one embodiment, in step S205, the first medical image 31 and the second medical image 32 are input into a trained first deep learning network for deep learning to obtain the post-processed image 33. The post-processed image 33 is output after the first medical image 31 and the second medical image are input together into the deep learning network for deep learning.

[0067] This embodiment achieves the main processing objectives described above well through a first deep learning network. However, those skilled in the art will understand that other image processing techniques can be used to achieve the above objectives. For example, in step S205, data fusion can be performed on the first medical image and the second medical image to obtain a post-processed image with reduced noise interference artifacts. This data fusion may include specific calculations of pixel values, image stitching, etc.

[0068] Figure 6A block diagram of a first deep learning network according to an embodiment of the present invention is shown. The first deep learning network includes a preprocessing module 610 and a processing module 630. The preprocessing module 610 is used to preprocess a first medical image 31 and a second medical image 32 respectively, and then fuse the preprocessed first medical image 31 and the preprocessed second medical image 32. The processing module 630 is used to process the result of the data fusion to output the post-processed image.

[0069] In one embodiment, the processing module 630 is used to perform a fusion operation on corresponding pixels of the preprocessed first medical image and the preprocessed second medical image. The fusion operation may include, for example, adding pixel values.

[0070] In one exemplary embodiment, the preprocessing module 610 may have two branches. The first branch sequentially includes: a first input terminal Input1 for receiving a first medical image, a first convolutional layer Conv1 for convolutional processing of the first medical image, a first batch normalization layer BN1 for batch normalizing the convolutionally processed image data, and a first activation layer Acti1 for nonlinearly mapping the output of the batch normalization layer. The second branch sequentially includes: a second input terminal Input2 for receiving a second medical image, a second convolutional layer Conv2 for convolutional processing of the second medical image, a second batch normalization layer BN2 for batch normalizing the convolutionally processed image data, and a second activation layer Acti2 for nonlinearly mapping the output of the batch normalization layer. The preprocessing module 610 also includes an output layer for outputting fused data of the outputs of the first activation layer Acti1 and the second activation layer Acti2.

[0071] In one exemplary embodiment, the processing module 630 includes a generator network comprising a plurality of encoders 631 and a plurality of decoders 632 connected in sequence. Each encoder 631 includes an input layer, a convolutional layer, a batch normalization layer, an activation layer, and an output layer. The input layer of the first encoder receives the output of the preprocessing module, i.e., the fused data. Each subsequent encoder receives the output of the previous encoder and performs convolution, normalization, activation, and other processing on that output. Each decoder 632 includes two inputs, a merging layer, a deconvolutional layer, a batch normalization layer, an activation layer, and an output layer. Each decoder is cascaded with a symmetrical encoder; for example, the first decoder is symmetrically cascaded with the last encoder, the second decoder with the second-to-last encoder, the third decoder with the third-to-last encoder, and so on. One input of each decoder receives the feature map output by the previous decoder, and the other input receives the feature map to be merged (e.g., the feature map output by an encoder symmetrically cascaded with it). The merging layer concatenates the two feature maps. The deconvolutional layer is used to upsample the concatenation result, the batch normalization layer is used to normalize the upsampled result, the activation layer is used to perform nonlinear mapping on the normalized result, and the output layer is used to output the result of the nonlinear mapping.

[0072] The processing module 630 may also include a processing layer that processes the output of the last decoder.

[0073] The embodiments of the present invention describe an exemplary structure of a first deep learning network. In other embodiments, the first deep learning network may also have other structures. For example, it may output a post-processed image directly by deep learning from both medical images without fusing the first and second medical images.

[0074] The first deep learning network described above and the second deep learning network described below can be trained based on a known input dataset and a known output dataset by setting the number of neurons in the processing layers of the network and optimizing the network parameters (including but not limited to weights, biases, etc.) to identify the mathematical relationship between the known input and the expected output and to represent the mathematical relationship between the input and output of each layer, so that the loss function converges.

[0075] In one embodiment, the first deep learning network can be trained based on an adversarial network, wherein multiple image pairs (first image and second image) with mirrored noise interference artifacts can be used as the first and second inputs of the training dataset, and an image without noise interference artifacts (third image) can be used as the output of the training dataset to train the first deep learning network.

[0076] The adversarial network includes a generator, the structure of which is similar to... Figure 6 The network structure shown may further include a discriminator, which receives the generator's output image or a gold standard image and outputs a judgment result indicating whether the received image contains noise interference artifacts.

[0077] The first image in the image pair used as the training dataset can be an image with noise artifacts obtained in a clinical scan. The second image in the image pair can also be obtained by scanning. For example, if the first image is obtained based on a certain scan parameter setting, the second image can be obtained (before or after obtaining the first image) with the same parameter setting (only the direction of frequency encoding is changed). The third image can also be an image without artifacts obtained in a clinical scan, which can have the same anatomical location or region of interest as the first and second images.

[0078] refer to Figure 7 In other embodiments, the first input to the training dataset, i.e., the first image, can be obtained by synthesizing a noisy image with a clinical image free of noise artifacts. Specifically, this includes the following steps:

[0079] In step one, clinical images free from interference noise artifacts are acquired (e.g., Figure 7 Image 71), background noise image (e.g.) Figure 7 Image 72) and noisy images (such as Figure 7 Image 73);

[0080] In step two, a noise interference image is synthesized from the image based on the interference noise and the background noise image;

[0081] In step three, clinical images free from interference noise artifacts are converted into first frequency domain images;

[0082] In step four, the synthesized noisy interference image is converted into a second frequency domain image;

[0083] In step five, a third frequency domain image is synthesized based on the second frequency domain image and the first frequency domain image (e.g., ...). Figure 7 Image 74);

[0084] In step six, the third frequency domain image is converted into a time domain image to serve as the first image mentioned above (e.g., Figure 7 Image 75 in the image.

[0085] Continue to refer to Figure 7 It can also generate a second input to the training dataset, namely a second image, which can specifically include the following steps:

[0086] In step seven, a mirror noise interference image is obtained based on the synthesized noise interference image. The interference noise in the mirror noise interference image is located at a symmetrical (or mirror) position with the original synthesized noise interference image.

[0087] In step eight, the image with mirror noise interference is converted into a fourth frequency domain image;

[0088] In step nine, the fourth frequency domain image is converted into a time domain image to serve as the second image described above.

[0089] Optionally, when a lower-resolution second image is required for model training, after obtaining the fourth frequency domain image, a tenth step may be included, wherein the fourth frequency domain image is low-pass filtered to obtain a filtered fourth frequency domain image (e.g., Figure 7 As shown in image 76), in step eleven, the low-pass filtered fourth frequency domain image is converted into a time domain image to serve as the aforementioned second image with lower resolution (e.g., image 76). Figure 7 Image 77 shown.

[0090] In one embodiment, while the configuration of the deep learning network is guided by prior knowledge of the estimation problem, the dimensions of the inputs, outputs, etc., it relies on or specifically on achieving the best approximation of the desired output data based on the input data. In various alternative implementations, certain aspects and / or features of the data, imaging geometry, reconstruction algorithms, etc., can be leveraged to give explicit meaning to certain data representations in the deep learning network, which can help accelerate training. This is because it creates the opportunity to train (or pre-train) or define certain layers separately in the deep learning network.

[0091] In some embodiments, the first deep learning network described above and the second deep learning network described below are trained on a training module on an external carrier (e.g., a device other than a medical imaging system). In some embodiments, the training system may include a first module for storing training datasets, a second module for training and / or updating based on the model, and a communication network for connecting the first and second modules. In some embodiments, the first module includes a first processing unit and a first storage unit, wherein the first storage unit is used to store the training dataset, and the first processing unit is used to receive relevant instructions (e.g., obtain the training dataset) and send the training dataset according to the instructions. Furthermore, the second module includes a second processing unit and a second storage unit, wherein the second storage unit is used to store the training model, and the second processing unit is used to receive relevant instructions, train and / or update the network, etc. In other embodiments, the training dataset may also be stored in the second storage unit of the second module, and the training system may not include the first module. In some embodiments, the communication network may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0092] Once data (e.g., a trained network) is generated and / or configured, it can be copied and / or loaded into a medical imaging system (e.g., a magnetic resonance imaging system described below), which can be done in various ways. For example, the loading model can be achieved via a directed connection or link between the medical imaging system and the computer. In this regard, communication between different components can be accomplished using available wired and / or wireless connections and / or according to any suitable communication (and / or network) standards or protocols. Alternatively or additionally, data can be loaded into the medical imaging system indirectly. For example, data can be stored on a suitable machine-readable medium (e.g., a flash memory card, etc.) and then used to load the data (on-site, such as by the system's user or authorized personnel) into the medical imaging system, or data can be downloaded to an electronic device capable of local communication (e.g., a laptop computer, etc.) and then used on-site (e.g., by the system's user or authorized personnel) to upload the data to the medical imaging system via a direct connection (e.g., a USB connector, etc.).

[0093] As discussed in this paper, deep learning techniques (also known as deep machine learning, hierarchical learning, or deep structured learning) employ artificial neural networks for learning. Deep learning methods are characterized by using one or more network architectures to extract or model data of interest. Deep learning methods can be accomplished using one or more processing layers (e.g., input layers, output layers, convolutional layers, normalization layers, sampling layers, etc., with varying numbers and functions depending on the deep network model). The configuration and number of layers allow deep networks to handle complex information extraction and modeling tasks. Specific parameters (also called “weights” or “biases”) of the network are typically estimated through a so-called learning process (or training process). The parameters learned or trained usually result in (or output) a network corresponding to different levels of layers; therefore, extracting or modeling different aspects of the initial data or the output of a previous layer can often represent the hierarchical structure or cascade of layers. In image processing or reconstruction, this can be characterized as different layers relative to different levels of features in the data. Thus, processing can be layered; that is, earlier or higher-level layers may correspond to extracting “simple” features from the input data, followed by layers that combine these simple features into features exhibiting higher complexity. In practice, each layer (or more specifically, each "neuron" in each layer) can employ one or more linear and / or nonlinear transformations (so-called activation functions) to process the input data into an output data representation. The number of "neurons" can be constant across multiple layers, or it can vary from layer to layer.

[0094] Figure 8 A flowchart of a medical imaging method according to another embodiment of the present invention is shown, which is consistent with... Figure 2 The method shown is similar, except that before step S203, i.e., before performing the first medical scan, steps S801 and S802 are included. In step S801, a prior scan image of the object is acquired. In step S802, it is determined whether noise artifacts exist in the prior scan image. If so, step S205 is executed, i.e., a medical scan is performed on the object to obtain the first medical image, and the additional scan is performed on the object to obtain the second medical image. If no artifacts exist in the prior scan image, only a regular medical scan is performed to acquire the medical image.

[0095] In this embodiment, the noise interference artifacts are first identified in the previous scan before the noise interference artifact removal process is started, avoiding redundant scanning processes and unnecessary image processing work, thereby improving imaging efficiency.

[0096] In other embodiments, it is also possible to directly determine whether noise interference artifacts exist in the first medical image. If they do, the additional scan is initiated and subsequent artifact removal processing is performed. However, since the MRI scan process typically involves pre-scanning, localization scans, and other procedures to obtain prior scan images before acquiring clinical diagnostic images, determining these prior scan images allows for advance preparation for the additional scan and image post-processing. Furthermore, determining the prior scan images can be performed synchronously during other scanning tasks, saving MRI examination time and avoiding waiting time between medical scans and additional scans.

[0097] As previously stated, the medical scan may include a formal scan. The prior scan image may include a pre-scan image obtained by pre-scanning the object, such as a shimming image used for main magnetic field homogeneity analysis and adjustment. The prior scan image may also include a localization image. In other embodiments, the image acquired by the formal scan may also serve as a prior scan image for other subsequent medical scans.

[0098] In one embodiment, the presence of noise interference artifacts in the previously scanned image can be determined based on a trained second deep learning network. An existing deep learning network can be selected to train the dataset to obtain the second deep learning network, which will not be elaborated further here.

[0099] Embodiments of the present invention may also provide a computer-readable storage medium comprising a stored computer program, wherein any embodiment of the magnetic resonance imaging method described above is executed when the computer program is run.

[0100] Figure 9 A block diagram 900 of a magnetic resonance imaging system according to an embodiment of the present invention is shown, which may include Figure 1 The system 100 shown includes some or all of its components, for example, the system 900 includes a scanner 910, a controller 930, and an image processor 950. The controller 930 controls the scanner 910 to perform magnetic resonance scanning, including:

[0101] Perform a medical scan on the object and acquire a first medical image, the first medical image having a first noise artifact; and,

[0102] An additional scan is performed on the object to obtain a second medical image, wherein the second medical image has a second noise interference artifact, the position of the second noise interference artifact in the first medical image being mapped symmetrically with respect to the position of the first noise interference artifact relative to the pixel center of the first medical image;

[0103] The image processor is used to process a first medical image and a second medical image to obtain a post-processed image with reduced noise artifacts.

[0104] In one embodiment, the controller 930 is used to control the scanner to use opposite frequency encoding directions when performing additional and medical scans, such that the second noise interference artifact is symmetrical or mirrored with respect to the pixel center of the image to the first noise interference artifact.

[0105] In one embodiment, the resolution of the second medical image is lower than that of the first medical image. Acquiring the lower-resolution image requires less time, thus reducing the time required to perform the additional scan. For example, the controller 930 can control the scanner 910 to acquire less data during the additional scan compared to the medical scan. Specifically, data near the center frequency encoding line in K-space (e.g., low-frequency data in the central region of K-space) can be sampled, while data outside the central region / data near the edge frequency encoding lines (e.g., high-frequency data near the edge) can be left unsampled and filled with 0.

[0106] In one embodiment, the image processor 950 includes a first deep learning network trained on the first medical image and the second medical image to output the post-processed image.

[0107] In one embodiment, the first deep learning network includes a preprocessing module and a processing module. The preprocessing module preprocesses the first medical image and the second medical image respectively, and then fuses the preprocessed first medical image and the preprocessed second medical image. The processing module processes the result of the data fusion to output the post-processed image.

[0108] In one embodiment, the data fusion includes performing a fusion operation on the corresponding pixel values ​​of the preprocessed first medical image and the preprocessed second medical image.

[0109] In one embodiment, the first deep learning network is trained based on an adversarial network.

[0110] Figure 10 A block diagram of a magnetic resonance imaging system according to another embodiment of the present invention is shown, which is related to... Figure 9 The system shown is similar, except that it also includes a judgment module 1010. In this embodiment, the controller 910 is also used to control the scanner to acquire a prior scan image of the object before performing the medical scan. The judgment module 1010 is used to determine whether there are noise interference artifacts in the prior scan image. If so, the controller 910 controls the scanner to perform the medical scan on the object and acquire a first medical image, and then performs an additional scan on the object to obtain the second medical image. The judgment module 1010 can be a module independent of the controller 910 or integrated with the controller 910 as part of the controller 910.

[0111] In one example, controller 910 may first control the scanner to perform a prior scan, and then control the scanner to perform a medical scan. If (e.g., based on a judgment of the prior scan image) it is determined that the medical image acquired by the medical scan will contain noise interference artifacts, an additional scan is performed. The prior scan image may include a pre-scanned image obtained by pre-scanning the object, or the positioning image described above, which may include a shimming image or an image used for other applications.

[0112] In one embodiment, the determination module 1010 includes a trained second deep learning network for performing deep learning on the previously scanned image to determine whether there is a noisy image interfering with it.

[0113] In various embodiments of the present invention, in order to improve the noise interference artifacts in the first medical image obtained in the medical scan, an additional scan is performed on the object to obtain a second medical image, and the position (e.g., coordinates in the image) of the noise interference artifacts in the second medical image is different from that of the noise interference artifacts in the first medical image. A post-processed image with reduced artifacts is obtained based on the first and second images, avoiding the loss of anatomical image data or inaccurate / incomplete restoration caused by removing artifacts from the first medical image alone, so as to reflect the anatomical structure of the artifact position as realistically as possible while removing noise interference artifacts.

[0114] In various embodiments of the present invention, images with good artifact removal effects can be obtained without analyzing the noise sources that generate noise interference artifacts. To verify the technical effect of the present invention, comparative imaging experiments were conducted on at least six parts of the human body. Figure 11 As shown, a gold standard image 1101 of a region, an original image 1102 with noise artifacts, and a post-processed image 1103 obtained based on an embodiment of the present invention are presented for observation. Figure 11 As can be seen from the white box area in the image, the post-processed image 1103, compared to the original image 1102, significantly reduces noise artifacts and is very close to the gold standard image 1101. The above embodiments of the present invention obtain the post-processed image based on the first and second medical images. Furthermore, Figure 11 Post-processed image 1104, obtained solely from a first medical image (e.g., undergoing similar image processing), is also shown. Comparing images 1101, 1102, 1103, and 1104, it can be seen that although image 1104 also reduces noise artifacts compared to the original image 1101, post-processed image 1104 has better artifact removal performance and is closer to the gold standard image.

[0115] Furthermore, through experimental analysis, the post-processed images obtained based on the various embodiments of the present invention have significant advantages over post-processed images obtained based on only one image in other metrics, such as the mean square error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM).

[0116] The purpose of providing the above specific embodiments is to enable a more thorough and comprehensive understanding of the disclosure of this invention, but this invention is not limited to these specific embodiments. Those skilled in the art should understand that various modifications, equivalent substitutions, and changes can be made to this invention, and all such modifications and changes should be within the scope of protection of this invention, provided they do not depart from the spirit of this invention.

Claims

1. A magnetic resonance imaging method, comprising: A medical scan is performed on the object to acquire a first medical image with a first noise interference artifact, and an additional scan is performed on the object to acquire a second medical image, wherein the second medical image has a second noise interference artifact, and the position of the second noise interference artifact in the first medical image is mapped to be symmetrical about the position of the first noise interference artifact with respect to the pixel center of the first medical image; and, The first and second medical images are subjected to synthesis-related processing to obtain post-processed images with reduced noise artifacts. In this context, the frequency encoding directions used in the additional scan and the medical scan are opposite, while the phase encoding directions are the same.

2. The magnetic resonance imaging method according to claim 1, wherein, The resolution of the second medical image is lower than that of the first medical image.

3. The magnetic resonance imaging method according to claim 2, wherein, During the additional scan, the central region of the K-space passing through the center frequency line is sampled, and the data outside the central region is filled with 0.

4. The magnetic resonance imaging method according to claim 1, wherein, The steps for performing synthesis-related processing on the first and second medical images include: The first medical image and the second medical image are input into a trained first deep learning network for deep learning to obtain the post-processed image.

5. The magnetic resonance imaging method according to claim 4, wherein, The first deep learning network includes: The preprocessing module is used to preprocess the first medical image and the second medical image respectively, and to fuse the preprocessed first medical image and the preprocessed second medical image; and, The processing module is used to process the result of the data fusion to output the post-processed image.

6. The magnetic resonance imaging method according to claim 5, wherein, The data fusion includes performing a fusion operation on the corresponding pixel values ​​of the preprocessed first medical image and the preprocessed second medical image.

7. The magnetic resonance imaging method according to claim 4, wherein, The steps for performing synthesis-related processing on the first and second medical images include: The first medical image and the second medical image are fused together to obtain the post-processed image.

8. The magnetic resonance imaging method according to claim 1, wherein, Also includes: Acquire a prior scan image of the object before performing the medical scan; Determine whether there are noise interference artifacts in the previously scanned image. If so, perform the additional scan on the object to obtain the second medical image.

9. The magnetic resonance imaging method according to claim 8, wherein, The prior scanned image includes a pre-scanned image obtained by pre-scanning the object.

10. The magnetic resonance imaging method according to claim 9, wherein, The pre-scanned image includes a shimming image.

11. The magnetic resonance imaging method according to claim 8, wherein, The prior scanned images include positioning images.

12. The magnetic resonance imaging method according to claim 8, wherein, A second deep learning network, based on training, determines whether noise interference artifacts exist in the previously scanned image.

13. A computer-readable storage medium comprising a stored computer program, wherein, When the computer program is run, it performs the magnetic resonance imaging method according to any one of claims 1-12.

14. A magnetic resonance imaging system, comprising: Scanner; A controller for controlling the scanner to perform a magnetic resonance scan, the magnetic resonance scan including: Perform a medical scan on the object and acquire a first medical image, the first medical image having a first noise artifact; and, An additional scan is performed on the object to obtain a second medical image, wherein the second medical image has a second noise interference artifact, the position of the second noise interference artifact in the first medical image being mapped symmetrically with respect to the pixel center of the first medical image to the position of the first noise interference artifact; and, An image processor is used to perform synthesis-related processing on a first medical image and a second medical image to obtain a post-processed image with reduced noise artifacts. In this context, the frequency encoding directions used in the additional scan and the medical scan are opposite, while the phase encoding directions are the same.

15. The magnetic resonance imaging system according to claim 14, wherein, The resolution of the second medical image is lower than that of the first medical image.

16. The magnetic resonance imaging system according to claim 14, wherein, The image processor includes a first deep learning network trained on the first and second medical images to output the post-processed image.

17. The magnetic resonance imaging system according to claim 16, wherein, The first deep learning network includes a preprocessing module and a processing module. The preprocessing module is used to preprocess the first medical image and the second medical image respectively, and to fuse the preprocessed first medical image and the preprocessed second medical image. The processing module is used to process the result of the data fusion to output the postprocessed image.

18. The magnetic resonance imaging system according to claim 14, wherein, The controller is also used to control the scanner to acquire a prior scan image of the object before performing the medical scan. The magnetic resonance imaging system also includes a judgment module, which is used to determine whether there are noise interference artifacts in the prior scan image. If so, the controller controls the scanner to perform the additional scan on the object to obtain the second medical image.

Images