WO2022257090A1 - 一种磁共振成像方法及相关设备 - Google Patents

一种磁共振成像方法及相关设备 Download PDF

Info

Publication number
WO2022257090A1
WO2022257090A1 PCT/CN2021/099570 CN2021099570W WO2022257090A1 WO 2022257090 A1 WO2022257090 A1 WO 2022257090A1 CN 2021099570 W CN2021099570 W CN 2021099570W WO 2022257090 A1 WO2022257090 A1 WO 2022257090A1
Authority
WO
WIPO (PCT)
Prior art keywords
magnetic resonance
image
resonance imaging
space data
neural network
Prior art date
Application number
PCT/CN2021/099570
Other languages
English (en)
French (fr)
Inventor
龚南杰
祁成晓
潘博洋
Original Assignee
苏州深透智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 苏州深透智能科技有限公司 filed Critical 苏州深透智能科技有限公司
Publication of WO2022257090A1 publication Critical patent/WO2022257090A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
    • G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution

Definitions

  • the embodiments of the present application relate to the field of image processing, and in particular to a magnetic resonance imaging method and related equipment.
  • Magnetic resonance imaging Magnetic Resonance Imaging, MRI is an imaging technique that uses the signal generated by the resonance of the atomic nucleus in a strong magnetic field to reconstruct the image. It is a nuclear physics phenomenon. It uses radio frequency pulses to excite atomic nuclei with non-zero spins in a magnetic field. After the radio frequency pulses stop, the atomic nuclei relax. During the relaxation process, the signals are collected by induction coils and reconstructed according to certain mathematical methods. Mathematical image. Magnetic resonance imaging differs from other imaging techniques in that it provides far more information than many other imaging techniques in medical imaging. Therefore, the diagnosis of the disease has a great obvious advantage.
  • the collected signals are K-space data in the frequency domain, and the structural image information of the patient is obtained through Fourier transform. It takes a lot of time to collect and process the signal.
  • a variety of algorithms that can obtain MRI images using less K-space data have been continuously proposed.
  • the first aspect of the embodiment of the present application provides a magnetic resonance imaging method, including:
  • the first magnetic resonance image is obtained based on partial K-space data of a region to be detected of the target patient;
  • the pre-trained image processing neural network model Inputting the first magnetic resonance image into a pre-trained image processing neural network model, the pre-trained image processing neural network model, the pre-trained image processing neural network model consists of low-quality magnetic resonance imaging and
  • the training set of high-quality magnetic resonance imaging corresponding to the low-quality magnetic resonance imaging is obtained through training.
  • the low-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on partial K-space data of a specific region of the patient.
  • the imaging is magnetic resonance imaging obtained based on all or part of the K-space data of a specific region of the patient;
  • a second magnetic resonance image corresponding to the first magnetic resonance image output by the image processing neural network model is obtained.
  • the method further includes:
  • the part of the K-space data is processed based on a preset algorithm to obtain a first magnetic resonance image.
  • processing the part of the K-space data based on a preset algorithm to obtain the first magnetic resonance image includes:
  • Part of the K-space data of the region to be detected of the target patient is processed based on a generalized automatic calibration partial parallel acquisition algorithm to obtain a first magnetic resonance image.
  • the method further includes:
  • the low-quality magnetic resonance imaging is:
  • the image processing neural network model is a residual convolutional neural network
  • the image processing neural network model can be a residual dense network or a network model with an encoder-decoder structure; wherein the residual dense network includes at least one dense block, and the dense block is composed of a convolutional layer and a nonlinear activation At least two basic modules composed of layers are connected through dense connections.
  • the second aspect of the embodiment of the present application provides a magnetic resonance imaging device, including:
  • an acquisition unit configured to acquire a first magnetic resonance image of the target patient, the first magnetic resonance image is obtained based on partial K-space data of a region to be detected of the target patient;
  • the input unit is configured to input the first magnetic resonance image into a pre-trained image processing neural network model, the pre-trained image processing neural network model, the pre-trained image processing neural network model consists of low
  • the high-quality magnetic resonance imaging and the high-quality magnetic resonance imaging training set corresponding to the low-quality magnetic resonance imaging are obtained through training, and the low-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on partial K-space data of a specific region of the patient, so The high-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on all or part of the K-space data of a specific region of the patient;
  • An obtaining unit configured to obtain a second magnetic resonance image corresponding to the first magnetic resonance image output by the image processing neural network model.
  • the third aspect of the embodiment of the present application provides a computer device, including:
  • Central processing unit memory, input and output interfaces, wired or wireless network interface and power supply;
  • the memory is a temporary storage memory or a persistent storage memory
  • the central processing unit is configured to communicate with the memory, and execute instructions in the memory on the computer device to perform the method described in any one of the first aspects of the embodiments of the present application.
  • the fourth aspect of the embodiments of the present application provides a computer-readable storage medium, including instructions, which, when run on a computer, cause the computer to execute the method described in any one of the first aspects of the embodiments of the present application.
  • the fifth aspect of the embodiments of the present application provides a computer program product including instructions, which, when run on a computer, cause the computer to execute the method described in any one of the first aspects of the embodiments of the present application.
  • this solution uses the image processing neural network model to process the first magnetic resonance image obtained using part of the K-space data, wherein the image processing neural network model consists of low-quality
  • the magnetic resonance imaging and the high-quality magnetic resonance imaging training set corresponding to the low-quality magnetic resonance imaging are obtained through training, so the second magnetic resonance image obtained by using the image processing neural network model is compared with the first magnetic resonance image, It is closer to the imaging obtained by using all or part of the K-space data, so more organ tissue information of the patient can be obtained by using the second magnetic resonance image, the resolution of the imaging obtained by magnetic resonance imaging is improved, and the signal-to-noise ratio is improved.
  • Fig. 1 is a schematic flow chart of an embodiment of the magnetic resonance imaging method of the present application
  • Fig. 2 is another schematic flow chart of the embodiment of the magnetic resonance imaging method of the present application.
  • Fig. 3 is a schematic flow chart of the application's image processing neural network model training and use process
  • Fig. 4 is a schematic structural diagram of an embodiment of the image processing neural network model of the present application.
  • FIG. 5 is a schematic structural diagram of an embodiment of a magnetic resonance imaging device of the present application.
  • FIG. 6 is a schematic structural diagram of an embodiment of a computer device in the present application.
  • FIG. 7A and 7B are illustrations of the effect after processing the magnetic resonance image and the magnetic resonance image input model provided in the present application.
  • an embodiment of the present application includes: Step 101 - Step 105 .
  • Magnetic resonance imaging technology is a multi-parameter, multi-contrast imaging technology. Based on magnetic resonance imaging, it can provide doctors with a wealth of patient organ and tissue information.
  • the signal collected by MRI is K-space (the dual space of ordinary space under Fourier transform) data in the frequency domain, and the structural image information of the patient is obtained through Fourier transform.
  • the quality of MRI is affected by the signal to noise ratio (Signal to Noise Ratio, SNR) and resolution (Resolution). Images with high signal-to-noise ratio and resolution can more clearly present the tiny structures in the body, helping doctors to make accurate differential diagnoses of lesions.
  • imaging methods using partial K-space data have been developed, such as the sensitivity encoding algorithm (SENSitivity Encoding, SENSE) in parallel acquisition (Parallel Imaging) or the generalized autocalibrating partially parallel acquisitions algorithm (GRAPPA). ), or the simultaneous multi-layer imaging (Simultaneous multi-slice, SMS) algorithm in the case of scanning part of the k-space sequence, using different coils to give different weights to the magnetic field signals in different regions, restores the non-existent Overlay image.
  • the newer generation of compressed sensing technology compressed sensing, CS
  • the present application provides a magnetic resonance imaging method, please refer to FIG. 1 , the method includes step 101 to step 103 .
  • the first magnetic resonance image is obtained based on partial K-space data of a region to be detected of the target patient;
  • the first magnetic resonance image is obtained by data processing based on part of the K-space data of the target patient.
  • the specific processing method can be based on sensitivity coding algorithm, obtained based on compressed sensing technology, or based on generalized automatic calibration partly parallel acquisition algorithm or based on Simultaneous multi-slice imaging (Simultaneous Multi-Slice) algorithm, or other improved methods based on the above-mentioned multiple methods, are not limited here.
  • the first magnetic resonance image is generated based on part of the K-space data, and the K-space data to be used when generating the imaging depends on the generation method adopted. For the case of using different generation algorithms, the K-space data can be collected according to different requirements, specifically There is no limit here.
  • the obtained magnetic resonance images include but are not limited to T1, T2, magnetic resonance angiography (MRA) and other types of weighted images.
  • the pre-trained image processing neural network model Inputting the first magnetic resonance image into a pre-trained image processing neural network model, the pre-trained image processing neural network model, the pre-trained image processing neural network model consists of low-quality magnetic resonance imaging and
  • the training set of high-quality magnetic resonance imaging corresponding to the low-quality magnetic resonance imaging is obtained through training.
  • the low-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on partial K-space data of a specific region of the patient.
  • the imaging is a magnetic resonance imaging obtained based on all or part of the K-space data of a specific area of the patient; the specific area can be any area preset according to imaging requirements, and it can be one area or multiple areas.
  • the area to be detected in step 101 is any one or more of the specific areas, or any part of the specific area.
  • an image preprocessing operation can also be performed on the first magnetic resonance image, and the image preprocessing operation includes image augmentation, image registration and image normalization any one or more of them.
  • the image augmentation can specifically be processing such as zooming, rotating, year-on-year increase and decrease of pixel values, mirroring, cropping or adding noise to the first magnetic resonance image;
  • the method of image registration can be specifically image rigid registration, registration based on key points, etc.
  • the registration operation of the first magnetic resonance image is realized through the image registration method, such as registration or image registration based on a neural network model.
  • Image normalization may specifically be based on normalization based on maximum and minimum values of pixels, normalization based on pixel mean values, and the like.
  • the image processing neural network model may be a convolutional neural network model architecture, and it is trained using a training set, which includes a plurality of low-quality magnetic resonance images and high-quality magnetic resonance images corresponding to the plurality of low-quality magnetic resonance images. That is, magnetic resonance imaging obtained using partial K-space data and magnetic resonance imaging obtained using all K-space data.
  • the image processing neural network model uses the above training set to train the image processing neural network model, so that the image processing neural network model analyzes the difference between the magnetic resonance imaging obtained by using part of the K-space data and the magnetic resonance imaging obtained by using all the K-space data, and collects both The high-dimensional features are learned, so that the trained image processing neural network model can output an image as similar as possible to its corresponding high-quality MRI based on the low-quality MRI.
  • a second magnetic resonance image corresponding to the first magnetic resonance image output by the image processing neural network model is obtained.
  • the second magnetic resonance image is obtained by the image processing neural network model based on the analysis of the first magnetic resonance image.
  • the image processing neural network model Before generating the second magnetic resonance image, the image processing neural network model has learned the characteristics of low-quality magnetic resonance imaging and high-quality magnetic resonance imaging. Therefore, based on the first magnetic resonance image, an image that is as similar as possible to the corresponding magnetic resonance imaging formed using all the K-space data can be output.
  • FIGS. 7A and 7B are schematic diagrams showing the effect of the second magnetic resonance image obtained after the input model processes the first magnetic resonance image. In Fig. 7A and Fig.
  • the first image is the high-quality MRI image obtained by using all the K-space data
  • the second image is the first MRI image obtained by using part of the K-space data
  • the third image is the Model processed second MR image. It can be seen from the figure that the resolution and signal-to-noise ratio of the second magnetic resonance image are higher than those of the first magnetic resonance image, and have good consistency with the high-quality magnetic resonance image.
  • this solution uses the image processing neural network model to process the first magnetic resonance image obtained using part of the K-space data, wherein the image processing neural network model consists of low-quality
  • the magnetic resonance imaging and the high-quality magnetic resonance imaging training set corresponding to the low-quality magnetic resonance imaging are obtained through training, so the second magnetic resonance image obtained by using the image processing neural network model is compared with the first magnetic resonance image, It is closer to the imaging obtained by using all the K-space data, so more organ tissue information of the patient can be obtained by using the second magnetic resonance image, the resolution of the imaging obtained by magnetic resonance imaging is improved, and the signal-to-noise ratio is improved.
  • An embodiment of the magnetic resonance imaging method of the present application includes: Step 201 - Step 205 .
  • the requirements for the collected K-space data are different based on the different processing algorithms used later, which may be determined according to the actual situation, and are not limited here.
  • the part of the K-space data is processed based on a preset algorithm to obtain a first magnetic resonance image.
  • the processing means adopted may include any one or more of the following methods:
  • Partial K-space data of the region to be detected of the target patient is processed based on a sensitivity encoding algorithm to obtain a first magnetic resonance image.
  • the Sensitivity Encoding Algorithm uses parallel imaging technology to shorten the scan time, and is widely used in various magnetic resonance imaging sequences. It is a mature imaging method that utilizes part of the K-space data.
  • Compressed sensing technology can use part of the k-space data to obtain imaging by comparing the difference between the generated image, k-space and known k-space data, and iterative regression.
  • Part of the K-space data of the region to be detected of the target patient is processed based on a generalized automatic calibration partial parallel acquisition algorithm to obtain a first magnetic resonance image.
  • GRAPPA Generalized autocalibrating partially parallel acquisitions
  • This imaging technology uses multi-channel phased array coils and high acceleration factor parallel acquisition technology to synchronize parallel excitation and simultaneous acquisition of multi-layer images, using the difference in signal intensity of signals from different parts of the tissue received by multi-channel coils, in the image domain.
  • the multi-level images collected at the same time are separated to realize the magnetic resonance imaging technology of obtaining multi-layer images with one radio frequency excitation.
  • the generation method of the first magnetic resonance image used in the actual implementation of this solution can be adjusted according to requirements, or other improved algorithms can be used, which can be determined according to actual conditions, and are not limited here.
  • steps 203 to 205 are similar to the aforementioned steps 101 to 103 in the embodiment corresponding to FIG. 1 , and will not be repeated here.
  • this solution uses part of the K-space data of the target patient for imaging, and uses the image processing neural network model to process the obtained first magnetic resonance image, wherein the image processing The neural network model is obtained by training the training set including the low-quality magnetic resonance imaging and the high-quality magnetic resonance imaging corresponding to the low-quality magnetic resonance imaging, so the image processing neural network model is used to process the obtained second magnetic resonance image and the first magnetic resonance image Compared with the first magnetic resonance image, it is closer to the imaging obtained by using all the K-space data, so using the second magnetic resonance image can obtain more organ tissue information of the patient, which improves the resolution of the imaging obtained by magnetic resonance imaging, and improves signal-to-noise ratio.
  • the low-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on partial K-space data of a specific region of the patient
  • the High-quality magnetic resonance imaging is magnetic resonance imaging obtained based on all or part of the K-space data of a specific region of the patient
  • the magnetic resonance training set includes but is not limited to T1, T2, MRA and other imaging sequences.
  • the low-quality magnetic resonance imaging can be: a magnetic resonance imaging obtained by performing a sensitivity encoding algorithm based on part of the K-space data of a specific area of the patient, or a magnetic resonance image obtained by performing compressed sensing technology processing based on part of the K-space data of a specific area of the patient.
  • Resonance imaging or any one or more of magnetic resonance imaging obtained by performing a generalized auto-calibration partial parallel acquisition algorithm based on partial K-space data of a specific region of the patient.
  • the acquisition method of the low-quality magnetic resonance imaging included in the training set should be the same as the acquisition method of the first magnetic resonance image in the subsequent input image processing neural network model Consistent, thereby ensuring the processing effect of the image processing neural network model.
  • the image processing neural network model used in this step can be any one of all types of neural networks.
  • the residual dense network residual in residual dense block, RRDB
  • the network is formed by several different weight ( ⁇ ) dense block (dense block, DB) residual connections (the plus sign in the figure indicates the residual connection), and as shown by the rightmost plus sign, between the input layer and the output layer use residual connections.
  • ⁇ dense block
  • DB dense block
  • Dense connection means that the output of each basic module is formed by the output residual connection of all previous basic modules. It should be noted that the number of dense blocks and the number of basic modules contained in each dense block are not limited to those shown in Figure 4, and may be other values set according to actual needs.
  • the residual dense network RRDB network used in Figure 4 is only an example of the residual convolutional network structure that can implement this scheme.
  • the depth of the residual convolutional network structure (such as the number of dense blocks ) and structure (such as the specific implementation structure of the nonlinear activation layer) are adaptively adjusted, which is not specifically limited here.
  • the training process of the image processing neural network model needs to use a loss function for training control, and the corresponding image processing neural network model when the loss function is minimized is a trained image processing neural network model.
  • the loss function can include a combination of one or more of mean absolute error, mean square difference error, structural similarity error, and confrontation generation error;
  • the method of minimizing the loss function can include stochastic gradient optimization, ADAM (adaptive motion estimation Algorithm) optimization and other methods.
  • the foregoing steps 303 to 305 are similar to the foregoing steps 101 to 103 in the embodiment corresponding to FIG. 1 , and will not be repeated here.
  • step 301 Use the training set described in step 301 to train the image processing neural network model with the above structure for subsequent use.
  • an embodiment of the magnetic resonance imaging equipment of the present application includes:
  • An acquisition unit 501 configured to acquire a first magnetic resonance image of a target patient, where the first magnetic resonance image is obtained based on partial K-space data of a region to be detected of the target patient;
  • the input unit 502 is configured to input the first magnetic resonance image into a pre-trained image processing neural network model, the pre-trained image processing neural network model, the pre-trained image processing neural network model consists of The low-quality magnetic resonance imaging and the training set training of the high-quality magnetic resonance imaging corresponding to the low-quality magnetic resonance imaging are obtained, and the low-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on partial K-space data of a specific region of the patient, The high-quality magnetic resonance imaging is the magnetic resonance imaging obtained based on all or part of the K-space data of a specific region of the patient;
  • the obtaining unit 503 is configured to obtain a second magnetic resonance image corresponding to the first magnetic resonance image output by the image processing neural network model.
  • the acquisition device is further configured to: acquire partial K-space data of the target patient's region to be detected;
  • the part of the K-space data is processed based on a preset algorithm to obtain a first magnetic resonance image.
  • the acquisition device is specifically used for:
  • Part of the K-space data of the region to be detected of the target patient is processed based on a generalized automatic calibration partial parallel acquisition algorithm to obtain any one of the first magnetic resonance images.
  • the device further includes: a training unit, configured to acquire a training set composed of low-quality magnetic resonance imaging and high-quality magnetic resonance imaging corresponding to the low-quality magnetic resonance imaging;
  • the low-quality magnetic resonance imaging is:
  • the image processing neural network model is a residual convolutional neural network
  • the image processing neural network model is a residual dense network or other network models; wherein the residual dense network includes at least one dense block, and the dense block consists of at least two basic modules consisting of a convolutional layer and a nonlinear activation layer Connect by means of dense connection.
  • the neural network training module adopts a supervised learning method. It takes low-quality MRI images as input and high-quality MRI images as labels.
  • the loss functions used for training include but are not limited to L1 loss, L2 loss, SSIM loss, GAN loss, etc., and can be one or more of them.
  • FIG. 6 is a schematic structural diagram of a computer device provided by an embodiment of the present application.
  • the computer device 600 may include one or more central processing units (central processing units, CPU) 601 and a memory 605, in which one or more More than one application or data.
  • CPU central processing units
  • the division of specific functional modules in the central processing unit 601 may be similar to the division of functional modules of each unit described in FIG. 5 above, which will not be repeated here.
  • the storage 605 may be a volatile storage or a persistent storage.
  • the program stored in the memory 605 may include one or more modules, and each module may include a series of instructions to operate on the server.
  • the central processing unit 601 may be configured to communicate with the memory 605 , and execute a series of instruction operations in the memory 605 on the server 600 .
  • the server 600 can also include one or more power supplies 602, one or more wired or wireless network interfaces 603, one or more input and output interfaces 604, and/or, one or more operating systems, such as Windows ServerTM, Mac OS XTM , UnixTM, LinuxTM, FreeBSDTM, etc.
  • operating systems such as Windows ServerTM, Mac OS XTM , UnixTM, LinuxTM, FreeBSDTM, etc.
  • the central processing unit 601 may execute the operations performed by the computer device in the foregoing embodiment shown in FIG. 1 , and the details will not be repeated here.
  • the present invention also provides a computer-readable storage medium, the computer-readable storage medium is used to realize the function of magnetic resonance imaging, and a computer program is stored thereon.
  • the computer program is executed by a processor, the processor can be used to execute MRI method as described in Figure 1.
  • the integrated unit can be stored in a corresponding computer-readable storage medium or integrated into a computer program product so as to execute the above-mentioned method.
  • the present invention realizes all or part of the processes in the methods of the above corresponding embodiments, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, When the computer program is executed by the processor, it can realize the steps of the above-mentioned various method embodiments.
  • the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form.
  • the computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM, Read-Only Memory) , Random Access Memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.
  • ROM Read-Only Memory
  • RAM Random Access Memory
  • electrical carrier signal telecommunication signal and software distribution medium, etc.
  • the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present invention.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .
  • the disclosed system, device and method can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

Landscapes

  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • General Physics & Mathematics (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

一种磁共振成像方法，包括：获取目标患者的第一磁共振图像（101），第一磁共振图像为依据目标患者的待检测区域的部分K空间数据获得；将第一磁共振图像输入预先训练好的图像处理神经网络模型（102），预先训练好的图像处理神经网络模型由低质量磁共振成像及对应的高质量磁共振成像的训练集训练获得；获得图像处理神经网络模型输出的与第一磁共振图像对应的第二磁共振图像（103）。使用图像处理神经网络模型进行处理所获得的第二磁共振图像与第一磁共振图像相比，更接近使用全部K空间数据所获得的成像，因此使用第二磁共振图像可以获得更多的患者器官组织信息，提高了磁共振成像所获得的成像的分辨率，提高了信噪比。

Description

一种磁共振成像方法及相关设备
本申请要求于2021年6月8日提交中国专利局、申请号为202110638201.4、发明名称为“一种磁共振成像方法及相关设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及图像处理领域，尤其涉及一种磁共振成像方法及相关设备。
背景技术
磁共振成像(Magnetic Resonance Imaging,MRI)是一种利用原子核在强磁场内发生共振产生的信号经图像重建的一种成像技术，是一种核物理现象。它是利用射频脉冲对置于磁场中含有自旋不为零的原子核进行激励，射频脉冲停止后，原子核进行弛豫，在其弛豫过程中用感应线圈采集信号，按一定的数学方法重建形成数学图像。磁共振成像技术不同于其他成像技术，它提供的信息量远远大于医学影像学中的其他许多成像技术。因此，对疾病的诊断具有很大的明显优越性。
在磁共振成像成像过程中，所采集的信号为频域的K空间数据，经傅立叶变换获得病人的结构图像信息。在信号的采集和处理过程需要消耗较多的时间。近年来，多种仅使用较少K空间数据即可得出磁共振成像的算法被不断提出。
但现有的仅使用较少K空间数据得出磁共振成像的算法在提高成像速度的同时，降低了图像质量。生成图像存在噪声大，分辨率低的问题。
发明内容
本申请实施例第一方面提供了一种磁共振成像方法，包括：
获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，所述预 先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或者部分K空间数据获得的磁共振成像；
获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
基于本申请实施例第一方面提供的磁共振成像方法，可选的，所述方法还包括：
获取所述目标患者的待检测区域的部分K空间数据；
基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像。
基于本申请实施例第一方面提供的磁共振成像方法，可选的，所述基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像包括：
基于灵敏度编码算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
基于压缩感知技术对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
基于同时多层成像算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
基于广义自动校准部分并行采集算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像。
基于本申请实施例第一方面提供的磁共振成像方法，可选的，所述获取目标患者的第一磁共振图像之前，所述方法还包括：
获取低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像组成的训练集；
使用所述训练集训练所述图像处理神经网络模型，获得训练好的图像处理神经网络模型。
基于本申请实施例第一方面提供的磁共振成像方法，可选的，所述低质量磁共振成像为：
基于患者的特定区域的部分K空间数据进行灵敏度编码算法获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行压缩感知技术处理获得的磁共振成像，或基于同时多层成像算法处理获得的磁共振图像，或基于患者的特定区域的部分K空间数据进行广义自动校准部分并行采集算法获得的磁共振成像中的任意一种或多种。
基于本申请实施例第一方面提供的磁共振成像方法，可选的，所述图像处理神经网络模型为残差卷积神经网络；
所述图像处理神经网络模型可以为残差稠密网络或者具有编码器-解码器结构的网络模型；其中所述残差稠密网络包括至少一个稠密块，所述稠密块由卷积层和非线性激活层构成的至少两个基本模块通过稠密连接方式进行连接。
本申请实施例第二方面提供了一种磁共振成像设备，包括：
获取单元，用于获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
输入单元，用于将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或者部分K空间数据获得的磁共振成像；
获得单元，用于获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
本申请实施例第三方面提供了一种计算机设备，包括：
中央处理器，存储器，输入输出接口，有线或无线网络接口以及电源；
所述存储器为短暂存储存储器或持久存储存储器；
所述中央处理器配置为与所述存储器通信，在所述计算机设备上执行所述存储器中的指令操作以执行本申请实施例第一方面中任意一项所述的方法。
本申请实施例第四方面提供了一种计算机可读存储介质，包括指令，当所述指令在计算机上运行时，使得计算机执行如本申请实施例第一方面中任意一项所述的方法。
本申请实施例第五方面提供了一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行如本申请实施例第一方面中任意一项所述的方法。
从以上技术方案可以看出，本申请实施例具有以下优点：本方案通过图像处理神经网络模型对使用部分K空间数据获得的第一磁共振图像进行处理，其中图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，因此使用图像处理神经网络模型进行处理所获得的第二磁共振图像与第一磁共振图像相比，更接近使用全部或者部分K空间数据所获得的成像，因此使用第二磁共振图像可以获得更多的患者器官组织信息，提高了磁共振成像所获得的成像的分辨率，提高了信噪比。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据提供的附图获得其他的附图。
图1为本申请磁共振成像方法实施例的一个流程示意图；
图2为本申请磁共振成像方实施例的另一个流程示意图；
图3为本申请图像处理神经网络模型训练及使用过程的一个流程示意图；
图4为本申请图像处理神经网络模型实施例的一个结构示意图；
图5为本申请磁共振成像设备实施例的一个结构示意图；
图6为本申请计算机设备实施例的一个结构示意图；
图7A及7B为本申请提供的磁共振图像及磁共振图像输入模型进行处理后的效果示例图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申 请保护的范围。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
需要说明的是，在本申请中涉及“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本申请要求的保护范围之内。
请参阅图1，本申请的一个实施例包括：步骤101-步骤105。
核磁共振成像技术是一种多参数、多对比度的成像技术，基于核磁共振成像可以为医生提供丰富的病人器官组织信息。核磁共振成像所采集的信号为频域的K空间(寻常空间在傅利叶转换下的对偶空间)数据，经傅立叶变换获得病人的结构图像信息。核磁共振成像的质量受信噪比(Signal to Noise Ratio,SNR)和分辨率(Resolution)影响。高信噪比和分辨率的图像可更清晰呈现身体内微小结构，有助于医生对病变部位进行准确的鉴别诊断。
然而磁共振成像在临床使用时，受限于人体组织的物理参数和成像算法的扫描参数，存在扫描时间长，检查效率低的问题，提高了临床使用成本。
因此，一些使用部分K空间数据进行成像的方法被开发出来，如并行采集(Parallel Imaging)中的灵敏度编码算法(SENSitivity Encoding，SENSE)或广义自动校准部分并行采集算法(Generalized autocalibrating partially parallel acquisitions，GRAPPA)，或同时 多层成像(Simultaneous multi-slice，SMS)算法在扫描部分k空间序列的情况下，利用不同线圈对不同区域磁场信号的权重不同，通过存在叠影的磁共振图像还原出不存在叠影的图像。又比如更新一代的压缩感知技术(compressed sensing，CS)可以利用部分k空间数据，通过对比生成图像、k空间和已知k空间数据的差异，迭代回归，得到成像。
然而上述利用部分K空间数据使用算法进行成像的方式，均在不同程度上降低了所得到的的磁共振成像的质量，为后续使用带来一定问题。
为解决上述问题，本申请提供了一种磁共振成像方法，请参照图1，本方法包括步骤101至步骤103。
101、获取目标患者的第一磁共振图像。
获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
第一磁共振图像为基于目标患者的部分K空间数据进行数据处理获得，具体的所采用的处理方法可为基于灵敏度编码算法、获得基于压缩感知技术获得或基于广义自动校准部分并行采集算法或基于同时多层成像(Simultaneous Multi-Slice)算法获得，或其他基于上述多种方式的改进方法获得，具体此处不做限定。第一磁共振图像为基于部分K空间数据生成，在生成成像时所需使用的K空间数据基于所采用的生成方式而定，对于采用不同生成算法的情况可按照不同需求采集K空间数据，具体此处不做限定。所得磁共振图像包括但不限于T1、T2、磁共振血管造影(MRA)等类型的加权像。
102、将所述第一磁共振图像输入预先训练好的图像处理神经网络模型。
将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或部分K空间数据获得的磁共振成像；其中特定区域可以是根据成像需求预先设置的任意区域，可以是一个区域也可以是多个区域。步骤101中的待检测区域为特定区域中的任意一个或多个，或者特定区域的任一部分。
需要说明的是，在将第一磁共振图像输入图像处理神经网络模型之前，还可以对第一磁共振图像进行图像预处理操作，图像预处理操作包括图像扩增、图像配准和图像归一化中的任意一项或多项。图像扩增可以具体是对第一磁共振图像进行缩放、旋转、像素值同比增减、镜像、裁剪或加入噪声等处理；图像配准的方法可以具体是图像刚性配准、基于关键点的配准或基于神经网络模型的图像配准等，通过图像配准方法实现对第一磁共振图像的配准操作。图像归一化可以具体是基于像素最大最小值的归一化、基于像素均值的归一化等方法。
图像处理神经网络模型可为卷积神经网络模型架构，并采用训练集训练完成，训练集中包括多张低质量磁共振成像及多张低质量磁共振成像对应的高质量磁共振成像。即使用部分K空间数据获得的磁共振成像和使用全部K空间数据获得的磁共振成像。
使用上述训练集对图像处理神经网络模型进行训练，使得图像处理神经网络模型分析使用部分K空间数据获得的磁共振成像和使用全部K空间数据获得的磁共振成像之间的差别，并采集二者的高维度特征，进行学习，以使得训练好的图像处理神经网络模型，可以基于低质量磁共振成像输出与其对应的高质量磁共振成像尽可能相似的图像。
103、获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。第二磁共振图像为图像处理神经网络模型依据第一磁共振图像分析获得，在生成第二磁共振图像前，图像处理神经网络模型已学习了低质量磁共振成像与高质量磁共振成像所具有的特征的对应关系，因此可基于第一磁共振图像输出与其对应的使用全部K空间数据所形成的磁共振成像尽可能相似的成像。请参阅图7A及7B，其为对第一磁共振图像输入模型处理后得到的第二磁共振图像的效果示意图。在图7A和图7B中，第一幅图像为采用全部K空间数据得到的高质量磁共振图像，第二幅图像为采用部分K空间数据得到的第一磁共振图像，第三幅图像为经过模型处理的第二磁共振图像。从图示中可以看出，第二磁共振图像相比第一磁共振图像的分辨率和信噪比均更 高，且和高质量磁共振图像有很好的一致性。
从以上技术方案可以看出，本申请实施例具有以下优点：本方案通过图像处理神经网络模型对使用部分K空间数据获得的第一磁共振图像进行处理，其中图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，因此使用图像处理神经网络模型进行处理所获得的第二磁共振图像与第一磁共振图像相比，更接近使用全部K空间数据所获得的成像，因此使用第二磁共振图像可以获得更多的患者器官组织信息，提高了磁共振成像所获得的成像的分辨率，提高了信噪比。
基于图1所描述的实施例，下面提供一种本方案在实施过程中可选择执行的详细实施例，请参阅图2，本申请磁共振成像方法的一个实施例包括：步骤201-步骤205。
201、获取所述目标患者的待检测区域的部分K空间数据。
获取所述目标患者的待检测区域的部分K空间数据。其中待检测区域为医生设置，依据对目标患者的初步诊断而定，本方案获取对目标患者的磁共振成像时只需使用目标患者的待检测区域的部分K空间数据，进而缩减了K空间数据的采集过程中所需消耗的时间，提高了本方案的可实施性。
在K空间数据的采集过程中，所采集的K空间数据要求基于后续所使用的处理算法不同而存在区分，具体可依据实际情况而定，此处不做限定。
202、基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像。
基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像。具体的对部分K空间数据进行处理，所采用的处理手段可包括下述几种方式中的任意一种或多种：
(1)基于灵敏度编码算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像。
灵敏度编码算法(SENSE)利用并行采集成像(parallel imaging)技术缩短扫描时间，广泛地被应用于磁共振各种成像序列，是一种成熟的利用部分K空间数据进行成像的方法。
(2)基于压缩感知技术对所述目标患者的待检测区域的部分K空间数据 进行处理，获得第一磁共振图像。
压缩感知技术(compressed sensing，CS)可以利用部分k空间数据，通过对比生成图像、k空间和已知k空间数据的差异，迭代回归，得到成像。
(3)基于广义自动校准部分并行采集算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像。
广义自动校准部分并行采集算法(Generalized autocalibrating partially parallel acquisitions，GRAPPA)同样是利用并行采集成像(parallel imaging)技术缩短扫描时间，并被成熟应用的磁共振成像方法。
(4)基于同时多层(Simultaneous multi-slice，SMS)成像技术所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像。
该种成像技术是利用多通道相控阵线圈和高加速因子并行采集技术，同步并行激励和同时采集多层影像，使用多通道线圈接收到的组织不同部位信号的信号强度差异，在影像域对同时采集的多个层面的影像进行分离，实现一次射频激励获得多层影像的磁共振成像技术。
可以理解的是，在本方案实际实施过程中所采用的第一磁共振图像的生成方法可依据需求进行调整，或采用其他改进算法，具体可依据实际情况而定，此处不做限定。
203、获取目标患者的第一磁共振图像。
204、将所述第一磁共振图像输入预先训练好的图像处理神经网络模型。
205、获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
上述步骤203至步骤205与前述图1对应实施例中步骤101至步骤103类似，此处不做赘述。
从以上技术方案可以看出，本申请实施例具有以下优点：本方案使用目标患者的部分K空间数据进行成像，并使用图像处理神经网络模型对获得的第一磁共振图像进行处理，其中图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，因此使用图像处理神经网络模型进行处理所获得的第二磁共振图像与第一磁共振图 像相比，更接近使用全部K空间数据所获得的成像，因此使用第二磁共振图像可以获得更多的患者器官组织信息，提高了磁共振成像所获得的成像的分辨率，提高了信噪比。同时仅使用目标患者的部分K空间数据完成该过程缩短了对患者的检查时间，避免了K空间采集时间过长所可能造成的运动伪影问题，提高了检查过程的舒适度，提高了本方案的可实施性。相比于传统磁共振成像方法，本方法能在保留图像质量的情况下大幅度减小成像时间。
下面对本方案中所使用的图像处理神经网络模型的训练及使用过程进行说明，请参照图3，包括步骤301-步骤305。
301、获取低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像组成的训练集。
获取低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像组成的训练集，低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或部分K空间数据获得的磁共振成像；磁共振训练集包括但不限于T1,T2,MRA等成像序列。其中所述低质量磁共振成像可为：基于患者的特定区域的部分K空间数据进行灵敏度编码算法获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行压缩感知技术处理获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行广义自动校准部分并行采集算法获得的磁共振成像中的任意一种或多种。
可以理解的是，为保证训练获得的图像处理神经网络模型的性能，训练集所包括的低质量磁共振成像的获得方法应与后续输入图像处理神经网络模型中的第一磁共振图像的获得方法一致，进而保证图像处理神经网络模型的处理效果。
302、使用所述训练集训练所述图像处理神经网络模型，获得训练好的图像处理神经网络模型。
本步骤使用的图像处理神经网络模型可以为所有类型神经网络中的任意一种，经试验，残差连接的残差稠密网络(residual in residual dense block,RRDB)具有较佳结果表现。具体的，其结构可参照图4。该网络由若干个不同权重(β)稠密块(dense block,DB)残差连接(图示中加号表示残差连接) 形成，并且如最右一个加号所示在输入层和输出层之间使用残差连接。如图虚线所示，每一稠密块由若干卷积层和非线性激活层构成的基本模块稠密连接得到。稠密连接指每一级基本模块的输出由之前所有基本模块的输出残差连接形成。需要说明的是，稠密块的数量以及每个稠密块内包含的基本模块的数量并不局限于图4，可以是根据实际需求而设置的其他值。
图4所使用残差稠密网络RRDB网络为仅为可实施本方案的残差卷积网络结构的一个示例，在实际实施过程中，可对残差卷积网络结构的深度(例如稠密块的数量)及结构(例如非线性激活层的具体实现结构)进行适应性调整，具体此处不做限定。
可以理解的是，图像处理神经网络模型的训练过程需要使用损失函数进行训练控制，损失函数最小化时对应的图像处理神经网络模型为训练好的图像处理神经网络模型。其中，损失函数可以包括均值绝对误差、均值平方差误差、结构相似性误差、对抗生成误差的一种或多种的组合；最小化损失函数的方法可以包括随机梯度优化、ADAM(自适应运动估计算法)优化等方法中的任意一种。
303、获取目标患者的第一磁共振图像。
304、将所述第一磁共振图像输入预先训练好的图像处理神经网络模型。
305、获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
上述步骤303至步骤305与前述图1对应实施例中步骤101至步骤103类似，此处不做赘述。
使用301步骤所述的训练集训练具有上述结构的图像处理神经网络模型，以便后续使用。
上面对申请实施例中的磁共振成像方法进行了描述，下面对本申请实施例中的磁共振成像设备进行描述。请参阅图5，本申请磁共振成像设备的一个实施例包括：
获取单元501，用于获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
输入单元502，用于将所述第一磁共振图像输入预先训练好的图像处理神 经网络模型，所述预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或部分K空间数据获得的磁共振成像；
获得单元503，用于获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
本实施例中，磁共振成像设备中各单元所执行的流程与前述图1所对应的实施例中描述的方法流程类似，此处不再赘述。
可选的，所述获取设备还用于：获取所述目标患者的待检测区域的部分K空间数据；
基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像。
可选的，所述获取设备具体用于：
基于灵敏度编码算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
基于压缩感知技术对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
基于广义自动校准部分并行采集算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，中的任意一种。
可选的，所述设备还包括：训练单元，用于获取低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像组成的训练集；
使用所述训练集训练所述图像处理神经网络模型，获得训练好的图像处理神经网络模型。
可选的，所述低质量磁共振成像为：
基于患者的特定区域的部分K空间数据进行灵敏度编码算法获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行压缩感知技术处理获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行广义自动校准部分并行采集算法获得的磁共振成像中的任意一种或多种。
可选的，所述图像处理神经网络模型为残差卷积神经网络；
所述图像处理神经网络模型为残差稠密网络或其他网络模型；其中所述残差稠密网络包括至少一个稠密块，所述稠密块由卷积层和非线性激活层构成的至少两个基本模块通过稠密连接方式进行连接。
神经网络训练模块采用监督学习方式。以低质量的磁共振图像为输入，以高质量的磁共振图像为标签。训练使用的损失函数包括但不限于L1 loss，L2 loss，SSIM loss，GAN loss等并可以为其中一种或多种的组合。
图6是本申请实施例提供的一种计算机设备的结构示意图，该计算机设备600可以包括一个或一个以上中央处理器(central processing units，CPU)601和存储器605，该存储器605中存储有一个或一个以上的应用程序或数据。
本实施例中，中央处理器601中的具体功能模块划分可以与前述图5中所描述的各单元的功能模块划分方式类似，此处不再赘述。
其中，存储器605可以是易失性存储或持久存储。存储在存储器605的程序可以包括一个或一个以上模块，每个模块可以包括对服务器中的一系列指令操作。更进一步地，中央处理器601可以设置为与存储器605通信，在服务器600上执行存储器605中的一系列指令操作。
服务器600还可以包括一个或一个以上电源602，一个或一个以上有线或无线网络接口603，一个或一个以上输入输出接口604，和/或，一个或一个以上操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM,LinuxTM，FreeBSDTM等。
该中央处理器601可以执行前述图1所示实施例中计算机设备所执行的操作，具体此处不再赘述。
本发明还提供了一种计算机可读存储介质，该计算机可读存储介质用于实现磁共振成像的功能，其上存储有计算机程序，计算机程序被处理器执行时，处理器，可以用于执行如图1所述的磁共振成像方法。
可以理解的是，所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在相应的一个计算机可读取存储介质中或集成为计算机程序产品以便执行上述方法。基于这样的理解，本发明实现上述相应的实施例方法中的全部或部分流程，也可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一计算机可读存储介质中，该计算机程 序在被处理器执行时，可实现上述各个方法实施例的步骤。其中，所述计算机程序包括计算机程序代码，所述计算机程序代码可以为源代码形式、对象代码形式、可执行文件或某些中间形式等。所述计算机可读介质可以包括：能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、电载波信号、电信信号以及软件分发介质等。需要说明的是，所述计算机可读介质包含的内容可以根据司法管辖区内立法和专利实践的要求进行适当的增减，例如在某些司法管辖区，根据立法和专利实践，计算机可读介质不包括电载波信号和电信信号。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为 单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
以上所述，以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (10)

  1. 一种磁共振成像方法，其特征在于，包括：
    获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
    将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由使用包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集对图像处理神经网络模型训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患者的特定区域的全部或部分K空间数据获得的磁共振成像；
    获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
  2. 根据权利要求1所述的磁共振成像方法，其特征在于，所述方法还包括：
    获取所述目标患者的待检测区域的部分K空间数据；
    基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像。
  3. 根据权利要求2所述的磁共振成像方法，其特征在于，所述基于预设算法对所述部分K空间数据进行处理，获得第一磁共振图像包括：
    基于灵敏度编码算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
    基于压缩感知技术对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
    基于广义自动校准部分并行采集算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像，或；
    基于同时多层成像算法对所述目标患者的待检测区域的部分K空间数据进行处理，获得第一磁共振图像。
  4. 根据权利要求1所述的磁共振成像方法，其特征在于，所述获取目标患者的第一磁共振图像之前，所述方法还包括：
    获取低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像组成的训练集；
    使用所述训练集训练所述图像处理神经网络模型，获得训练好的图像处理神经网络模型。
  5. 根据权利要求1所述的磁共振成像方法，其特征在于，所述低质量磁共振成像为：
    基于患者的特定区域的部分K空间数据进行灵敏度编码算法获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行压缩感知技术处理获得的磁共振成像，或基于患者的特定区域的部分K空间数据进行校准部分并行采集算法，或基于患者的特定区域的部分K空间数据进行同时多层成像算法获得的磁共振成像中的任意一种或多种。
  6. 根据权利要求1所述的磁共振成像方法，其特征在于，所述图像处理神经网络模型为残差稠密网络；其中所述残差稠密网络包括至少一个稠密块，所述稠密块由卷积层和非线性激活层构成的至少两个基本模块通过稠密连接方式进行连接。
  7. 根据据权利要求1所述的磁共振成像方法，其特征在于，将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，包括：
    将所述第一磁共振图像进行图像预处理操作后输入预先训练好的图像处理神经网络模型；其中所述图像预处理操作包括图像扩增、图像配准和图像归一化中的任意一项或多项。
  8. 一种磁共振成像设备，其特征在于，包括：
    获取单元，用于获取目标患者的第一磁共振图像，所述第一磁共振图像为依据所述目标患者的待检测区域的部分K空间数据获得；
    输入单元，用于将所述第一磁共振图像输入预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型，所述预先训练好的图像处理神经网络模型由包括低质量磁共振成像及所述低质量磁共振成像对应的高质量磁共振成像的训练集训练获得，所述低质量磁共振成像为基于患者的特定区域的部分K空间数据获得的磁共振成像，所述高质量磁共振成像为基于患 者的特定区域的全部或部分K空间数据获得的磁共振成像；
    获得单元，用于获得所述图像处理神经网络模型输出的与所述第一磁共振图像对应的第二磁共振图像。
  9. 一种计算机设备，其特征在于，包括：
    中央处理器，存储器，输入输出接口，有线或无线网络接口以及电源；
    所述存储器为短暂存储存储器或持久存储存储器；
    所述中央处理器配置为与所述存储器通信，在所述计算机设备上执行所述存储器中的指令操作以执行权利要求1-7中任意一项所述的方法。
  10. 一种计算机可读存储介质，包括指令，当所述指令在计算机上运行时，使得计算机执行如权利要求1-7中任意一项所述的方法。
PCT/CN2021/099570 2021-06-08 2021-06-11 一种磁共振成像方法及相关设备 WO2022257090A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110638201.4 2021-06-08
CN202110638201.4A CN113359077A (zh) 2021-06-08 2021-06-08 一种磁共振成像方法及相关设备

Publications (1)

Publication Number Publication Date
WO2022257090A1 true WO2022257090A1 (zh) 2022-12-15

Family

ID=77533222

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/099570 WO2022257090A1 (zh) 2021-06-08 2021-06-11 一种磁共振成像方法及相关设备

Country Status (2)

Country Link
CN (1) CN113359077A (zh)
WO (1) WO2022257090A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114114116B (zh) * 2022-01-27 2022-08-23 南昌大学 磁共振成像生成方法、系统、存储介质及计算机设备
CN114241077B (zh) * 2022-02-23 2022-07-15 南昌睿度医疗科技有限公司 一种ct图像分辨率优化方法及装置
CN114241078A (zh) * 2022-02-23 2022-03-25 南昌睿度医疗科技有限公司 一种加速磁共振超分辨率成像方法、装置、设备及介质
CN114418900B (zh) * 2022-03-30 2022-07-01 南昌睿度医疗科技有限公司 一种磁共振图像降噪方法、装置、设备及存储介质
CN114419183B (zh) * 2022-03-31 2022-07-01 南昌睿度医疗科技有限公司 一种mra加速图像的优化方法、系统、设备及存储介质
CN116721316A (zh) * 2023-08-11 2023-09-08 之江实验室 一种模型训练和地磁图优化方法、装置、介质及设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190347772A1 (en) * 2018-04-19 2019-11-14 Subtle Medical, Inc. Systems and methods for improving magnetic resonance imaging using deep learning
CN111870245A (zh) * 2020-07-02 2020-11-03 西安交通大学 一种跨对比度引导的超快速核磁共振成像深度学习方法
CN111899177A (zh) * 2020-08-05 2020-11-06 苏州深透智能科技有限公司 一种图像处理方法、装置、电子设备和可读存储介质
CN111899165A (zh) * 2020-06-16 2020-11-06 厦门大学 一种基于功能模块的多任务图像重建卷积网络模型

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017113205A1 (zh) * 2015-12-30 2017-07-06 中国科学院深圳先进技术研究院 一种基于深度卷积神经网络的快速磁共振成像方法及装置
CN108957375B (zh) * 2018-07-23 2020-12-22 深圳先进技术研究院 磁共振成像方法、装置、设备及存储介质
CN109493394A (zh) * 2018-10-26 2019-03-19 上海东软医疗科技有限公司 磁共振成像获得深度学习训练集的方法、重建方法和装置
CN109658469B (zh) * 2018-12-13 2023-05-26 深圳先进技术研究院 一种基于深度先验学习的头颈联合成像方法和装置
CN109712208B (zh) * 2018-12-13 2023-01-24 深圳先进技术研究院 基于深度学习的大视野磁共振扫描图像重建方法和装置
CN109712119B (zh) * 2018-12-13 2020-07-03 深圳先进技术研究院 一种磁共振成像及斑块识别方法和装置
CN109801259A (zh) * 2018-12-18 2019-05-24 中国科学院深圳先进技术研究院 一种核磁共振图像的快速成像方法、装置及设备
CN110288672A (zh) * 2019-06-28 2019-09-27 闽江学院 一种基于超深稠密网络的压缩感知磁共振图像重建方法
CN112213673B (zh) * 2020-09-07 2022-11-22 上海东软医疗科技有限公司 动态磁共振成像方法、装置、重建计算机及磁共振系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190347772A1 (en) * 2018-04-19 2019-11-14 Subtle Medical, Inc. Systems and methods for improving magnetic resonance imaging using deep learning
CN111989039A (zh) * 2018-04-19 2020-11-24 深透医疗公司 使用深度学习改进磁共振成像的系统和方法
CN111899165A (zh) * 2020-06-16 2020-11-06 厦门大学 一种基于功能模块的多任务图像重建卷积网络模型
CN111870245A (zh) * 2020-07-02 2020-11-03 西安交通大学 一种跨对比度引导的超快速核磁共振成像深度学习方法
CN111899177A (zh) * 2020-08-05 2020-11-06 苏州深透智能科技有限公司 一种图像处理方法、装置、电子设备和可读存储介质

Also Published As

Publication number Publication date
CN113359077A (zh) 2021-09-07

Similar Documents

Publication Publication Date Title
WO2022257090A1 (zh) 一种磁共振成像方法及相关设备
Souza et al. Dual-domain cascade of U-nets for multi-channel magnetic resonance image reconstruction
US11023785B2 (en) Sparse MRI data collection and classification using machine learning
Chaudhari et al. Utility of deep learning super‐resolution in the context of osteoarthritis MRI biomarkers
CN108717717B (zh) 基于卷积神经网络和迭代方法相结合的稀疏mri重建的方法
WO2020219915A1 (en) Denoising magnetic resonance images using unsupervised deep convolutional neural networks
Sandino et al. Deep convolutional neural networks for accelerated dynamic magnetic resonance imaging
US10878311B2 (en) Image quality-guided magnetic resonance imaging configuration
US11874359B2 (en) Fast diffusion tensor MRI using deep learning
US11170543B2 (en) MRI image reconstruction from undersampled data using adversarially trained generative neural network
CN110992440B (zh) 弱监督磁共振快速成像方法和装置
WO2020077202A1 (en) Medical image segmentation using deep learning models trained with random dropout and/or standardized inputs
Lv et al. Which GAN? A comparative study of generative adversarial network-based fast MRI reconstruction
Knoll et al. Deep learning methods for parallel magnetic resonance image reconstruction
US20210239780A1 (en) Estimating diffusion metrics from diffusion-weighted magnetic resonance images using optimized k-q space sampling and deep learning
CN108287324A (zh) 磁共振多对比度图像的重建方法和装置
Arshad et al. Transfer learning in deep neural network based under-sampled MR image reconstruction
Pezzotti et al. An adaptive intelligence algorithm for undersampled knee mri reconstruction: Application to the 2019 fastmri challenge
Cui et al. Motion artifact reduction for magnetic resonance imaging with deep learning and k-space analysis
Chen et al. Compressed sensing in sodium magnetic resonance imaging: techniques, applications, and future prospects
Slavkova et al. An untrained deep learning method for reconstructing dynamic MR images from accelerated model‐based data
US11280868B2 (en) Image enhancement with variable number of excitation (NEX) acquisitions accelerated using compressed sensing
US20230380714A1 (en) Method and system for low-field mri denoising with a deep complex-valued convolutional neural network
CN106137199B (zh) 扩散磁共振成像中的广义球面去卷积
Liu et al. Accelerated submillimeter wave‐encoded magnetic resonance imaging via deep untrained neural network

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21944599

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21944599

Country of ref document: EP

Kind code of ref document: A1