WO2024103414A1 - 一种磁共振图像重建方法和图像重建装置 - Google Patents

一种磁共振图像重建方法和图像重建装置 Download PDF

Info

Publication number
WO2024103414A1
WO2024103414A1 PCT/CN2022/133016 CN2022133016W WO2024103414A1 WO 2024103414 A1 WO2024103414 A1 WO 2024103414A1 CN 2022133016 W CN2022133016 W CN 2022133016W WO 2024103414 A1 WO2024103414 A1 WO 2024103414A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
magnetic resonance
signal
magnetic field
sensitivity
Prior art date
Application number
PCT/CN2022/133016
Other languages
English (en)
French (fr)
Inventor
郑海荣
梁栋
王海峰
崔卓须
刘聪聪
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2024103414A1 publication Critical patent/WO2024103414A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the present invention relates to the technical field of magnetic resonance imaging, and in particular to a magnetic resonance image reconstruction method and an image reconstruction device.
  • Magnetic resonance imaging has a slow scanning speed. Long scanning time can cause discomfort to patients and easily introduce motion artifacts into the image, thus affecting the image quality.
  • Magnetic resonance parallel imaging methods are a type of method that accelerates MRI scanning speed, such as sensitivity encoding (SENSE) and generalized autocalibrating partially parallel acquisitions (GRAPPA). This type of method achieves the purpose of fast scanning by reducing the amount of collected data and reconstructing the undersampled data using the redundant information contained in the multi-channel coil.
  • SENSE sensitivity encoding
  • GRAPPA generalized autocalibrating partially parallel acquisitions
  • Magnetic resonance imaging includes wave gradient field coded parallel imaging technology, wave coded imaging technology, and virtual conjugate coil (VCC) imaging technology.
  • VCC virtual conjugate coil
  • wave encoding is a parallel imaging technology used to speed up magnetic resonance scanning. It utilizes the more efficient spatial encoding characteristics of multi-channel coils.
  • Virtual Conjugate Coil is a parallel imaging technology with similar effects to wave encoding, which can provide more channels of spatial encoding prior information.
  • the combination of Wave-VCC can give full play to the characteristics of both and provide a higher acceleration technology.
  • the conventional Wave-VCC reconstruction method only uses the low-frequency ACS signal in the middle of k-space to estimate the background phase. Due to the lack of surrounding high-frequency information, the estimated background phase is difficult to characterize the image of high-frequency phase changes.
  • Wave encoding technology is a parallel imaging technology used to speed up 3D MRI scanning. This technology uses MRI gradient coils to apply phase differences in the layer selection and phase directions while acquiring MRI signals (while applying the readout gradient field), and uses MRI gradient coils to apply phase differences in the phase encoding direction.
  • the sinusoidal gradient field is used, and the fast parallel imaging technology with controlled aliasing (2D CAIPIRINHA, two-dimension Controlled Aliasing In Parallel Imaging Results In Higher Acceleration) is used to under-sample the data, so that the aliasing artifacts caused by undersampling are dispersed along the readout, layer selection and phase directions, reducing the degree of image aliasing artifacts in each pixel, thereby greatly reducing the signal-to-noise ratio loss of the geometry factor (g-factor, geometry factor) in parallel imaging reconstruction, achieving the purpose of high acceleration.
  • 2D CAIPIRINHA two-dimension Controlled Aliasing In Parallel Imaging Results In Higher Acceleration
  • VCC Virtual conjugate coils
  • the existing technology requires collecting low-frequency auto-calibration signals (ACS) signals, and then calculating the high-frequency coil sensitivity based on the low-frequency ACS signals.
  • ACS auto-calibration signals
  • memory introduces motion errors, which leads to large errors in the calculated coil sensitivity, and further causes poor image quality reconstructed based on the coil sensitivity.
  • the present invention provides a magnetic resonance image reconstruction method and an image reconstruction device, which solve the problem that the reconstructed image calculated by the prior art has poor quality.
  • the present invention adopts the following technical solutions:
  • the present invention provides a magnetic resonance image reconstruction method, comprising:
  • the magnetic field phase difference formed by each magnetic field is obtained
  • a magnetic resonance image of the target object is reconstructed according to the underlying image, the coil sensitivity, the conjugate sensitivity, the background phase, the sampling template signal and the magnetic field phase difference.
  • the image structure information of the depth image is applied with a network structure composed of several neural networks to obtain the bottom layer image, background phase, coil sensitivity of the target object to the magnetic field coil, and conjugate sensitivity of the coil sensitivity output by the network structure, wherein the depth image is used to characterize the depth information of the target object relative to the magnetic resonance device, and the magnetic field coil is a coil inside the magnetic resonance device, including:
  • sampling the signal from the target object received by the magnetic resonance device to generate a sampling template signal includes:
  • the signal from the target object received by the magnetic resonance device is sampled using a three-dimensional MRI sequence to obtain a sampling signal in a layer selection direction and a sampling signal in a phase direction;
  • a sinusoidal gradient field is applied in the layer selection direction and a truncated sinusoidal gradient field is applied in the phase direction; or, a truncated sinusoidal gradient field is applied in the layer selection direction and a sinusoidal gradient field is applied in the phase direction.
  • obtaining the magnetic field phase difference formed by each magnetic field based on each magnetic field information applied in the sampling environment includes:
  • a magnetic resonance image of the target object is reconstructed according to the target signal and the conjugate symmetric signal.
  • an embodiment of the present invention further provides a magnetic resonance image reconstruction device, wherein the device comprises the following components:
  • an information analysis module used for applying a network structure composed of a plurality of neural networks to the image structure information of the depth image, and obtaining an underlying image, a background phase, a coil sensitivity of the target object to the magnetic field coil, and a conjugate sensitivity of the coil sensitivity output by the network structure, wherein the depth image is used to characterize the depth information of the target object relative to the magnetic resonance device, and the magnetic field coil is a coil inside the magnetic resonance device;
  • a signal adopting module used for sampling the signal from the target object received by the magnetic resonance device to generate a sampling template signal
  • a phase difference calculation module is used to obtain the magnetic field phase difference formed by each magnetic field according to each magnetic field information applied in the sampling environment;
  • An image reconstruction module is used to reconstruct a magnetic resonance image of the target object based on the underlying image, the coil sensitivity, the conjugate sensitivity, the background phase, the sampling template signal and the magnetic field phase difference.
  • an embodiment of the present invention further provides a terminal device, wherein the terminal device comprises a memory, a processor, and a magnetic resonance image reconstruction program stored in the memory and executable on the processor, and when the processor executes the magnetic resonance image reconstruction program, the steps of the magnetic resonance image reconstruction method described above are implemented.
  • an embodiment of the present invention further provides a computer-readable storage medium, on which a magnetic resonance image reconstruction program is stored, and when the magnetic resonance image reconstruction program is executed by a processor, the steps of the magnetic resonance image reconstruction method described above are implemented.
  • the underlying image, background phase, coil sensitivity, and conjugate sensitivity required for image reconstruction by the present invention are all derived from a neural network. Since the present invention does not involve low-frequency ACS signals in the process of calculating coil sensitivity, the quality of the image reconstructed by the present invention is improved. In addition, since the underlying image of the present invention is also calculated by a neural network, the speed of image reconstruction by the present invention is improved.
  • Fig. 1 is an overall flow chart of the present invention
  • FIG2 is a structural diagram of a deep convolutional neural network in an embodiment of the present invention.
  • FIG3 is a schematic diagram of a data undersampling strategy according to an embodiment of the present invention.
  • FIG4 is a schematic diagram of a truncated sinusoidal gradient field in an embodiment of the present invention.
  • FIG5 is a schematic diagram of a sinusoidal gradient field in an embodiment of the present invention.
  • FIG6 is a schematic diagram of a truncated gradient field used in a bSSFP sequence according to an embodiment of the present invention.
  • FIG. 7 is a block diagram of the internal structure of a terminal device provided in an embodiment of the present invention.
  • the present invention provides a magnetic resonance image reconstruction method and an image reconstruction device, which solves the problem of poor quality of reconstructed images calculated by the prior art.
  • a network structure composed of several neural networks is applied to the image structure information of the depth image to obtain the underlying image, background phase, coil sensitivity of the target object to the magnetic field coil, and conjugate sensitivity of the coil sensitivity output by the network structure; the signal from the target object received by the magnetic resonance device is sampled to generate a sampling template signal; then, based on the magnetic field information applied in the sampling environment, the magnetic field phase difference formed by each magnetic field is obtained; finally, based on the underlying image, coil sensitivity, conjugate sensitivity, background phase, sampling template signal and magnetic field phase difference, the magnetic resonance image of the target object is reconstructed.
  • the present invention improves the quality of the reconstructed image.
  • the magnetic resonance device is fixed at a position, and the depth image of the patient's lesion is collected by the magnetic resonance device (used to characterize the distance between each point of the lesion and the magnetic resonance device).
  • the image information of the depth image is input into four neural networks respectively to obtain the bottom layer image, background phase, coil sensitivity (the sensitivity of the lesion to the coil inside the magnetic resonance device, that is, how much the coil change will cause the image information collected at the lesion to change), and conjugate sensitivity.
  • the magnetic resonance image of the lesion can be reconstructed by combining the bottom layer image, coil sensitivity, conjugate sensitivity, background phase, sampling template signal and magnetic field phase difference.
  • the magnetic resonance image reconstruction method of this embodiment can be applied to a terminal device, which can be a terminal product with an image acquisition function, such as a magnetic resonance device, etc.
  • a terminal device which can be a terminal product with an image acquisition function, such as a magnetic resonance device, etc.
  • the magnetic resonance image reconstruction method specifically includes the following steps:
  • image structure information is utilized.
  • This image structure information is generally represented as a sparse property of the image itself in MRI images, and this property can be represented by an optimized network structure.
  • This embodiment includes four neural networks as shown in FIG2 , which are, from top to bottom, a first deep convolutional neural network CNN k , a second deep convolutional neural network CNN k , and a second deep convolutional neural network CNN k .
  • the four deep convolutional neural networks in this embodiment have similar structures, except that the number of intermediate channels is different.
  • the Adam optimizer is used to iteratively optimize the parameters of the network. When the parameters of the network are optimized, a small noise is randomly input to generate a complete image.
  • the optimization parameters are based on the following formula:
  • the output of the network is the loss function of the deep convolutional neural network, They are the parameters of the above four deep convolutional neural networks respectively.
  • the values of the network parameters corresponding to the minimum value of the second norm of the loss function are calculated by the above formula to complete the optimization of the network parameters.
  • the image structure information of the depth image is input into the first deep convolutional neural network CNN k of the decoding structure, and the first deep convolutional neural network CNN k outputs the bottom layer image m.
  • the image structure information of the depth image is input into the second deep convolutional neural network Second Deep Convolutional Neural Network Output background phase
  • the image structure information of the depth image is input to the third deep convolutional neural network CNN C , and the coil sensitivity CSM is output by the third deep convolutional neural network CNN C .
  • the image structure information of the depth image is input to the fourth deep convolutional neural network CNN C* , and the conjugate sensitivity CSM* is output by the fourth deep convolutional neural network CNN C* .
  • the input of CNN k is CNNc
  • the input of CNN ⁇ is
  • the ACS signal containing only low-frequency phase information is not directly used to estimate the background phase. Instead, the image and its phase information are characterized according to the image degradation process and the last collected signal, so that the image and the background phase information contained therein can be generated more accurately.
  • S200 Sampling the signal from the target object received by the magnetic resonance device to generate a sampling template signal.
  • a magnetic resonance device sends an original signal to a lesion on a human body (target object).
  • the lesion on a human body interacts with the original signal, causing the original signal to become a new signal.
  • the new signal is sampled to generate a sampling template signal as shown in FIG3 .
  • Step S200 requires applying a sinusoidal gradient field and a truncated sinusoidal gradient field before sampling.
  • a sinusoidal gradient field is applied in the layer selection direction by using an MRI gradient field coil, and a truncated sinusoidal gradient field as shown in FIG4 is applied in the phase direction by using an MRI gradient field coil.
  • a truncated sinusoidal gradient field as shown in FIG4 is applied in the layer selection direction by using an MRI gradient field coil, and a sinusoidal gradient field as shown in FIG5 is applied in the phase direction by using an MRI gradient field coil.
  • the 0th order moment of the truncated sinusoidal gradient field is zero.
  • the 0th order moment (in the phase direction and the layer selection direction) of the truncated sinusoidal gradient field is zero, and the truncated sinusoidal gradient field adopts the direction shown in FIG. 4 in the phase direction.
  • This can not only effectively disperse the aliasing artifacts caused by undersampling, reduce the signal-to-noise ratio loss caused by the g-factor to achieve high-fold acceleration, but also avoid the imaging artifacts caused by the gradient field whose 0th order moment is not zero.
  • the magnetic field phase difference Psf between the sinusoidal gradient field and the truncated sinusoidal gradient field is
  • the expression of the sinusoidal gradient field is as follows:
  • A is the amplitude of the sinusoidal gradient field;
  • DC is the duration of a sinusoidal gradient field cycle (as marked in Figure 5);
  • DR is the platform duration of the readout gradient field (as marked in Figure 5).
  • step S200 After applying the above-mentioned sinusoidal gradient field and truncated sinusoidal gradient field, step S200 starts sampling signals, and step S200 includes the following steps S201 and S202:
  • the three-dimensional MRI sequence includes a GRE sequence, a SE sequence, a bSSFP sequence, and the like.
  • This embodiment uses a three-dimensional MRI sequence to undersample the signal in order to accelerate the scan.
  • the truncated gradient field is applied to the bSSFP sequence, and the signal acquisition strategy of the 2D CAIPIRINHA technology is used to reduce the amount of data acquisition.
  • the aliasing caused by undersampling can be dispersed to the readout, phase and layer selection directions at the same time, making more effective use of the background area in the FOV, increasing the sensitivity difference between different pixels, and further reducing the g-factor signal-to-noise ratio loss.
  • the truncated gradient field is applied in the bSSFP sequence in the manner shown in FIG6 , and the 2D CAIPIRINHA data acquisition strategy is shown in FIG3 .
  • the readout direction is perpendicular to both the phase direction and the layer selection direction, and the intersection of the dashed lines is the readout line required for full sampling.
  • the readout line required for the undersampling strategy adopted by the present invention is represented by a solid origin.
  • FIG3 shows 3 ⁇ 3 times undersampling (3 times undersampling in the phase direction and 3 times undersampling in the layer selection direction), the total acceleration factor is 9, and the required acquisition time is repetition time (TR, repetition time) ⁇ number of phase encoding lines (Np) ⁇ number of layer selection encoding lines (Ns)/9.
  • each magnetic field information is the magnetic field phase of the sinusoidal gradient field and the magnetic field phase of the truncated sinusoidal gradient field in step S200.
  • Step S400 includes the following steps S401 to S4011:
  • the coil sensitivity CSM corresponding to the output of the third deep convolutional neural network CNN C is the second result obtained by applying Fourier transform to the first result, For the third result, Apply inverse Fourier transform to the third result, where M is the sampled template signal in FIG3 .
  • b is replaced by bi, and Substitute bH :
  • the conjugate symmetry of Psf * is the conjugate symmetry of Psf.
  • the received signal It is the conjugate symmetry of the original signal bi .
  • it is equivalent to providing another set of phase information, that is, providing an additional set of additional encoding prior information of the receiving coil in the wave encoding framework, further providing an acceleration multiple.
  • a real magnetic resonance image of the target object can be reconstructed according to the target signal b and the conjugate symmetric signal bH .
  • Obtaining a magnetic resonance image through b and bH is a prior art.
  • the underlying image, background phase, coil sensitivity, and conjugate sensitivity required for image reconstruction in the present invention are all derived from a neural network. Since the present invention does not involve low-frequency ACS signals in the process of calculating coil sensitivity, the quality of the image reconstructed in the present invention is improved. In addition, since the underlying image of the present invention is also calculated by a neural network, the speed of image reconstruction in the present invention is improved.
  • the present invention uses a deep convolutional neural network (Decoder) that does not require training to represent the underlying image after Wave-VCC encoding and expansion, CSM, and the background phase of high-frequency changes that cannot be estimated using only the ACS of the low-frequency part, and uses a convolutional neural network to indirectly generate the above three variables.
  • a decoding convolutional neural network that does not require training is used. Compared with traditional supervised neural networks or unsupervised neural networks, there is no need to collect training data, and the update of network parameters is optimized through an optimization algorithm, which is more in line with the characteristics of the difficulty of collecting full sampling data in clinical magnetic resonance imaging.
  • This embodiment also provides a magnetic resonance image reconstruction device, which includes the following components:
  • an information analysis module used for applying a network structure composed of a plurality of neural networks to the image structure information of the depth image, to obtain an underlying image, a background phase, a coil sensitivity of the target object to the magnetic field coil, and a conjugate sensitivity of the coil sensitivity output by the network structure, wherein the depth image is used to characterize the depth information of the target object relative to the magnetic resonance device, and the magnetic field coil is a coil inside the magnetic resonance device;
  • a signal adopting module used for sampling the signal from the target object received by the magnetic resonance device to generate a sampling template signal
  • a phase difference calculation module is used to obtain the magnetic field phase difference formed by each magnetic field according to each magnetic field information applied in the sampling environment;
  • An image reconstruction module is used to reconstruct a magnetic resonance image of the target object based on the underlying image, the coil sensitivity, the conjugate sensitivity, the background phase, the sampling template signal and the magnetic field phase difference.
  • the present invention also provides a terminal device, whose principle block diagram can be shown in Figure 7.
  • the terminal device includes a processor, a memory, a network interface, a display screen, and a temperature sensor connected through a system bus.
  • the processor of the terminal device is used to provide computing and control capabilities.
  • the memory of the terminal device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium.
  • the network interface of the terminal device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a magnetic resonance image reconstruction method is implemented.
  • the display screen of the terminal device can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the terminal device is pre-set inside the terminal device to detect the operating temperature of the internal device.
  • FIG. 7 is only a block diagram of a partial structure related to the solution of the present invention, and does not constitute a limitation on the terminal device to which the solution of the present invention is applied.
  • the specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
  • a terminal device in one embodiment, includes a memory, a processor, and a magnetic resonance image reconstruction program stored in the memory and executable on the processor.
  • the processor executes the magnetic resonance image reconstruction program, the following operation instructions are implemented:
  • the magnetic field phase difference formed by each magnetic field is obtained
  • a magnetic resonance image of the target object is reconstructed according to the underlying image, the coil sensitivity, the conjugate sensitivity, the background phase, the sampling template signal and the magnetic field phase difference.
  • Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Animal Behavior & Ethology (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Radiology & Medical Imaging (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Surgery (AREA)
  • Multimedia (AREA)
  • Heart & Thoracic Surgery (AREA)

Abstract

一种磁共振图像重建方法和图像重建装置。磁共振图像重建方法包括：采用深度神经网络对深度图像的图像结构信息进行分析，得到底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、线圈敏感度的共轭敏感度，然后对磁共振设备接收到的来自目标物体的信号进行采样，生成采样模板信号，之后依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差，最后依据底层图像、线圈敏感度、共轭敏感度、背景相位、采样模板信号以及磁场相位差，重建目标物体的磁共振图像。在计算线圈敏感度的过程中不涉及低频的ACS信号，因此提高了重建图像的质量和速度。

Description

一种磁共振图像重建方法和图像重建装置 技术领域
本发明涉及磁共振成像技术领域，具体是涉及一种磁共振图像重建方法和图像重建装置。
背景技术
磁共振成像(MRI,Magnetic Resonance Imaging)扫描速度慢，过长的扫描时间在引起病患不适的同时，容易在图像中引入运动伪影，从而影响图像质量。磁共振并行成像方法是一类加速MRI扫描速度的方法，如灵敏度编码技术(SENSE,sensitivity encoding)和整体自动校准部分并行采集技术(GRAPPA,generalized autocalibrating partially parallel acquisitions)等。该类方法通过减少采集的数据量，并利用多通道线圈所包含的冗余信息对欠采样数据进行重建，从而到达快速扫描的目的。
磁共振成像包括波浪梯度场编码并行成像技术、Wave编码成像技术、虚共轭线圈(VCC)成像技术。
其中，波浪梯度场编码并行成像技术(Wave encoding)是一种用于加快磁共振扫描速度的并行成像技术，其利用了更高效率的多通道线圈空间编码特性，虚拟共轭线圈技术(Virtual Conjugate Coil,VCC)是和wave编码的具有类似效果的并行成像技术，可以提供更多通道的空间编码先验信息，Wave-VCC的结合能够发挥两者的特性，提供更高倍数的加速技术，但是常规的Wave-VCC重建方法中仅仅利用k-space中间的低频ACS信号对背景相位估计，由于缺乏周围高频的信息，使得估计出的背景相位难以表征高频相位变化的图像。
Wave编码技术是一种用于加快三维磁共振扫描速度的并行成像技术，该技术在MRI信号采集的同时(施加读出梯度场的同时)，利用MRI梯度线圈在选层和相位方向分别施加相位差，利用MRI梯度线圈在相位编码方向施加相位差为
Figure PCTCN2022133016-appb-000001
的正弦梯度场，并采用可控混叠的快速并行成像技术(2D CAIPIRINHA,two-dimension Controlled Aliasing In Parallel Imaging Results In Higher Acceleration)对数据进行欠采，使得欠采样所导致的混叠伪影沿读出、选层和相位方向进行分散，降低各像素点中图像混叠伪影的程度，从而 极大的降低了并行成像重建中的几何因子(g-factor,geometry factor)信噪比丢失，达到高倍加速的目的。
虚共轭线圈(VCC)是另一种改善并行成像中编码矩阵系统条件的技术。其思想是将对象背景和线圈相位合并到重建过程中，通过添加虚拟线圈实现提供额外的编码能力，虚拟线圈是由来自实际物理线圈的共轭对称k空间信号生成的。
现有技术需要采集低频的自动校准信号(auto-calibration signals,ACS)信号，再根据低频的ACS信号去计算高频的线圈敏感度，而在采集低频的ACS信号的过程中，记忆引入运动误差，从而导致计算出的线圈敏感度存在较大误差，进而导致依据线圈敏感度重建的图像质量较差。
综上所述，现有技术计算出的重建图像质量较差。
因此，现有技术还有待改进和提高。
发明内容
为解决上述技术问题，本发明提供了一种磁共振图像重建方法和图像重建装置，解决了现有技术计算出的重建图像质量较差的问题。
为实现上述目的，本发明采用了以下技术方案：
第一方面，本发明提供一种磁共振图像重建方法，其中，包括：
对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈；
对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
在一种实现方式中，所述对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设 备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈，包括：
对所述深度图像的图像结构信息应用具有解码结构的第一深度卷积神经网络，得到所述第一深度卷积神经网络输出的底层图像；
对所述深度图像的图像结构信息应用第二深度卷积神经网络，得到所述第二深度卷积神经网络输出的背景相位；
对所述深度图像的图像结构信息应用第三深度卷积神经网络，得到所述第三深度卷积神经网络输出的线圈敏感度；
对所述深度图像的图像结构信息应用第四深度卷积神经网络，得到所述第四深度卷积神经网络输出的共轭敏感度。
在一种实现方式中，所述对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号，包括：
对所述磁共振设备接收到的来自所述目标物体的信号使用三维MRI序列进行采样，得到选层方向的采样信号和相位方向的采样信号；
依据选层方向的采样信号和相位方向的采样信号，生成采样模板信号。
在一种实现方式中，对所述磁共振设备接收到的来自所述目标物体的信号使用三维MRI序列进行采样的同时，在所述选层方向施加正弦梯度场，在所述相位方向施加截断式正弦梯度场；或者，在所述选层方向施加截断式正弦梯度场，在所述相位方向施加正弦梯度场。
在一种实现方式中，所述截断式正弦梯度场的0阶矩为零。
在一种实现方式中，所述依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差，包括：
依据所述正弦梯度场的磁场相位和所述截断式正弦梯度场的磁场相位，计算所述正弦梯度场与所述截断式正弦梯度场之间的磁场相位差。
在一种实现方式中，所述依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像，包括：
将所述底层图像乘以所述线圈敏感度，得到第一结果；
对所述第一结果应用傅里叶变换，得到第二结果；
将所述第二结果与所述磁场相位差相乘，得到第三结果；
对所述第三结果应用傅里叶逆变换，得到第四结果；
将所述第四结果乘以所述采样模板信号，得到目标信号；
将所述背景相位、所述共轭敏感度、所述底层图像相乘，得到第五结果；
对所述第五结果应用傅里叶变换，得到第六结果；
将所述第六结果乘以磁场相位差，得到第七结果；
对所述第七结果应用傅里叶逆变换，得到第八结果；
将所述第八结果乘以所述采样模板信号，得到所述目标信号的共轭对称信号；
依据所述目标信号和所述共轭对称信号，重建所述目标物体的磁共振图像。
第二方面，本发明实施例还提供一种磁共振图像重建装置，其中，所述装置包括如下组成部分：
信息解析模块，用于对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈；
信号采用模块，用于对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
相位差计算模块，用于依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
图像重建模块，用于依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
第三方面，本发明实施例还提供一种终端设备，其中，所述终端设备包括存储器、处理器及存储在所述存储器中并可在所述处理器上运行的磁共振图像重建程序，所述处理器执行所述磁共振图像重建程序时，实现上述所述的磁共振图像重建方法的步骤。
第四方面，本发明实施例还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有磁共振图像重建程序，所述磁共振图像重建程序被处理器执行时，实现上述所述的磁共振图像重建方法的步骤。
有益效果：本发明重建图像所需的底层图像、背景相位、线圈敏感度、共轭敏感度 全部来源于神经网络，由于本发明在计算线圈敏感度的过程中不涉及低频的ACS信号，因此提高了本发明重建图像的质量。另外，由于本发明的底层图像也是通过神经网络计算得到，从而提高了本发明重建图像的速度。
附图说明
图1为本发明的整体流程图；
图2为本发明实施例中的深度卷积神经网络结构图；
图3为本发明实施例中的数据欠采样策略示意图；
图4为本发明实施例中的截断式正弦梯度场示意图；
图5为本发明实施例中的正弦梯度场示意图；
图6为本发明实施例中的截断式梯度场用于bSSFP序列示意图；
图7为本发明实施例提供的终端设备的内部结构原理框图。
具体实施方式
以下结合实施例和说明书附图，对本发明中的技术方案进行清楚、完整地描述。基于本发明中的实施例，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，都属于本发明保护的范围。
经研究发现，现有技术需要采集低频的ACS信号，再根据低频的ACS信号去计算高频的线圈敏感度，而在采集低频的ACS信号的过程中，记忆引入运动误差，从而导致计算出的线圈敏感度存在较大误差，进而导致依据线圈敏感度重建的图像质量较差。
为解决上述技术问题，本发明提供了一种磁共振图像重建方法和图像重建装置，解决了现有技术计算出的重建图像质量较差的问题。具体实施时，首先对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、线圈敏感度的共轭敏感度；对磁共振设备接收到的来自目标物体的信号进行采样，生成采样模板信号；之后依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；最后依据底层图像、线圈敏感度、共轭敏感度、背景相位、采样模板信号以及磁场相位差，重建目标物体的磁共振图像。 本发明提高了重建图像的质量。
举例说明，需要用磁共振成像技术采集患者的病变处图像，磁共振设备固定在一个位置处，通过磁共振设备采集患者病变处的深度图像(用于表征病变处各点与磁共振设备之间的距离)。将该深度图像的图像信息分别输入到四个神经网络中，得到底层图像、背景相位、线圈敏感度(病变处对磁共振设备内部线圈的敏感度，即线圈变化会导致采集到病变处的图像信息发生多大的变化)、共轭敏感度。另外，在对信号(该信号为磁共振设备向病变处发送原始信号之后，病变处对该原始信号作用之后形成的信号)进行采样以得到采样模板信号的同时给信号所在的环境施加各种磁场。各种磁场之间会产生磁场相位差。最后结合底层图像、线圈敏感度、共轭敏感度、背景相位、采样模板信号以及磁场相位差，就可以重建出病变处(目标物体)的磁共振图像。
示例性方法
本实施例的磁共振图像重建方法可应用于终端设备中，所述终端设备可为具有图像采集功能的终端产品，比如磁共振设备等。在本实施例中，如图1中所示，所述磁共振图像重建方法具体包括如下步骤：
S100，对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈。
本实施例中，图像结构信息被利用，这种图像结构信息在MRI图像中一般表示为图像本身的稀疏性质，而这种性质可以用优化后的网络结构去表征。
本实施例包括四个如图2所示的神经网络，图2中从上到下依次为第一深度卷积神经网络CNN k、第二深度卷积神经网络
Figure PCTCN2022133016-appb-000002
第三深度卷积神经网络CNN C、第四深度卷积神经网络CNN C*。本实施例中的这四个深度卷积神经网络结构类似，不同的是中间通道数不同，在本实施例中使用Adam优化器去迭代优化网络的参数，当网络的参数被优化好后，随机输入一个小的噪声就可以生成完整的图像。
优化参数依据如下的公式：
Figure PCTCN2022133016-appb-000003
络的输出，
Figure PCTCN2022133016-appb-000004
为深度卷积神经网络的损失函数，
Figure PCTCN2022133016-appb-000005
分别为上述四个深度卷积神经网络的参数，通过上述公式计算损失函数的二范数取最小值时所对应的网络参数的值，以完成对网络参数的优化。
优化网络之后，将深度图像的图像结构信息输入到解码结构的第一深度卷积神经网络CNN k，第一深度卷积神经网络CNN k输出的底层图像m。将深度图像的图像结构信息输入到第二深度卷积神经网络
Figure PCTCN2022133016-appb-000006
第二深度卷积神经网络
Figure PCTCN2022133016-appb-000007
输出的背景相位
Figure PCTCN2022133016-appb-000008
将深度图像的图像结构信息输入到第三深度卷积神经网络CNN C，第三深度卷积神经网络CNN C输出的线圈敏感度CSM。将深度图像的图像结构信息输入到第四深度卷积神经网络CNN C*，第四深度卷积神经网络CNN C*输出的共轭敏感度CSM*。其中，CNN k的输入为
Figure PCTCN2022133016-appb-000009
CNN c、
Figure PCTCN2022133016-appb-000010
和CNN φ的输入为
Figure PCTCN2022133016-appb-000011
本实施例中，没有直接使用仅仅包含低频相位信息的ACS信号去估计背景相位，而是根据图像的退化过程以及最后采集的信号去表征图像以及其相位信息，因此能更加准确的生成图像以及包含的背景相位信息。
S200，对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号。
比如磁共振设备向人体病变处(目标物体)发送原始信号，人体病变处与原始信号相互作用，使得原始信号变成一种新的信号，对该新的信号进行采样，生成如图3所示的采样模板信号。
步骤S200在采样之前需要施加正弦梯度场和截断式正弦梯度场。
在一个实施例中，在进行信号采样之前，利用MRI梯度场线圈在选层方向施加正弦梯度场，同时利用MRI梯度场线圈在相位方向施加如图4所示的截断式正弦梯度场。或者，利用MRI梯度场线圈在选层方向施加如图4所示的截断式正弦梯度场，同时利用MRI梯度场线圈在相位方向施加如图5所示的正弦梯度场。且截断式正弦梯度场的0阶矩为零。
该实施例中，截断式正弦梯度场的0阶矩(在相位方向和选层方向)为零再结合截断式正弦梯度场在相位方向采用如图4所示的方向，不仅能够有效分散由欠采样所导致的混叠伪影，降低g-factor引起的信噪比丢失以实现高倍加速，同时避免了由0阶矩不为零的梯度场所导致的成像伪影。
在一个实施例中，正弦梯度场与截断式正弦梯度场的磁场相位差Psf为
Figure PCTCN2022133016-appb-000012
在一个实施例中，正弦梯度场的表达式如下：
Figure PCTCN2022133016-appb-000013
Figure PCTCN2022133016-appb-000014
其中，t为时间，且t＝0时间点已在图5中标注；
Figure PCTCN2022133016-appb-000015
和
Figure PCTCN2022133016-appb-000016
分别为wave-CAIPI技术在相位和选层方向所施加的梯度场；A为正弦梯度场幅值；D C为一个正弦梯度场周期的持续时间(如图5中所标注)；D R为读出梯度场的平台持续时间(如图5中所标注)。
施加上述正弦梯度场和截断式正弦梯度场之后，步骤S200开始采样信号，步骤S200包括如下的步骤S201和S202：
S201，对所述磁共振设备接收到的来自所述目标物体的信号使用三维MRI序列进行采样，得到选层方向的采样信号和相位方向的采样信号。
本实施例中，三维MRI序列包括GRE序列、SE序列、bSSFP序列等。
S202，依据选层方向的采样信号和相位方向的采样信号，生成采样模板信号。
本实施例使用三维MRI序列对信号进行欠采样，以达到加速扫描的目的。将截断式梯度场应用于bSSFP序列当中，并使用2D CAIPIRINHA技术的信号采集策略减少数据采集量。通过截断式梯度场和2D CAIPIRINHA采样策略的结合，能够将欠采样所导致的混叠同时分散到读出、相位和选层方向，更有效的利用了FOV中的背景区域，增大了不同像素点间的灵敏度差异，从而更进一步降低g-factor信噪比丢失。
截断式梯度场采用如图6所示的方式应用在bSSFP序列中，同时2D CAIPIRINHA数据采集策略如图3所示。图3中同时垂直于相位方向和选层方向为读出方向，虚线交点为全采样所需采集的读出线，本发明所采用的欠采样策略所需采集的读出线由实心原点表示。图3所示为3×3倍欠采样(相位方向3倍欠采样，选层方向3倍欠采样)，总加速倍数为9，所需采集时间为重复时间(TR,repetition time)×相位编码线数(Np)×选层编码线数(Ns)/9。
S300，依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差。
本实施例中，各个磁场信息为步骤S200中的正弦梯度场的磁场相位和截断式正弦梯 度场的磁场相位，在二维情况中，点扩散函数Psf Y中任意点(k x,y)的点扩散函数值为Psf Y(k x,y)＝waveP y(k x,y)/P y(k x,y)，如果在另外一个频率编码方向，用z方向表示，则上式就会变成Psf z(k x,y)＝waveP z(k x,z)/P z(k x,z)，在三维情况中，两个相位编码方向均添加梯度正弦梯度磁场和截断式正弦梯度磁场，则三维点扩散函数为Psf yz(k x,y,z)＝Psf z(k x,z)·Psf y(k x,y)，即三维点扩散函数Psf yz中任意点(k x,y,z)的值为Psf yz(k x,y,z)。
S400，依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
步骤S400包括如下的步骤S401至S4011：
S401，将所述底层图像乘以所述线圈敏感度，得到第一结果。
S402，对所述第一结果应用傅里叶变换，得到第二结果。
S403，将所述第二结果与所述磁场相位差Psf相乘，得到第三结果。
S404，对所述第三结果应用傅里叶逆变换，得到第四结果。
S405，将所述第四结果乘以所述采样模板信号，得到目标信号b：
Figure PCTCN2022133016-appb-000017
Figure PCTCN2022133016-appb-000018
对应第一深度卷积神经网络输出的底层图像m，
Figure PCTCN2022133016-appb-000019
对应第三深度卷积神经网络CNN C输出的线圈敏感度CSM，
Figure PCTCN2022133016-appb-000020
对应第一结果，
Figure PCTCN2022133016-appb-000021
为对第一结果应用傅里叶变换所得的第二结果，
Figure PCTCN2022133016-appb-000022
为第三结果，
Figure PCTCN2022133016-appb-000023
对第三结果应用傅里叶逆变换，M为图3中的采样模板信号。
S406，将所述背景相位、所述共轭敏感度、所述底层图像相乘，得到第五结果；
S407，对所述第五结果应用傅里叶变换，得到第六结果；
S408，将所述第六结果乘以磁场相位差Psf，得到第七结果；
S409，对所述第七结果应用傅里叶逆变换，得到第八结果；
S4010，将所述第八结果乘以所述采样模板信号，得到所述目标信号的共轭对称信号b H：
Figure PCTCN2022133016-appb-000024
Figure PCTCN2022133016-appb-000025
对应第一深度卷积神经网络输出的底层图像m，
Figure PCTCN2022133016-appb-000026
对应第四深度卷积神经网络输出的共轭敏感度CSM*，
Figure PCTCN2022133016-appb-000027
对应第二深度卷积神经网络输出的背景相位，
Figure PCTCN2022133016-appb-000028
为第五结果，
Figure PCTCN2022133016-appb-000029
为第六结果，
Figure PCTCN2022133016-appb-000030
为第七结果,
Figure PCTCN2022133016-appb-000031
为傅里叶逆变换。
在一个实施例中，用b i替代b，用
Figure PCTCN2022133016-appb-000032
替代b H：
b i＝MF y,zPsf(k x,y,z)F xC im
Figure PCTCN2022133016-appb-000033
b i和
Figure PCTCN2022133016-appb-000034
是wave编码的前向模型，其中M为CAIPI采样模板，如图3所示，在实际的磁共振成像过程中，
Figure PCTCN2022133016-appb-000035
为背景相位，D i为接收线圈固有的线圈敏感度，在一般的模型中并没有单独考虑背景相位，而是将其包含在C i中进行后续重建工作，通过VCC进行扩展后的数据为，
Figure PCTCN2022133016-appb-000036
Figure PCTCN2022133016-appb-000037
为
Figure PCTCN2022133016-appb-000038
的共轭对称，Psf *为Psf的共轭对称，接收的信号
Figure PCTCN2022133016-appb-000039
为原始信号b i的共轭对称，通过这样虚拟共轭对称后，相当于提供了另外一组的相位信息，也就是说对wave编码框架中提供了一组额外的接收线圈的额外编码先验信息，进一步提供加速倍数。
S4011，依据所述目标信号b和所述共轭对称信号b H，重建所述目标物体的磁共振图像。
根据目标信号b和共轭对称信号b H可以重建出目标物体的真实磁共振图像，通过b和b H得到磁共振图像为现有技术。
综上，本发明重建图像所需的底层图像、背景相位、线圈敏感度、共轭敏感度全部来源于神经网络，由于本发明在计算线圈敏感度的过程中不涉及低频的ACS信号，因此提高了本发明重建图像的质量。另外，由于本发明的底层图像也是通过神经网络计算得到，从而提高了本发明重建图像的速度。
另外，本发明使用无需训练的深度卷积神经网络(Decoder)去表示经过Wave-VCC编码和扩展后的底层图像、CSM以及无法仅使用低频部分的ACS估计的高频变化的背景相位，使用卷积神经网络先去间接生成上述三个变量，在本发明中，使用无需训练的解码卷积神经网路，相对于传统的无论是监督神经网络还是无监督神经网络，不需要收集训练数据，网络参数的更新是通过优化算法进行优化的，这更加符合临床磁共振成像中全采样数据的难以收集的特点。
示例性装置
本实施例还提供一种磁共振图像重建装置，所述装置包括如下组成部分：
信息解析模块，用于对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈；
信号采用模块，用于对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
相位差计算模块，用于依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
图像重建模块，用于依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
基于上述实施例，本发明还提供了一种终端设备，其原理框图可以如图7所示。该终端设备包括通过系统总线连接的处理器、存储器、网络接口、显示屏、温度传感器。其中，该终端设备的处理器用于提供计算和控制能力。该终端设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该终端设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种磁共振图像重建方法。该终端设备的显示屏可以是液晶显示屏或者电子墨水显示屏，该终端设备的温度传感器是预先在终端设备内部设置，用于检测内部设备的运行温度。
本领域技术人员可以理解，图7中示出的原理框图，仅仅是与本发明方案相关的部分结构的框图，并不构成对本发明方案所应用于其上的终端设备的限定，具体的终端设备以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，提供了一种终端设备，终端设备包括存储器、处理器及存储在存储器中并可在处理器上运行的磁共振图像重建程序，处理器执行磁共振图像重建程序时，实现如下操作指令：
对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场 线圈为所述磁共振设备内部的线圈；
对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是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非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本发明所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
最后应说明的是：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. 一种磁共振图像重建方法，其特征在于，包括：
    对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈；
    对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
    依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
    依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
  2. 如权利要求1所述的磁共振图像重建方法，其特征在于，所述对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈，包括：
    对所述深度图像的图像结构信息应用具有解码结构的第一深度卷积神经网络，得到所述第一深度卷积神经网络输出的底层图像；
    对所述深度图像的图像结构信息应用第二深度卷积神经网络，得到所述第二深度卷积神经网络输出的背景相位；
    对所述深度图像的图像结构信息应用第三深度卷积神经网络，得到所述第三深度卷积神经网络输出的线圈敏感度；
    对所述深度图像的图像结构信息应用第四深度卷积神经网络，得到所述第四深度卷积神经网络输出的共轭敏感度。
  3. 如权利要求1所述的磁共振图像重建方法，其特征在于，所述对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号，包括：
    对所述磁共振设备接收到的来自所述目标物体的信号使用三维MRI序列进行采样，得到选层方向的采样信号和相位方向的采样信号；
    依据选层方向的采样信号和相位方向的采样信号，生成采样模板信号。
  4. 如权利要求3所述的磁共振图像重建方法，其特征在于，对所述磁共振设备接收 到的来自所述目标物体的信号使用三维MRI序列进行采样的同时，在所述选层方向施加正弦梯度场，在所述相位方向施加截断式正弦梯度场；或者，在所述选层方向施加截断式正弦梯度场，在所述相位方向施加正弦梯度场。
  5. 如权利要求4所述的磁共振图像重建方法，其特征在于，所述截断式正弦梯度场的0阶矩为零。
  6. 如权利要求4所述的磁共振图像重建方法，其特征在于，所述依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差，包括：
    依据所述正弦梯度场的磁场相位和所述截断式正弦梯度场的磁场相位，计算所述正弦梯度场与所述截断式正弦梯度场之间的磁场相位差。
  7. 如权利要求2所述的磁共振图像重建方法，其特征在于，所述依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像，包括：
    将所述底层图像乘以所述线圈敏感度，得到第一结果；
    对所述第一结果应用傅里叶变换，得到第二结果；
    将所述第二结果与所述磁场相位差相乘，得到第三结果；
    对所述第三结果应用傅里叶逆变换，得到第四结果；
    将所述第四结果乘以所述采样模板信号，得到目标信号；
    将所述背景相位、所述共轭敏感度、所述底层图像相乘，得到第五结果；
    对所述第五结果应用傅里叶变换，得到第六结果；
    将所述第六结果乘以磁场相位差，得到第七结果；
    对所述第七结果应用傅里叶逆变换，得到第八结果；
    将所述第八结果乘以所述采样模板信号，得到所述目标信号的共轭对称信号；
    依据所述目标信号和所述共轭对称信号，重建所述目标物体的磁共振图像。
  8. 一种磁共振图像重建装置，其特征在于，所述装置包括如下组成部分：
    信息解析模块，用于对深度图像的图像结构信息应用由若干个神经网络组成的网络结构，得到所述网络结构输出的底层图像、背景相位、目标物体对磁场线圈的线圈敏感度、所述线圈敏感度的共轭敏感度，所述深度图像用于表征目标物体相对磁共振设备的深度信息，所述磁场线圈为所述磁共振设备内部的线圈；
    信号采用模块，用于对所述磁共振设备接收到的来自所述目标物体的信号进行采样，生成采样模板信号；
    相位差计算模块，用于依据采样环境中施加的各个磁场信息，得到各个磁场所构成的磁场相位差；
    图像重建模块，用于依据所述底层图像、所述线圈敏感度、所述共轭敏感度、所述背景相位、所述采样模板信号以及所述磁场相位差，重建所述目标物体的磁共振图像。
  9. 一种终端设备，其特征在于，所述终端设备包括存储器、处理器及存储在所述存储器中并可在所述处理器上运行的磁共振图像重建程序，所述处理器执行所述磁共振图像重建程序时，实现如权利要求1-7任一项所述的磁共振图像重建方法的步骤。
  10. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质上存储有磁共振图像重建程序，所述磁共振图像重建程序被处理器执行时，实现如权利要求1-7任一项所述的磁共振图像重建方法的步骤。
PCT/CN2022/133016 2022-11-16 2022-11-18 一种磁共振图像重建方法和图像重建装置 WO2024103414A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211435020.2A CN115690253A (zh) 2022-11-16 2022-11-16 一种磁共振图像重建方法和图像重建装置
CN202211435020.2 2022-11-16

Publications (1)

Publication Number Publication Date
WO2024103414A1 true WO2024103414A1 (zh) 2024-05-23

Family

ID=85054848

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/133016 WO2024103414A1 (zh) 2022-11-16 2022-11-18 一种磁共振图像重建方法和图像重建装置

Country Status (2)

Country Link
CN (1) CN115690253A (zh)
WO (1) WO2024103414A1 (zh)

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2581756A1 (en) * 2011-10-11 2013-04-17 Koninklijke Philips Electronics N.V. MR imaging using parallel signal acquisition
CN108335339A (zh) * 2018-04-08 2018-07-27 朱高杰 一种基于深度学习和凸集投影的磁共振重建方法
US20190204401A1 (en) * 2017-12-29 2019-07-04 Siemens Healthcare Gmbh Method and Deep Quantitative Susceptibility Mapping (QSM)
CN111513716A (zh) * 2019-02-05 2020-08-11 通用电气精准医疗有限责任公司 使用扩展灵敏度模型和深度神经网络进行磁共振图像重建的方法和系统
CN112014782A (zh) * 2020-08-06 2020-12-01 深圳先进技术研究院 磁共振成像方法、装置及计算机存储介质
CN113298901A (zh) * 2021-05-13 2021-08-24 中国科学院深圳先进技术研究院 卷褶视野磁共振图像的重建方法、计算机设备及存储介质
CN113313778A (zh) * 2021-05-13 2021-08-27 中国科学院深圳先进技术研究院 磁共振图像的重建方法、计算机设备及存储介质
CN114565690A (zh) * 2021-12-15 2022-05-31 中国科学院深圳先进技术研究院 基于波浪梯度编码场和深度学习模型的磁共振成像方法
CN114820849A (zh) * 2022-04-19 2022-07-29 浙江大学 基于深度学习的磁共振cest图像重建方法、装置及设备
CN115135237A (zh) * 2020-02-13 2022-09-30 氧气医药株式会社 磁共振影像处理装置及其方法

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2581756A1 (en) * 2011-10-11 2013-04-17 Koninklijke Philips Electronics N.V. MR imaging using parallel signal acquisition
US20190204401A1 (en) * 2017-12-29 2019-07-04 Siemens Healthcare Gmbh Method and Deep Quantitative Susceptibility Mapping (QSM)
CN108335339A (zh) * 2018-04-08 2018-07-27 朱高杰 一种基于深度学习和凸集投影的磁共振重建方法
CN111513716A (zh) * 2019-02-05 2020-08-11 通用电气精准医疗有限责任公司 使用扩展灵敏度模型和深度神经网络进行磁共振图像重建的方法和系统
CN115135237A (zh) * 2020-02-13 2022-09-30 氧气医药株式会社 磁共振影像处理装置及其方法
CN112014782A (zh) * 2020-08-06 2020-12-01 深圳先进技术研究院 磁共振成像方法、装置及计算机存储介质
CN113298901A (zh) * 2021-05-13 2021-08-24 中国科学院深圳先进技术研究院 卷褶视野磁共振图像的重建方法、计算机设备及存储介质
CN113313778A (zh) * 2021-05-13 2021-08-27 中国科学院深圳先进技术研究院 磁共振图像的重建方法、计算机设备及存储介质
CN114565690A (zh) * 2021-12-15 2022-05-31 中国科学院深圳先进技术研究院 基于波浪梯度编码场和深度学习模型的磁共振成像方法
CN114820849A (zh) * 2022-04-19 2022-07-29 浙江大学 基于深度学习的磁共振cest图像重建方法、装置及设备

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XIAOFANG LIU, CHEN YONGLI, YE XIUZI: "The Encoding and Reconstruction of Parallel MRI Chinese Full Text Journal of Biomedical Engineering", JOURNAL OF BIOMEDICAL ENGINEERING, vol. 27, no. 2, 25 April 2010 (2010-04-25), pages 471 - 475, XP093169994 *
程慧涛 等 (CHENG, HUITAO ET AL.): "基于深度递归级联卷积神经网络的并行磁共振成像方法 (Deep Recursive Cascaded Convolutional Network for Parallel MRI)", 波谱学杂志 CHINESE JOURNAL OF MAGNETIC RESONANCE), vol. 36, no. 4, 15 December 2019 (2019-12-15), XP093068441, DOI: 10.11938/cjmr20192721 *

Also Published As

Publication number Publication date
CN115690253A (zh) 2023-02-03

Similar Documents

Publication Publication Date Title
Chung et al. Score-based diffusion models for accelerated MRI
US7902825B2 (en) Motion corrected tensor magnetic resonance imaging
Hammernik et al. Physics-Driven Deep Learning for Computational Magnetic Resonance Imaging: Combining physics and machine learning for improved medical imaging
US20070253599A1 (en) Motion Estimation Using Hidden Markov Model Processing in MRI and Other Applications
WO2020114329A1 (zh) 磁共振快速参数成像方法及装置
US11131738B2 (en) Multiband, multishot magnetic resonance elastography
CN112037298A (zh) 图像重建方法、装置、计算机设备和存储介质
Schaetz et al. Accelerated computing in magnetic resonance imaging: real‐time imaging using nonlinear inverse reconstruction
WO2022236851A1 (zh) 卷褶视野磁共振图像的重建方法、计算机设备及存储介质
KR102584166B1 (ko) 리스케일링과 인공신경망을 적용한 자기 공명 영상 처리 장치 및 그 방법
WO2022236823A1 (zh) 卷褶视野磁共振图像的重建方法、计算机设备及存储介质
CN114255291A (zh) 用于磁共振参数定量成像的重建方法、系统
CN111681297B (zh) 图像重建方法、计算机设备和存储介质
WO2024021796A1 (zh) 一种图像处理方法、装置、电子设备、存储介质及程序产品
Sun et al. DIR3D: Cascaded Dual-Domain Inter-Scale Mutual Reinforcement 3D Network for highly accelerated 3D MR image reconstruction
WO2024103414A1 (zh) 一种磁共振图像重建方法和图像重建装置
CN113866694B (zh) 一种快速三维磁共振t1定量成像方法、系统及介质
Chang et al. Parallel MRI reconstruction using broad learning system
WO2021129235A1 (zh) 三维磁共振快速参数成像方法和装置
CN115236576A (zh) 快速多对比度磁共振扩散成像和重建方法及系统
CN118215851A (zh) 用于加速非笛卡尔磁共振成像重建的双域自监督学习
Wu et al. Deriving new soft tissue contrasts from conventional MR images using deep learning
Dang et al. Joint MR sequence optimization beats pure neural network approaches for spin-echo MRI super-resolution
WO2023108484A1 (zh) 基于波浪梯度编码场和深度学习模型的磁共振成像方法
US20230288515A1 (en) Augmenting Diffusion-Weighted Magnetic Resonance Imaging Data

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: 22965593

Country of ref document: EP

Kind code of ref document: A1