Dynamic magnetic resonance delay enhancement imaging method

Abstract

The invention discloses a dynamic magnetic resonance delay enhancement imaging method. The method comprises the following steps: for a target part, acquiring k-space data by using a phase-sensitive inversion recovery sequence and rearranging the k-space data to obtain rearranged data; and inputting the rearranged data into a trained image reconstruction model to obtain a reconstructed image, the image reconstruction model being constructed based on a deep balance model and being used for learning a mapping relationship between the rearranged undersampled data and the reconstructed image. According to the method, segmented k-space data is divided into multiple times of high-acceleration acquisition, artifacts caused by heart movement are remarkably reduced, and a high-quality PSIR image is reconstructed from undersampled data by combining a deep expansion network with a deep balance model.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic resonance imaging, and more specifically, to a dynamic magnetic resonance delayed enhancement imaging method. Background Art

[0002] Phase-sensitive inversion recovery (PSIR) technology effectively enhances the contrast between normal myocardium and scarred myocardium by restoring the true magnetization polarity in magnetic resonance delayed gadolinium enhancement (LGE) imaging, and is an important clinical tool for evaluating myocardial viability. However, this technology requires the simultaneous acquisition of inversion recovery (IR) images and proton density reference images, which doubles the scanning time compared to conventional IR sequences. During long-term scanning, uncontrollable physiological movements (such as cardiac movement) can easily introduce artifacts and image blur, thereby reducing image quality.

[0003] Dynamic late gadolinium enhancement imaging is a key medical imaging technique that is widely used in the diagnosis of cardiovascular lesions. However, dynamic MRI with higher temporal and spatial resolution often requires longer scanning time, which limits its application in time-sensitive clinical scenarios. In order to overcome the contradiction between scanning time and image quality, researchers have proposed a variety of reconstruction methods. For example, because dynamic MRI data has low-rank and sparsity characteristics, these characteristics can be used for data decomposition, thereby achieving efficient image reconstruction under undersampling conditions. Therefore, methods based on low-rank decomposition and sparse modeling have been proposed. With the development of deep learning technology, the application of end-to-end neural networks in MRI image reconstruction has achieved remarkable results. However, end-to-end deep learning methods directly map undersampled data to reconstructed images, lack physical constraints, and are prone to artifacts in highly undersampled and complex scenarios. In recent years, methods based on data-driven physical models combined with deep learning have achieved better reconstruction results at higher acceleration times. These methods are usually combined with physical models and improve reconstruction quality by optimizing regularization terms, but further improvements are still needed to improve image reconstruction quality in scenarios with high temporal resolution and significant cardiac motion artifacts.

[0004] In summary, the existing delayed enhancement magnetic resonance imaging reconstruction method has the following defects:

[0005] 1) Methods based on low-rank decomposition and sparse modeling rely on the assumption of low-rank and sparsity of data, which makes it difficult to ensure the quality of reconstruction in complex scenarios. They also have high computational overhead and cannot meet the needs of clinical efficient real-time reconstruction.

[0006] 2) Although the end-to-end deep learning method accelerates the reconstruction process, it is easy to produce artifacts under high acceleration or complex scenes due to the lack of physical model constraints, resulting in unstable reconstruction quality;

[0007] 3) Data-driven physical models combined with deep learning methods have improved the reconstruction quality to a certain extent, but there is still room for further improvement in scenarios with high temporal resolution and significant cardiac motion artifacts. Summary of the invention

[0008] The purpose of the present invention is to overcome the defects of the prior art and provide a dynamic magnetic resonance delayed enhancement imaging method. The method comprises the following steps:

[0009] For the target part, the k-space data is collected and rearranged using a phase-sensitive inversion recovery sequence to obtain rearranged data;

[0010] The rearranged data is input into a trained image reconstruction model to obtain a reconstructed image, wherein the image reconstruction model is constructed based on a deep balanced model and is used to learn a mapping relationship between the rearranged undersampled data and the reconstructed image.

[0011] Compared with the prior art, the advantages of the present invention are that the proposed dynamic delayed enhancement imaging method is a dynamic PSIR LGE imaging method (or Dyn-PSIR), which significantly reduces the artifacts caused by cardiac motion by dividing the segmented k-space data into multiple high-acceleration acquisitions. In addition, a deep unfolding network combined with a deep equilibrium model (DEQ) is used to reconstruct high-quality PSIR images from undersampled data.

[0012] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0014] Figure 1 is a flow chart of a dynamic magnetic resonance delayed enhancement imaging method according to an embodiment of the present invention;

[0015] Figure 2 is a schematic diagram of a k-space data rearrangement strategy according to an embodiment of the present invention;

[0016] Figure 3 is a schematic diagram of the overall architecture of a dynamic magnetic resonance delayed enhancement imaging method according to an embodiment of the present invention;

[0017] Figure 4is a schematic diagram comparing the effects of a reconstructed image according to an embodiment of the present invention and the prior art;

[0018] Figure 5 It is a schematic diagram comparing the effects of a reconstructed image according to another embodiment of the present invention and the prior art. DETAILED DESCRIPTION

[0019] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0020] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0021] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0022] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0023] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0024] The present invention proposes an efficient dynamic delayed enhancement magnetic resonance imaging method, which segments the undersampled k-space data into multiple high undersampled k-space data to reduce cardiac motion artifacts, and combines deep learning and a self-supervised reconstruction network guided by a physical model to learn the mapping relationship between undersampled data and high-quality images. In addition, an optimization architecture based on a deep balance model is adopted to balance the constraints of data consistency and image priors during iteration, thereby effectively capturing the dynamic characteristics of the image. Finally, accurate restoration of dynamic PSIR images is achieved through a multi-stage training and reconstruction strategy.

[0025] Specifically, see Figure 1 As shown, the provided dynamic magnetic resonance delayed enhancement imaging method comprises the following steps:

[0026] Step S110 , rearrange the k-space data acquired by using the phase-sensitive inversion recovery sequence.

[0027] In clinical practice, PSIR (Phase Sensitive Inversion Recovery) sequences are often used for data acquisition, where PSIR k-space data is generated by combining segmented k-space lines into a complete data set. In one embodiment, these data are rearranged into multiple time frames, such as Figure 2 As shown. Traditional PSIR reconstruction merges segmented k-space data into a complete data set, while in the embodiment of the present invention, these data are rearranged into multiple moments to achieve high acceleration multiple undersampling. This rearrangement strategy enables each moment to achieve high undersampling and has sampling characteristics that are resistant to motion artifacts.

[0028] Step S120: constructing an image reconstruction model based on the deep balance model.

[0029] Let x and y represent the PSIR image to be reconstructed and the undersampled k-space data respectively. The forward reconstruction model is:

[0030]

[0031] Where A represents the encoding operator, λ is the regularization coefficient, represents a regularization term, which is used to improve the reconstruction performance.

[0032] The optimization problem in formula (1) can be solved iteratively by the proximal gradient descent method. The iterative formula is expressed as:

[0033]

[0034] Among them, h is an auxiliary variable, k represents the number of iterations, η is the step size, represents the proximal operator, which is defined as:

[0035]

[0036] Among them, z represents the result of the previous iteration, and w represents the result after the next iteration.

[0037] The iterative algorithm can be expanded into a neural network structure, using a deep neural network to learn the proximal operator:

[0038]

[0039] Where Γ is the learned operator. In order to improve the expressiveness and convergence of the network, the deep equilibrium model (DEQ) is optimized by using fixed point equations, as follows:

[0040] x k+1 =f θ (x k ; y) (5)

[0041] Among them, fθ is a learning operator with parameters θ.

[0042] Step S130, using the set loss function to train the image reconstruction model in a self-supervised manner.

[0043] Figure 3 It is the overall architecture of the self-supervised reconstruction network model, including a 3D convolution-based balancing module (or 3D module for short), and a subsequent 2D convolution module (or 2D module for short). The 3D convolution-based balancing module is used to perform fixed-point iterations. The 2D convolution module reduces the impact of image changes in each time frame on the 3D convolution, that is, reduces the impact of image changes between shots on the 3D convolution.

[0044] In one embodiment, the loss function for training the self-supervised reconstruction network is defined as follows:

[0045]

[0046] in, represents the loss, x label is the label image from the fully sampled data, x 2D and x 3D They are the outputs of the 2D module and the 3D module respectively. is a feature extractor pre-trained by the VGG network. The final PSIR reconstruction result can be calculated using the following formula:

[0047]

[0048] Among them, x r and x m denote the PSIR reference image and the amplitude image respectively, where ψ(·) is a phase function, ψ(x r ) represents the phase information calculated based on the PSIR reference image, x real is the ground-truth image of PSIR.

[0049] Step S140: In practical applications, for the spatial data collected by using the phase-sensitive inversion recovery sequence, a reconstructed image is obtained by using a trained image reconstruction model.

[0050] After the image reconstruction model is trained, the optimized parameters, such as weights and biases, can be obtained and then used for actual image reconstruction. For example, the model application process is: using the k-space data collected by the phase-sensitive inversion recovery sequence; rearranging the k-space data and inputting it into the trained image reconstruction model to obtain the reconstructed PSIR image.

[0051] In order to further verify the effect of the present invention, experiments were conducted and compared with the prior art. The prior art is marked as:

[0052] GRAPPA (Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn. Reson. Med. 2002; 47(6): 1202-1210.);

[0053] MoDL (HK Aggarwal, MP Mani, and M. Jacob, "MoDL: Model-based deep learning architecture for inverse problems," IEEE Transactions on Medical Imaging, vol. 38, no. 2, pp. 394–405, 2018.);

[0054] L+S (R. Otazo, E. Candes, and DKSodickson, "Low-rank plus sparse matrixdecomposition for accelerated dynamic MRI with separation of background and dynamic components," Magnetic Resonance in Medicine, vol. 73, no. 3, pp. 1125–1136, 2015.).

[0055] Two PSIR LGE datasets were acquired from patients with myocardial infarction using 3T and 5T scanners (uMR 890 and uMR Jupiter, United Imaging Healthcare, Shanghai, China) with parallel acceleration factors of R = 2 and R = 3, respectively. Each dataset was divided into multiple time frames and the Dyn-PSIR method was used to reconstruct each moment, achieving up to 12.5 times speedup. GRAPPA data at R = 2 and R = 3 were used as reference images. Since fully sampled PSIR LGE data are rarely available in clinical settings, the network was trained on a fully sampled pelvic image dataset using undersampled data with acceleration factors R ranging from 6 to 12.

[0056] Figure 4The following is a schematic diagram of the 5T PSIR image reconstructed using the Dyn-PSIR, MoDL and L+S methods at an acceleration factor of R = 10.5. Figure 4 (a) is the amplitude image, Figure 4 (b) is the true value image, corresponding to the first and second frames respectively. Figure 4 It can be seen that aliasing artifacts can be observed in the reconstructed image of the MoDL method, while the image generated by the present invention (Dyn-PSIR) has high texture fidelity, and the image quality is comparable to the reference image and better than the L+S method.

[0057] Figure 5 The 3T PSIR image reconstructed using the Dyn-PSIR, MoDL and L+S methods at an acceleration factor of R = 12.5 is shown in Figure 1. Figure 5 (a) Corresponding amplitude image, Figure 5 (b) The corresponding true value image, corresponding to the 1st and 3rd moments respectively. Figure 5 It can be seen that in the reconstructed images of the L+S and MoDL methods, aliasing artifacts can be observed, as shown by the arrows. The experimental results show that the images generated by the present invention still show image quality comparable to that of the reference images.

[0058] In addition, different methods were quantitatively compared on 5T and 3T PSIR LGE images, covering different tangents. Table 1 corresponds to 5T images, and Table 2 corresponds to 3T images. 4CH represents four-chamber heart view, 3CH represents three-chamber heart view, 2CH represents two-chamber heart view, and SAX represents short-axis view. The best results are shown in bold. The experimental comparison results show that the images reconstructed by the present invention achieved the highest values ​​in peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), and achieved the lowest reconstruction error.

[0059] Table 1: Comparison results on 5T image 5TPSIR magnitude image

[0060]

[0061] 5T PSIRrealimage

[0062]

[0063] Table 2: Comparison results on 3T images 3TPSIRmagnitudeimage

[0064]

[0065] 3TPSIRreal image

[0066]

[0067] In summary, the present invention proposes a dynamic PSIR delayed gadolinium enhancement (LGE) imaging method (Dyn-PSIR), which realizes the reconstruction of high-quality PSIR images under highly undersampled data by dividing the segmented k-space data into multiple highly undersampled k-space data and combining the deep unfolding network with the deep equilibrium model (DEQ). It has been verified that the present invention can effectively reduce motion artifacts, effectively balance image quality and acceleration rate, and achieve an acceleration capability of up to 12.5 times, providing an efficient and reliable solution for PSIR technology in time-sensitive clinical applications.

[0068] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0069] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0070] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0071] The computer program instructions for performing the operation of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, Python, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present invention.

[0072] Various aspects of the present invention are described herein with reference to the flow charts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each box of the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer-readable program instructions.

[0073] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0074] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0075] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that it is equivalent to implement it by hardware, implement it by software, and implement it by combining software and hardware.

[0076] Embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the marketplace, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A dynamic magnetic resonance delayed enhancement imaging method, comprising the following steps: For the target part, the k-space data is collected and rearranged using a phase-sensitive inversion recovery sequence to obtain rearranged data; The rearranged data is input into a trained image reconstruction model to obtain a reconstructed image, wherein the image reconstruction model is constructed based on a deep balanced model and is used to learn a mapping relationship between the rearranged undersampled data and the reconstructed image.

2. The method according to claim 1, characterized in that: The image reconstruction model includes a 3D convolution module and a 2D convolution module. The 3D convolution module is used to perform fixed point iteration, and the 2D convolution module is used to perform two-dimensional convolution to reduce the impact of image changes in each time frame on the 3D convolution.

3. The method according to claim 1, characterized in that The data rearrangement is to rearrange the acquired k-space data into a plurality of moments, each moment corresponding to under-sampled data.

4. The method according to claim 1, characterized in that The loss function for training the image reconstruction model is set to: in, represents the loss value, x label is the label image from the fully sampled data, x 2D represents the output of the 2D convolution module, x 3D represents the output of the 3D convolution module, It is a pre-trained feature extractor for deep convolutional networks.

5. The method according to claim 4, characterized in that The reconstructed image is obtained based on the following formula: Among them, x r and x m denote the reference image and amplitude image of the phase-sensitive inversion recovery sequence, respectively, x real is the true image of the phase-sensitive inversion recovery sequence, ψ(x r ) represents the phase information calculated based on the phase-sensitive inversion recovery reference image.

6. The method according to claim 4, characterized in that The deep convolutional network is a VGG network.

7. The method according to claim 1, characterized in that The image reconstruction model is constructed according to the following steps: The forward reconstruction model of the image reconstruction model is expressed as: Let x and y represent the phase-sensitive inversion recovery image to be reconstructed and the undersampled k-space data, respectively, A represents the encoding operator, λ is the regularization coefficient, represents the regularization term; For the forward reconstruction model, the proximal gradient descent method is used for iterative solution, and the iterative formula is expressed as: Among them, h is an auxiliary variable, k represents the number of iterations, η is the step size, represents the proximal operator, which is defined as: x r and x m denote the phase-sensitive inversion recovery reference image and the amplitude image, ψ(·) is the phase function, ψ(x r ) represents the phase information calculated based on the phase-sensitive inversion recovery reference image, x real is the true value image of phase-sensitive inversion recovery, z represents the result of the previous iteration, and w represents the result after the next iteration.

8. The method according to claim 7, characterized in that The iterative formula is expanded into a neural network structure, and the proximal operator is learned using a deep neural network, which is expressed as: For x k+1 , by using the fixed point equation optimization, expressed as: x k+1 =f θ (x j ;y) Among them, Γ is the learned operator, f θ is a learning operator with parameters θ.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images