Abdominal multi-shot high-definition diffusion magnetic resonance intelligent reconstruction method

Abstract

An intelligent reconstruction method for abdominal multi-shot high-definition diffusion magnetic resonance, relates to intelligent reconstruction of abdominal diffusion magnetic resonance image based on multi-shot interleaved echo sequence acquisition. It comprises the following steps: 1) acquiring frequency domain data of abdominal diffusion magnetic resonance image acquired by multi-shot interleaved echo sequence; 2) designing intelligent high-definition abdominal diffusion magnetic resonance reconstruction model; 3) constructing a large number of paired training data and using them for training of the phase segmentation network in the reconstruction model to obtain trained network parameters; 4) solving the intelligent high-definition abdominal diffusion magnetic resonance reconstruction model by using the convex set projection algorithm to obtain the final reconstruction result. Compared with the traditional reconstruction algorithm, the proposed method has better reconstruction signal-to-noise ratio and artifact suppression capability, and can realize high-definition diffusion image reconstruction.

Description

Technical Field

[0001] This invention relates to a method for intelligent reconstruction of abdominal multi-excitation high-definition diffuse magnetic resonance imaging, and more particularly to a method for intelligent reconstruction of abdominal multi-excitation high-definition diffuse magnetic resonance imaging based on the acquisition of multi-excitation staggered planar echo sequences without navigation echoes. Background Technology

[0002] Diffusion magnetic resonance imaging (DMRI) is the only non-invasive and radiation-free imaging method for detecting the movement of water molecules in living organisms, and it is widely used in clinical and neuroscience research (M. Bernstein, F. King, X. Zhou. Handbook of MRIPulse Sequences). Elsevier (2004). However, due to limitations in echo time and echo train length of rapid acquisition sequences, clinical diffusion imaging has a low signal-to-noise ratio and resolution (3-4 mm within the slice). 2 ).

[0003] Multiple excitations of staggered planar echo sequences can be used to acquire sub-millimeter ultra-high resolution diffusion data. However, due to physiological movements of the human body (such as heartbeats and respiration), there may be phase changes caused by motion between images acquired from each excitation (R. Bammer, “Basic principles of diffusion-weighted imaging”). European Journal of Radiology (A. Anderson, and J. Gore, “Analysis and correction of motion artifacts in diffusion-weighted imaging,” vol. 45, pp.169-184, 2003.), therefore a reconstruction algorithm is needed to correct the motion phase (A. Anderson, and J. Gore, “Analysis and correction of motion artifacts in diffusion-weighted imaging,” vol. 45, pp.169-184, 2003.). Magnetic Resonance in Medicine , vol. 32, pp.379-87, 1994.).

[0004] Several motion phase correction reconstruction algorithms for multi-excitation diffusion magnetic resonance imaging (MEI) have been proposed. Among them, MUSE (N(NK Chen, A. Guidon, HC Chang, and AW Song, "A robust multi-shot scan strategy for high-resolution diffusion-weighted MRI enabled by multiplexedsensitivity-encoding (MUSE),") is proposed. Neuroimage(Vol. 72, pp. 41-47, 2013.) is a representative navigation-free echo reconstruction scheme. Recently, structured low-rank matrix reconstruction methods based on smooth phase characteristics, such as MUSSELS (M. Mani, M. Jacob, D. Kelley, ...), have emerged. et al. , “Multi‐shot sensitivity‐encodediffusion data recovery using structured low‐rank matrix completion (MUSSELS),” Magnetic Resonance in Medicine , vol. 78, pp. 494-507, 2017.), PLRHM (Y. Huang, X. Zhang, H. Guo, et al. , “Phase-constrained reconstruction of high-resolution multi-shot diffusion weighted image,” Journal of Magnetic Resonance , vol. 312, pp. 106690, 2020.), PAIR (C. Qian, Z Wang, B Shi, et al .,“A paired phase and magnitude reconstruction for advanced diffusion-weightedimaging,” IEEE Transactions on Biomedical Engineering (DOI: 10.1109 / TBME.2023.3288031, 2023) and physically-based data generation schemes (Q. Yang, Z. Wang, K. Guo, 2023) et al ., “Physics-driven synthetic data learning for biomedical magneticresonance: The imaging physics-based data synthesis paradigm for artificial intelligence,” IEEE Signal Processing Magazine The intelligent reconstruction algorithm PIDD (C. Qian, Z. Wang, X. Zhang, vol. 40, no. 2, pp. 129-140, 2023.) et al., “Physics-informeddeep diffusion MRI reconstruction: Break the bottleneck of training data inartificial intelligence,” 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI) (pp. 1-5, 2023, DOI: 10.1109 / ISBI53787.2023.10230538.) These algorithms have been proposed successively. They can achieve robust high-resolution diffusion image reconstruction of the brain from four excitations.

[0005] However, the aforementioned algorithms still struggle to reliably achieve the more challenging high-definition diffusion reconstruction of the abdomen using multi-excitation methods. This is primarily because the displacement of the liver and surrounding organs is mainly driven by respiratory and cardiac movements, resulting in subvoxel phase shifts that attenuate non-diffusion signals, leading to artifacts associated with frequency-space trajectory errors. Summary of the Invention

[0006] The purpose of this invention is to address the problems of existing diffusion reconstruction methods, such as the difficulty in achieving high-definition diffusion reconstruction of abdominal multi-excitation systems, and to provide an intelligent reconstruction method for abdominal multi-excitation high-definition diffusion magnetic resonance imaging that has significant signal-to-noise ratio advantages and better artifact suppression capabilities.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solutions.

[0008] A method for intelligent reconstruction of abdominal multi-excitation high-definition diffuse magnetic resonance imaging includes the following steps:

[0009] 1) Acquire frequency domain data of abdominal diffuse magnetic resonance images acquired using a multi-excitation staggered planar echo sequence;

[0010] 2) Design of an intelligent high-definition abdominal diffusion magnetic resonance reconstruction model;

[0011] 3) Construct a large amount of paired training data and use it to train the phase segmentation network in the reconstruction model to obtain the trained network parameters;

[0012] 4) The intelligent high-definition abdominal diffusion magnetic resonance reconstruction model was solved using the convex set projection algorithm to obtain the final reconstruction results;

[0013] In step 1), the frequency domain data of the abdominal diffuse magnetic resonance imaging acquired using a multi-excitation staggered-plane echo sequence are as follows:

[0014]

[0015] in, The frequency domain data (also known as k-space) of abdominal diffuse magnetic resonance images acquired from H channels and J excitations represents the frequency code number and phase code number of a single diffuse magnetic resonance image, respectively. It is the diffuse magnetic resonance image of the target solution; It is a channel sensitivity map of the signal receiving coil used to collect data; This indicates the noise during data acquisition; It is a sampling operator determined by the frequency domain trajectory, indicating that zeros are filled at unsampled points; It is a Fourier transform operator; and These are the real number field and the complex number field, respectively.

[0016] In step 2), the intelligent high-definition abdominal diffusion magnetic resonance reconstruction model is as follows:

[0017]

[0018] in, This represents the frequency domain data of abdominal diffusion magnetic resonance images acquired from H channels and J excitations; N and M represent the number of frequency codes and the number of phase codes for a single diffusion magnetic resonance image, respectively. It is the diffuse magnetic resonance image for solving the objective. It is a channel sensitivity map of the signal receiving coil used to collect data; It is a sampling operator determined by the frequency domain trajectory, indicating that zeros are filled at unsampled points; It is a Fourier transform operator; and These are the real number field and the complex number field, respectively. and These are the squares of the Frobenius norm and the nuclear norm, respectively; and These are two regularization parameters; This represents the operator that selects the P-th region of the image domain. This represents an operator that selects a smooth phase region in the image domain. This represents an operator that selects a non-smooth phase region in the image domain. It is an operator for constructing structured low-rank matrices; It is an operator for constructing a local low-rank matrix, where c is the coordinate of the center of the selected image patch.

[0019] In step 3), a large amount of paired training data is constructed and used to train the phase segmentation network in the reconstruction model to obtain the trained network parameters:

[0020] a) Design a phase segmentation network for a multi-layer convolutional neural network.

[0021] It consists of L convolutional layers, including one input layer, one output layer, and L-2 hidden layers; each hidden layer contains Z convolutional kernels, with a kernel size of... ;

[0022] b) Construct a training dataset for training the phase segmentation network.

[0023] First, frequency domain data of high-resolution diffusion magnetic resonance imaging (MRI) with J excitations from the abdomen, comprising S layers, were acquired. A 64×64 region was selected from the center of the frequency domain of the s-th layer, and reconstructed using a traditional optimization algorithm to obtain a multi-excitation diffusion image with low resolution but high signal-to-noise ratio. ;

[0024] The phase of the diffusion image for the j-th excitation (j=1,2,…,J, traversing all excitations) Perform manual segmentation and select smooth phase regions. ;right Taking the complement in the image domain yields the non-smooth phase region. Phase and amplitude As network input, and As training labels for the network, a total of S×J pairs of training samples were obtained;

[0025] c) Use the constructed pairwise datasets for network training

[0026] The denoiser is trained using the dataset described above, and the loss function is defined as follows:

[0027]

[0028] in, It is a phase-segmentation network. This represents the set of parameters within the entire network. The parameters of a trained network can be represented as... Σ represents the summation operation;

[0029] In step 4), the convex set projection algorithm is used to solve the problem, wherein the k-th iteration is described as follows:

[0030] a) Data validation items

[0031]

[0032]

[0033] in, It is the image after data verification in the k-th iteration; It is a channel sensitivity map of the signal receiving coil used to collect data; It is the image after channel merging in the k-th iteration; This represents the frequency domain data of abdominal diffuse magnetic resonance images acquired from H channels and J excitations. These are the diffuse magnetic resonance images from the k-th iteration, with their initial values... It can be obtained by selecting a 64×64 region from the center of the frequency domain and then reconstructing it using a traditional optimization algorithm; N, M, J, and H represent the number of frequency codes, the number of phase codes, the number of excitations, and the number of receiving coil channels in a single diffuse magnetic resonance image, respectively; and These are the frequency domain sampling operator and its adjoint operator; and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and These are the real number field and the complex number field, respectively. It is a regularization parameter;

[0034] b) Phase splitting

[0035]

[0036] in, It is the image of the j-th excitation after data verification and channel merging in the k-th iteration; and These are operators for constructing structured low-rank matrices and their adjoint operators; It is a pre-trained phase segmentation network, and its weight parameters are: ; and These are the smooth and non-smooth regions segmented from the image of the j-th excitation, respectively; by sequentially segmenting J excitations, the phase segmentation results of multiple excitations are obtained. and ;

[0037] c) Structured low-rank constraints

[0038]

[0039]

[0040] in, It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. It is an unsmooth phase region; It is a singular value hard threshold truncation operator, which means performing singular value decomposition on matrix Z and retaining the first r... pA singular value, and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and Represents the operators for constructing structured low-rank matrices and their inverses;

[0041] d) Local low-rank constraints

[0042]

[0043]

[0044] in, It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. It is an unsmooth phase region; It is a singular value soft thresholding operator, which means performing singular value decomposition on matrix Z and subtracting the singular values. And retain singular values ​​greater than 0. and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and Represents the operators for constructing structured low-rank matrices and their inverses; It is a regularization parameter;

[0045] e) Update image

[0046]

[0047] This is the final reconstruction result obtained in the kth iteration;

[0048] After K iterations, a reconstructed high-resolution multi-excitation abdominal diffusion magnetic resonance image was finally obtained.

[0049] This invention addresses the limitations of existing technologies in achieving high-resolution diffusion reconstruction of the abdomen using multi-excitation methods. It offers significant advantages in signal-to-noise ratio and superior artifact suppression in abdominal high-resolution diffusion reconstruction. In particular, this invention achieves better motion artifact suppression in regions with large motion phases, such as the liver and spleen, where traditional algorithms often exhibit poor artifact suppression, enabling high-resolution diffusion image reconstruction of the liver. This invention shows great promise for application in high-resolution diffusion imaging of moving parts of the body, especially in the abdominal liver region. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the model iterative reconstruction process in an embodiment of the present invention.

[0051] Figure 2 This is the segmentation result of the phase segmentation network proposed in the embodiments of the present invention.

[0052] Figure 3 This is a comparison of the reconstruction results of the method proposed in this embodiment of the invention in high-definition diffusion imaging of the liver using four excitations, with those of a comparative method. (a) is comparative method 1, (b) is comparative method 2, (c) is comparative method 3, (d) is comparative method 4, and (e) is the method of this invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the following embodiments will be used to further illustrate this invention in conjunction with the accompanying drawings.

[0054] Example 1: Reconstruction of high-resolution diffusion imaging data of the abdominal liver from four excitations

[0055] The frequency domain data of abdominal diffuse magnetic resonance imaging acquired using a multi-excitation staggered planar echo sequence are as follows:

[0056]

[0057] in, This represents the frequency domain data of the abdominal diffuse magnetic resonance images acquired from 32 channels and 4 excitations. It is the diffuse magnetic resonance image of the target solution; It is a channel sensitivity map of the signal receiving coil used to collect data; This indicates the noise during data acquisition; It is a sampling operator determined by the frequency domain trajectory, indicating that zeros are filled at unsampled points; It is a Fourier transform operator; and These are the real number field and the complex number field, respectively.

[0058] The intelligent high-definition abdominal diffusion magnetic resonance reconstruction model is as follows:

[0059]

[0060] in, This represents the frequency domain data of the abdominal diffuse magnetic resonance images acquired from 32 channels and 4 excitations. It is the diffusion image of the objective solution. It is a channel sensitivity map of the signal receiving coil used to collect data; It is a sampling operator determined by the frequency domain trajectory, indicating that zeros are filled at unsampled points; It is a Fourier transform operator; and These are the real number field and the complex number field, respectively. and These are the squares of the Frobenius norm and the nuclear norm, respectively; and These are two regularization parameters, set to 1.0 and 1.0 respectively; This represents different regions in the image domain. This represents an operator that selects a smooth phase region in the image domain; This represents an operator that selects a non-smooth phase region in the image domain. It is an operator for constructing structured low-rank matrices; It is an operator for constructing a local low-rank matrix, where c is the coordinate of the center of the selected image patch;

[0061] A large amount of paired training data is constructed and used to train the phase segmentation network in the reconstruction model to obtain the trained network parameters:

[0062] a) Design a phase segmentation network for a multi-layer convolutional neural network.

[0063] It consists of 16 convolutional layers, including one input layer, one output layer, and 14 hidden layers; each hidden layer contains 128 convolutional kernels, and the kernel size is 5×5.

[0064] b) Construct a training dataset for training the phase segmentation network.

[0065] First, frequency domain data from 150 layers of abdominal J-excitation high-resolution diffusion magnetic resonance imaging (MRI) were acquired. A 64×64 region was selected from the center of each layer's frequency domain, and a traditional optimization algorithm was used for reconstruction, resulting in a low-resolution but high signal-to-noise ratio multi-excitation diffusion image. .

[0066] The phase of the diffusion image of each excitation is sequentially determined. Perform manual segmentation and select smooth phase regions. .right Taking the complement in the image domain yields the non-smooth phase region. .

[0067] Phase and amplitude As network input, and A total of 6,000 pairs of training samples were obtained as training labels for the network.

[0068] c) Use the constructed pairwise datasets for network training

[0069] The denoiser is trained using the dataset described above, and the loss function is defined as follows:

[0070]

[0071] in, It is a phase-segmentation network. This represents the set of parameters within the entire network. The parameters of a trained network can be represented as... ; Σ represents the L2 norm term; Σ represents the summation operation.

[0072] In step 4), the convex set projection algorithm is used to solve the problem, wherein the k-th iteration is described as follows:

[0073] a) Data validation items

[0074]

[0075]

[0076] in, It is the image after data verification in the k-th iteration; It is a channel sensitivity map of the signal receiving coil used to collect data; It is the image after channel merging in the k-th iteration; This represents the frequency domain data of the abdominal diffuse magnetic resonance images acquired from 32 channels and 4 excitations. These are the diffuse magnetic resonance images from the k-th iteration, with their initial values... It can be obtained by selecting a 16×16 region from the center of the frequency domain and then reconstructing it using a traditional optimization algorithm; and These are the frequency domain sampling operator and its adjoint operator; and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and These are the real number field and the complex number field, respectively. This is the regularization parameter, set to 1.0.

[0077] b) Phase splitting

[0078]

[0079] in, It is the image of the j-th excitation after data verification and channel merging in the k-th iteration; and These are operators for constructing structured low-rank matrices and their adjoint operators; It is a pre-trained phase segmentation network, and its weight parameters are: ; and These are the phase-smooth and non-smooth regions segmented from the image of the j-th excitation, such as... Figure 2As shown, the phase segmentation result of the multi-excitation is obtained by sequentially dividing the excitation number into four segments. and .

[0080] c) Structured low-rank constraints

[0081]

[0082]

[0083] in, It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. It is an unsmooth phase region; It is a singular value hard threshold truncation operator, which means performing singular value decomposition on matrix Z and retaining the first r... p There are 10 singular values, r1 = 10, r2 = 100; and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and This represents the operator for constructing a structured low-rank matrix and its inverse operator.

[0084] d) Local low-rank constraints

[0085]

[0086]

[0087] in, It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. It is an unsmooth phase region; It is a singular value soft thresholding operator, which means performing singular value decomposition on matrix Z and subtracting the singular values. And retain singular values ​​greater than 0, where, =0.01, =0.1; and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and This represents the operator for constructing a structured low-rank matrix and its inverse operator. This is the regularization parameter, set to 2.0.

[0088] d) Update the image

[0089]

[0090] This is the final reconstruction result obtained in the kth iteration.

[0091] After 10 iterations, a reconstructed high-resolution multi-excitation abdominal diffusion magnetic resonance image was finally obtained. A schematic diagram of the iterative reconstruction process of this invention is shown below. Figure 1 As shown.

[0092] To verify the technical effectiveness of this invention, the method of this invention is compared with four traditional reconstruction algorithms: MUSE, MUSSELS, LLR, and PAIR. A comprehensive evaluation of these algorithms' reconstruction signal-to-noise ratio and artifact suppression capabilities is conducted. Figure 3 This paper presents a comparison of the reconstruction results of the method proposed in this invention in high-resolution diffusion imaging of the liver using four excitations, with those of a contrasting method. Specifically:

[0093] Comparative method 1: MUSE (N.-k. Chen, A. Guidon, H.-C. Chang, and AW Song, "A robust multi-shot scan strategy for highresolution diffusion weighted MRI enabled by multiplexed sensitivity-encoding (MUSE)," Neuroimage, vol. 72, pp.41-47, 2013.).

[0094] Comparative method 2: MUSSELS (M. Mani, HK Aggarwal, V. Magnotta, and M. Jacob, "Improved MUSSELS reconstruction for high-resolution multi-shot diffusionweighted imaging," Magnetic Resonance in Medicine, vol. 83, no. 6, pp. 2253-2263, 2020).

[0095] Comparative method 3: LLR (Y. Hu et al., "Motion-robust reconstruction of multishot diffusion-weighted images without phase estimation through locally low-rank regularization," Magnetic Resonance in Medicine, vol. 81, no. 2, pp.1181-1190, 2019.).

[0096] Comparative method 4: PAIR (C. Qian et al., "A Paired Phase and MagnitudeReconstruction for Advanced Diffusion-Weighted Imaging," IEEE Transactions onBiomedical Engineering, vol. 70, no. 12, pp. 3425-3435, 2023.).

[0097] from Figure 3 The comparative results show that the method proposed in this invention exhibits a higher reconstruction signal-to-noise ratio. This method effectively reduces noise interference while preserving image details. Particularly when dealing with regions such as the liver and spleen, where motion phase is large and traditional algorithms struggle to effectively suppress artifacts, this invention demonstrates significant artifact suppression capabilities. The algorithm design specifically addresses motion artifacts, achieving clearer image reconstruction. Compared to MUSE, this invention not only improves the reconstruction signal-to-noise ratio but also significantly reduces motion-induced artifacts. Compared to MUSSELS, this invention further enhances image clarity while maintaining high resolution. Although the LLR method also considers motion artifact suppression, this invention performs better in this regard, especially when handling complex motion patterns. While the PAIR method provides a novel phase and amplitude reconstruction strategy, this invention still maintains a significant advantage in suppressing artifacts and preserving image quality.

[0098] In summary, the method of this invention not only outperforms traditional algorithms in terms of reconstruction signal-to-noise ratio and artifact suppression, but also exhibits higher stability and clarity when handling complex motion patterns. It performs exceptionally well in high-resolution diffusion imaging of moving parts, especially regions such as the abdominal liver, and has broad application prospects in medical diagnosis, treatment monitoring, and biomedical research. It has significant application value and potential in high-resolution diffusion imaging of moving parts.

[0099] The above embodiments are merely preferred embodiments of the present invention and should not be considered as limiting the scope of the present invention. All equivalent variations and improvements made within the scope of the present invention should still fall within the patent coverage of the present invention.

Claims

1. A method for intelligent reconstruction of abdominal multi-excitation high-definition diffused magnetic resonance imaging, characterized in that... Includes the following steps: 1) Acquire frequency domain data of abdominal diffuse magnetic resonance images acquired using a multi-excitation staggered planar echo sequence; 2) Design an intelligent high-definition abdominal diffusion magnetic resonance reconstruction model, specifically: Where Y∈D NMJH×NM This represents the frequency domain data of abdominal diffusion magnetic resonance images acquired from H channels and J excitations, where N and M represent the number of frequency codes and phase codes for a single diffusion magnetic resonance image, respectively; ρ = [ρ1, ρ2, ..., ρ J [C1, C2, ..., C] is the diffusion magnetic resonance image for solving the objective problem. H [This is a channel sensitivity map of the signal receiving coil used for data acquisition;] It is a sampling operator determined by the frequency domain trajectory, indicating that zeros are filled at unsampled points; It is a Fourier transform operator; D represents the complex field; and ||·|| * These are the squares of the Frobenius norm and the nuclear norm, respectively; λ1 and λ2 are two regularization parameters; This operator selects the p-th region of the image domain. When the superscript p = 1, This indicates an operator that selects a smooth phase region in the image domain; when the superscript p = 2, This represents an operator that selects a non-smooth phase region in the image domain. It is an operator for constructing structured low-rank matrices; It is an operator for constructing a local low-rank matrix, where c is the coordinate of the center of the selected image patch; 3) Construct a large amount of paired training data and use it to train the phase segmentation network in the reconstruction model to obtain the trained network parameters, including the following steps: a) Design a phase segmentation network for a multi-layer convolutional neural network. It consists of L convolutional layers, including one input layer, one output layer, and L-2 hidden layers; each hidden layer contains Z convolutional kernels, and the kernel size is z×z; b) Construct a training dataset for training the phase segmentation network. First, high-resolution diffusion magnetic resonance imaging (DMRI) data of the abdomen with J excitations and S layers was acquired. A 64×64 region was selected from the center of the frequency domain of the s-th layer, and reconstructed using a traditional optimization algorithm to obtain a low-resolution but high signal-to-noise ratio multi-excitation diffusion image ρ. s ; The phase P of the diffusion image of the j-th excitation is sequentially... sj =ρ sj / |ρ sj Perform manual segmentation, iterating through all excitations j = 1, 2, ..., J, and selecting smooth phase regions. right Taking the complement in the image domain yields the non-smooth phase region. Phase P sj =ρ sj / |ρ sj |and amplitude m s =|ρ sj As network input, and As training labels for the network, a total of S×J pairs of training samples were obtained; c) Use the constructed pairwise datasets for network training The denoiser is trained using the dataset described above, and the loss function is defined as follows: in, It is a phase segmentation network, where Θ represents the set of parameters within the entire network, and the trained network parameters are represented as follows: Σ represents the summation operation; 4) The intelligent high-definition abdominal diffusion magnetic resonance reconstruction model was solved using the convex set projection algorithm to obtain the final reconstruction results; The convex set projection algorithm is used to solve the problem, and the k-th iteration is described as follows: a) Data validation items H k =C * G k / C * C. Among them, G k ∈D NMJH×NMJ It is the image after data verification in the k-th iteration; It is the image after channel merging in the k-th iteration; ρ k ∈D NMJ×NM These are the diffusion magnetic resonance images from the k-th iteration, with initial values ​​ρ. 0 The image was reconstructed using a traditional optimization algorithm after selecting a 64×64 region from the center of the frequency domain; N, M, J, and H represent the number of frequency codes, the number of phase codes, the number of excitations, and the number of receiving coil channels in a single diffuse magnetic resonance image, respectively. and These are the frequency domain sampling operator and its adjoint operator; and These are the Fourier transform operator and the inverse Fourier transform operator, respectively; D represents the complex field; λ1 is the regularization parameter; b) Phase splitting in, It is the image of the j-th excitation after data verification and channel merging in the k-th iteration; and These are operators for constructing structured low-rank matrices and their adjoint operators; It is a trained phase segmentation network, and its network inputs are respectively The amplitude and phase, with weighting parameters being The symbol |·| represents taking the absolute value; and These are the smooth and non-smooth regions segmented from the image of the j-th excitation, respectively; by sequentially segmenting J excitations, the phase segmentation results of multiple excitations are obtained. and c) Structured low-rank constraints Among them, H k It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. The phase region is not smooth; SVT(Z,r) p ) is the singular value hard threshold truncation operator, which means performing singular value decomposition on matrix Z and retaining the first rp singular values. and These are the Fourier transform operator and the inverse Fourier transform operator, respectively. and Represents the operators for constructing structured low-rank matrices and their inverses; d) Local low-rank constraints Among them, H k It is the diffusion image after data verification and channel merging in the k-th iteration; It is a phase-divided region, in which, It is a smooth phase region. It is an unsmooth phase region; It is a singular value soft thresholding operator, which means performing singular value decomposition on matrix Z and subtracting ε from the singular values. p And retain singular values ​​greater than 0. and The operator for constructing a structured low-rank matrix and its inverse operator are represented; λ2 is the regularization parameter; e) Update image ρ k This is the final reconstruction result obtained in the kth iteration; After K iterations, a reconstructed high-resolution multi-excitation abdominal diffusion magnetic resonance image was finally obtained.

2. The intelligent reconstruction method for abdominal multi-excitation high-definition diffusion magnetic resonance imaging as described in claim 1, characterized in that... In step 1), the frequency domain data of the abdominal diffuse magnetic resonance imaging acquired using a multi-excitation staggered-plane echo sequence are as follows:

Images