Low-field magnetic resonance diffusion imaging reconstruction method and system

Abstract

The invention relates to a low-field magnetic resonance diffusion imaging reconstruction method. The method comprises the following steps: acquiring K space data of low-field magnetic resonance; mapping the acquired K space data to a standard Cartesian coordinate system by adopting a grid interpolation method of a Gaussian kernel, and weighting each radial sampling point through a Gaussian kernel function to ensure that the data is smoothly distributed on a standard grid; image reconstruction is carried out based on image prior constraints by using the interpolated K space data, and the image prior constraints comprise a low-rank structure and sparse prior information of the image; parameters of low-rank and sparse components are optimized step by step by using a mixed kernel function, and constraint is carried out by combining a low-rank structure and sparse prior information of an image, so that the image reconstruction effect is improved. The invention also relates to a low-field magnetic resonance diffusion imaging reconstruction system. According to the invention, the image quality and the anti-noise performance can be obviously improved.

Description

Technical Field

[0001] The invention relates to a low-field magnetic resonance diffusion imaging reconstruction method and system. Background Art

[0002] Magnetic resonance imaging (MRI) has become an important tool for medical imaging because it can provide high-resolution soft tissue images without the hazards of ionizing radiation. However, although low-field MRI has advantages in cost, power consumption, and shielding requirements, its lower magnetic field strength leads to weaker signal strength, which further affects the signal-to-noise ratio (SNR) of the image, especially in diffusion-weighted imaging (DWI). The decrease in signal intensity, the limitation of gradient intensity and gradient switching rate are the main challenges of low-field diffusion imaging, which greatly limits the application of low-field MRI in diffusion imaging.

[0003] To address these challenges, researchers have developed various diffusion imaging techniques, including EPI (echo planar imaging) and LSDI (line scan diffusion imaging) sequences. The EPI sequence achieves efficient imaging by rapidly acquiring data across the entire K-space, but it is prone to artifacts in low-field MRI and requires high gradient strength and gradient switching rates. In contrast, the LSDI sequence acquires data line by line and is not limited by gradient strength and gradient switching rates, but it is prone to cross-sectional artifacts due to its slow scanning speed, making it difficult to use efficiently under low-field conditions. In addition, radial acquisition methods have been explored for diffusion imaging in low-field MRI to address signal-to-noise ratio and sampling efficiency issues.

[0004] The radial acquisition scheme in high-resolution magnetic resonance obtains data by radiating from the center of the K space to the edge, and reduces artifacts and increases imaging speed through dense sampling of the central part of the K space and sparse reconstruction technology. This radial sampling mode can achieve high-density sampling in the low-frequency area, ensuring the clarity of key image information.

[0005] Existing low-field MRI diffusion imaging techniques have significant shortcomings in image quality and noise suppression. First, the diffusion imaging sequence (EPI) commonly used in high fields is difficult to overcome the limitations of gradient intensity and switching rate under low-field conditions, which makes it impossible to apply it in low-field MRI. In addition, the inherent weaker net magnetization intensity and shorter T1 relaxation time of low-field MRI systems lead to a significant reduction in signal intensity, while thermal noise remains unchanged, making the impact of noise more obvious during image reconstruction, ultimately affecting the quality and diagnostic value of the image. In diffusion imaging, the diffusion gradient signal attenuation in weak magnetic field systems is more serious, further reducing the signal-to-noise ratio, making the acquisition of high-quality images more challenging. Although the existing line scan diffusion imaging (LSDI) can perform complete sampling, it has a long scanning time, is easily affected by motion artifacts and frequency offset effects, and the clarity and stability of the reconstructed image are insufficient. Summary of the invention

[0006] In view of this, it is necessary to provide a low-field magnetic resonance diffusion imaging reconstruction method and system, which can significantly improve image quality and noise resistance performance.

[0007] The present invention provides a low-field magnetic resonance diffusion imaging reconstruction method, which comprises the following steps: S1, collecting K-space data of low-field magnetic resonance; S2, using a Gaussian kernel grid interpolation method to map the collected K-space data to a standard Cartesian coordinate system, and weighting each radial sampling point by a Gaussian kernel function to ensure that the data is smoothly distributed on the standard grid; S3, using the interpolated K-space data to reconstruct the image based on image prior constraints, wherein the image prior constraints include: a low-rank structure and sparse prior information of the image; S4, gradually optimizing the parameters of low-rank and sparse components by using a hybrid kernel function, and combining the low-rank structure and sparse prior information of the image to perform constraints, thereby improving the effect of image reconstruction.

[0008] Preferably, step S1 comprises: Step S11, establishing an adaptive K-space sampling model under low field strength conditions; Step S12, designing a low-field diffusion imaging sequence; Step S13: Acquire K-space data according to the designed sampling trajectory and time sequence.

[0009] Preferably, the step S11 comprises: The signal acquisition trajectory is determined to be bidirectional radial, the central area of ​​the K space can be repeatedly and densely sampled, and the edge ensures sparsity to reduce the acquisition time, and finally obtains the K space sampling model.

[0010] Preferably, the step S12 comprises: First, a 90-degree excitation pulse is applied, slice selection is achieved through G1, and the phase compensation gradient G2 is used to compensate for the phase error caused by the slice selection gradient. Then, gradients G3 and G13 control the acquisition direction, allowing continuous radial sampling at different angles. After passing through the center of K space, the above gradients continue to expand toward the periphery to achieve bidirectional acquisition. After the 180-degree focusing pulse, read gradients G5 and G15 are responsible for data acquisition to ensure signal coverage in the entire K space; at the same time, diffusion gradients are applied before G3 and G13 to directionally compensate for magnetic field inhomogeneity and capture anisotropic water molecule diffusion information to ensure reliable diffusion-weighted imaging signals under low fields. Gradient G4 was set after the 180-degree pulse to eliminate unwanted echoes and enhance stability, while spoiler gradients G6, G7, and G8 were used to suppress secondary echoes, flow signals, and residual spin effects.

[0011] Preferably, the step S2 comprises: Step S21: Project the collected radial K-space data onto a standard Cartesian grid using the polar coordinate to Cartesian coordinate conversion formula (1), where: is the distance from the sampling point to the center of K space, n represents the nth piece of projection data collected, θ is the step size of the projection angle change for each acquisition, FE represents the position of the abscissa in the standard Cartesian coordinate system with the center of K space as the origin, and PE represents the position of the ordinate; (1); Step S22: Kernel-based grid interpolation uses a Gaussian kernel function to weight each non-uniform sampling point so that it is smoothly distributed to nearby grid points to ensure complete coverage of the K space.

[0012] Preferably, the step S3 comprises: Step S31, establishing an optimization problem: during the reconstruction process, the K-space data is decomposed into low-rank and sparse components by solving a convex optimization problem with image prior constraints; Step S32, iterative optimization: utilizing the low-rank characteristics and sparse priors of the image, performing iterative optimization to gradually adjust the weight parameters of the low-rank and sparse components to achieve the best reconstruction effect.

[0013] Preferably, in step S31: The optimization formula is as follows: (2) in, is the nuclear norm of the low-rank component, is the L1 norm of the sparse component, and are the weight parameters of low-rank and sparse terms, respectively, is the sampling operator, It is the K-space data obtained after interpolation in step S2.

[0014] Preferably, the step S4 comprises: Step S41, define a hybrid kernel function, use an improved Gaussian-polynomial hybrid kernel function, directly introduce the prior coefficients L and S into the whale optimization algorithm, and gradually adjust the balance between the two during the optimization process to achieve the best reconstruction quality; Step S42, iterative solution, gradually iteratively adjust the kernel function parameters through the whale optimization algorithm, so that the low rank and sparse components achieve the optimal balance, thereby significantly improving the image clarity and effectively suppressing artifacts.

[0015] Preferably, in step S41: The specific formula of the mixed kernel function is as follows: (3).

[0016] The present invention also provides a low-field magnetic resonance diffusion imaging reconstruction system, which includes an acquisition module, an interpolation module, a reconstruction module, and an optimization module, wherein: The acquisition module is used to acquire K-space data of low-field magnetic resonance; The interpolation module is used to map the collected K-space data into a standard Cartesian coordinate system by using a Gaussian kernel grid interpolation method, and weight each radial sampling point by a Gaussian kernel function to ensure that the data is smoothly distributed on a standard grid; The reconstruction module is used to use the interpolated K-space data to reconstruct the image based on the image prior constraints, and the image prior constraints include: low-rank structure and sparse prior information of the image; The optimization module is used to gradually optimize the parameters of low-rank and sparse components by using a hybrid kernel function, and to perform constraints in combination with the low-rank structure and sparse prior information of the image, thereby improving the effect of image reconstruction.

[0017] This application uses a special bidirectional radial acquisition sequence to ensure dense repeated sampling of the low-frequency part of the K space and sparse sampling of the high-frequency area, thereby enhancing the stability of the fixed components and the signal strength of the diffuse components. During the reconstruction process, the K space is completed by interpolation of the kernel function. At the same time, the image low-rank and sparse priors are used for constraints, and the low-rank and sparse coefficients are optimized by the whale optimization algorithm (WOA) to balance the energy difference between the low-frequency and high-frequency components, effectively suppressing artifacts and enhancing the clarity of the image. Thereby maximizing the accuracy and noise resistance of the reconstructed image. Ultimately, this application significantly improves the image quality of low-field diffusion imaging and enhances its reliability and applicability in clinical diagnosis. The beneficial effects of this application specifically include: (1) Experimental improvement of bidirectional radial sampling sequence design: In low-field MRI diffusion imaging, in order to improve the acquisition efficiency, this application designs a bidirectional radial sampling sequence (SEDWPR). This sequence allows the radial trajectory to penetrate from one side of the K space to the other side, passing through the central area multiple times, thereby achieving high-density low-frequency signal acquisition. Through experimental comparison with traditional unidirectional radial acquisition, the bidirectional design of this application increases low-frequency redundancy, effectively improves the image's ability to resist motion artifacts, enhances sampling consistency in low-field MRI, and provides more stable data support for subsequent reconstruction.

[0018] (2) Improvement of data acquisition efficiency: Combined with experimental comparison, bidirectional radial sampling can cover the main area of ​​K space with fewer acquisition times, effectively reducing the dependence on phase encoding. By achieving dense sampling in the low-frequency area, the K space filling efficiency is optimized. The results show that this method significantly reduces the scanning time in a low-field environment, improves the overall imaging efficiency, and further improves the reconstruction accuracy by using compressed sensing reconstruction.

[0019] (3) Experimental verification of anti-artifact capability: In low-field magnetic resonance imaging, motion artifacts and non-uniform sampling artifacts are common image quality challenges. This application uses a bidirectional radial sampling sequence design so that each trajectory passes through the center of the K space from multiple angles. Experimental results show that this sampling strategy significantly enhances the image's ability to resist motion artifacts. At the same time, through comparative experiments, it is verified that under different motion conditions, the image quality of this application is more stable, further improving the reliability of low-field magnetic resonance imaging.

[0020] (4) Experimental optimization of grid interpolation reconstruction. In order to deal with the non-uniform distribution caused by radial sampling, this application introduces a Gaussian kernel grid interpolation method to smoothly map radial data to a standard Cartesian grid. Experiments show that Gaussian weighted interpolation balances the weight of each radial point, making the K-space data coverage more complete. Comparative experiments show that compared with other interpolation methods, Gaussian kernel interpolation significantly reduces non-uniform sampling artifacts and improves the reconstruction accuracy of the image.

[0021] (5) Experimental verification of low-field diffusion imaging sequence optimization. In low-field environments, traditional diffusion imaging methods have poor imaging effects due to limited gradient strength and switching rate. To this end, this application designs a diffusion-weighted imaging sequence suitable for low-field environments. By precisely controlling the time sequence and gradient combination, the gradient utilization rate under limited conditions is maximized. Experimental verification shows that the diffusion imaging quality designed in this application under low-field conditions is better than that of traditional methods, especially in terms of signal-to-noise ratio and image clarity. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the low-field magnetic resonance diffusion imaging reconstruction method of the present invention; Figure 2 A schematic diagram of spatial data sampling provided by an embodiment of the present invention; Figure 3 A schematic diagram of a kernel function interpolation method provided by an embodiment of the present invention; Figure 4 FIG. 4 is a hardware architecture diagram of the low-field magnetic resonance diffusion imaging reconstruction system of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] See also Figure 1 FIG. 1 is a flowchart of a preferred embodiment of the low-field magnetic resonance diffusion imaging reconstruction method of the present invention.

[0025] Step S1, collect K-space data of low-field magnetic resonance. That is: in a low-field magnetic resonance environment, use a bidirectional radial diffusion imaging sequence to collect K-space data. Through this strategy, it is ensured that sufficient sampling coverage is achieved in the low-frequency area to obtain enough low-frequency information for accurate reconstruction; at the same time, sparse sampling is used in the high-frequency area to reduce the amount of data collection and optimize the acquisition efficiency. This sampling method reduces the sampling time and data volume while ensuring the quality of reconstruction, improves the overall signal-to-noise ratio, and ensures that the image still has good clarity and detail performance under low-field conditions. Specifically: In K-space data acquisition, this application designs a bidirectional radial sampling sequence suitable for low-field diffusion imaging to ensure high-density low-frequency repeated sampling and sparse sampling in the edge area. Figure 2 In this embodiment, each radial trajectory passes through the center of the K space through multi-angle projection, thereby enhancing the redundancy of low-frequency data and reducing the amount of peripheral sampling, effectively reducing data acquisition time. The specific steps are as follows: Step S11, establishing a K-space sampling model: Under low field strength conditions, the present invention establishes an adaptive K-space sampling model, determines that the trajectory of signal acquisition is bidirectional radial, the central area of ​​the K-space can be repeatedly and densely sampled, and the edge ensures sparseness to reduce acquisition time. The final K-space sampling model is as follows: Figure 2 As shown in (c), it shows how the projection trajectories at multiple angles evenly cover the entire K space.

[0026] Step S12, designing a low-field diffusion imaging sequence: To adapt to the hardware limitations of the low-field MRI system, the present invention designs a bidirectional radial diffusion imaging sequence. Figure 2 In the sequence design shown in (a), a 90-degree excitation pulse is first applied through a radio frequency pulse (RF), while slice selection is achieved through G1 on the slice selection gradient (GS), and the phase compensation gradient G2 is used to compensate for the phase error caused by the slice selection gradient. Then, gradients G3 and G13 are applied to the X-axis readout gradient (GR) and the Y-axis readout gradient (GP), respectively. G3 and G13 control the pre-dephasing direction of the acquisition signal, allowing the echo signal to continue radial sampling at different angles. The above gradients continue to expand to the periphery after passing through the center of the K space to achieve bidirectional acquisition.

[0027] After the 180-degree focusing pulse, readout gradients G5 and G15 are applied to the X-axis readout gradient (GR) and the Y-axis readout gradient (GP), respectively. G5 and G15 are responsible for controlling the data acquisition of the echo signal. The areas of these two gradients are twice that of G3 and G13 to ensure the signal reading coverage in the entire K space. At the same time, in order to adapt to the diffusion effect in low-field imaging, the present invention applies diffusion gradients (such as G20 and G23, G21 and G24, and G22 and G25) before G3 and G13. These diffusion gradients directionally compensate for magnetic field inhomogeneity, capture anisotropic water molecule diffusion information, and ensure that reliable diffusion-weighted imaging signals are obtained under low fields.

[0028] In addition, gradient G4 is set after the 180-degree pulse to eliminate unnecessary echoes and enhance stability, while destruction gradients G6, G7, and G8 are used to suppress secondary echoes, flow signals, and residual spin effects. By applying diffusion gradients before G3 and G13, phase error propagation is minimized, thereby simplifying frequency encoding, improving the ability to resist motion artifacts, and enhancing the stability of the system.

[0029] Step S13, K-space data acquisition: According to the designed sampling trajectory and time sequence, a 0.16T low-field magnetic resonance system is used to collect K-space data. High-density sampling in the central area ensures complete signal acquisition and provides an accurate reference for subsequent reconstruction, while downsampling in the peripheral area effectively reduces the amount of data and creates conditions for rapid imaging. In this process, the gradient field usually changes continuously in a smooth manner without frequent direction switching, which reduces the recovery time and stabilization time of the system and effectively shortens the overall acquisition time. Finally, the data collected at each angle is stored as a column, such as Figure 2 As shown in (b).

[0030] In other embodiments, the sampling angle is calculated to adapt to various clinical needs. For example, when clinical diagnosis requires high resolution, the low-frequency sampling density is enhanced and the reconstruction algorithm is optimized to ensure complete image details; in emergency or rapid imaging, the sampling density can be reduced and the sampling speed can be increased to meet clinical timeliness.

[0031] Step S2, K-space data grid interpolation. That is, due to the non-uniformity of radial sampling, in order to obtain complete K-space data, the grid interpolation method of Gaussian kernel is used to map the collected sparse prior to the standard Cartesian coordinate system. Each radial sampling point is weighted by the Gaussian kernel function to ensure that the data is smoothly distributed on the standard grid, so as to ensure the accuracy and consistency of subsequent image reconstruction.

[0032] Since radial sampling results in incomplete K-space sampling, this embodiment uses a kernel function interpolation method, such as Figure 3As shown, the collected radial data is mapped to a standard Cartesian grid. Specifically: Step S21, conversion from polar coordinates to Cartesian coordinates, projecting the acquired radial K-space data onto a standard Cartesian grid using the conversion formula (1) from polar coordinates to Cartesian coordinates, where: is the distance from the sampling point to the center of K space, n represents the nth piece of projection data collected, θ is the step size of the projection angle change for each data collection, FE represents the position of the horizontal coordinate in the standard Cartesian coordinate system with the center of K space as the origin, and PE represents the position of the vertical coordinate.

[0033] (1) Step S22: Kernel-based grid interpolation uses a Gaussian kernel function to weight each non-uniform sampling point so that it is smoothly distributed to nearby grid points to ensure complete coverage of the K space. Figure 3 For each regular grid point, the weights of adjacent sampling points are accumulated to achieve complete K-space coverage and high-precision reconstruction, as shown in (a). Figure 3 As shown in (b).

[0034] Step S3, image reconstruction based on image prior constraints. That is, prior constraints can effectively separate the structural information and noise characteristics of the image to achieve a clearer image restoration effect. By using appropriate prior information to adjust the reconstruction model, it can be adapted to data collected at different angles and requirements, thereby improving the versatility of the algorithm and the quality of reconstruction.

[0035] This embodiment utilizes the low-rank structure and sparse prior information of the image for reconstruction to improve the image's noise resistance and detail restoration effect. The low-rank components are used to capture the stable background information of the image, while the sparse components extract dynamic features.

[0036] Specifically: Step S31, establish an optimization problem. During the reconstruction process, the K-space data is decomposed into low-rank and sparse components by solving a convex optimization problem with image prior constraints to minimize the reconstruction error. The optimization formula is as follows: (2) in, is the nuclear norm of the low-rank component, is the L1 norm of the sparse component, and are the weight parameters of low-rank and sparse terms, respectively, is the sampling operator, is the K-space data obtained after interpolation in step S2. Through this prior constraint model, the image restoration capability in a low-field imaging environment is effectively enhanced.

[0037] Step S32, iterative optimization. Using the low-rank characteristics and sparse priors of the image, iterative optimization is performed to gradually adjust the weight parameters of the low-rank and sparse components. and , in order to achieve the best reconstruction effect.

[0038] Step S4, Whale Optimization Algorithm (WOA) tuning parameters: WOA gradually optimizes the parameters of low-rank and sparse components by using a hybrid kernel function, and constrains them based on the low-rank characteristics of the image and sparse prior information to improve the image reconstruction effect. This significantly improves the noise resistance and detail restoration effect of the reconstructed image, so that the final generated image has higher clarity and diagnostic value. Through this optimization strategy, the best reconstruction quality can be achieved in different imaging scenarios, further enhancing the clinical application value of the image.

[0039] Based on the use of image prior information, this embodiment uses the Whale Optimization Algorithm (WOA) to further optimize the low-rank structure and coefficient components. The specific steps are as follows: Step S41, define a hybrid kernel function, use an improved Gaussian-polynomial hybrid kernel function, directly introduce the prior coefficients L and S into WOA, and gradually adjust the balance between the two during the optimization process to achieve the best reconstruction quality. The specific formula is as follows: (3) Step S42, iterative solution, by gradually iteratively adjusting the kernel function parameters through WOA, so that the low-rank and sparse components reach the optimal balance, thereby significantly improving the image clarity and effectively suppressing artifacts. The reconstruction error is gradually reduced in multiple iterations to obtain a high-quality image reconstruction effect.

[0040] See also Figure 4 , which is a hardware architecture diagram of a low-field magnetic resonance diffusion imaging reconstruction system 10 of the present invention. The system comprises: an acquisition module 101, an interpolation module 102, a reconstruction module 103, and an optimization module 104. Among them: The acquisition module 101 is used to acquire K-space data of low-field magnetic resonance. That is: in a low-field magnetic resonance environment, a bidirectional radial diffusion imaging sequence is used to acquire K-space data. Through this strategy, sufficient sampling coverage is ensured in the low-frequency region to obtain sufficient low-frequency information for accurate reconstruction; at the same time, sparse sampling is used in the high-frequency region to reduce the amount of data acquisition and optimize the acquisition efficiency. This sampling method reduces the sampling time and data volume while ensuring the quality of reconstruction, improves the overall signal-to-noise ratio, and ensures that the image still has good clarity and detail performance under low-field conditions. Specifically: In K-space data acquisition, this application designs a bidirectional radial sampling sequence suitable for low-field diffusion imaging to ensure high-density low-frequency repeated sampling and sparse sampling in the edge area. Figure 2 This embodiment uses multi-angle projection to realize that each radial trajectory passes through the center of the K space, thereby enhancing the redundancy of low-frequency data and reducing the amount of peripheral sampling, effectively reducing data acquisition time. Specifically including: The acquisition module 101 establishes a K-space sampling model: Under low field strength conditions, the present invention establishes an adaptive K-space sampling model to determine that the trajectory of signal acquisition is bidirectional radial, the central area of ​​the K-space can be repeatedly and densely sampled, and the edge ensures sparseness to reduce acquisition time. The final K-space sampling model is as follows: Figure 2 As shown in (c), it shows how the projection trajectories at multiple angles evenly cover the entire K space.

[0041] The acquisition module 101 designs a low-field diffusion imaging sequence: To adapt to the hardware limitations of the low-field MRI system, the present invention designs a bidirectional radial diffusion imaging sequence. Figure 2 In the sequence design shown in (a), a 90-degree excitation pulse is first applied through a radio frequency pulse (RF), while slice selection is achieved through G1 on the slice selection gradient (GS), and the phase error caused by the slice selection gradient is compensated using the phase compensation gradient G2. Then, gradients G3 and G13 are applied to the X-axis readout gradient (GR) and the Y-axis readout gradient (GP), respectively. G3 and G13 control the pre-dephasing direction of the acquisition signal, allowing the echo signal to continue radial sampling at different angles. The above gradients continue to expand to the periphery after passing through the center of the K space, realizing bidirectional acquisition.

[0042] After the 180-degree focusing pulse, readout gradients G5 and G15 are applied to the X-axis readout gradient (GR) and the Y-axis readout gradient (GP), respectively. G5 and G15 are responsible for controlling the data acquisition of the echo signal. The areas of these two gradients are twice that of G3 and G13 to ensure the signal reading coverage in the entire K space. At the same time, in order to adapt to the diffusion effect in low-field imaging, the present invention applies diffusion gradients (such as G20 and G23, G21 and G24, and G22 and G25) before G3 and G13. These diffusion gradients directionally compensate for magnetic field inhomogeneity, capture anisotropic water molecule diffusion information, and ensure that reliable diffusion-weighted imaging signals are obtained under low fields.

[0043] In addition, gradient G4 is set after the 180-degree pulse to eliminate unnecessary echoes and enhance stability, while destruction gradients G6, G7, and G8 are used to suppress secondary echoes, flow signals, and residual spin effects. By applying diffusion gradients before G3 and G13, phase error propagation is minimized, thereby simplifying frequency encoding, improving the ability to resist motion artifacts, and enhancing the stability of the system.

[0044] The acquisition module 101 performs K-space data acquisition: according to the designed sampling trajectory and time sequence, a 0.16T low-field magnetic resonance system is used to acquire K-space data. High-density sampling in the central area ensures complete signal acquisition and provides an accurate reference for subsequent reconstruction, while downsampling in the peripheral area effectively reduces the amount of data and creates conditions for rapid imaging. In this process, the gradient field usually changes continuously in a smooth manner without frequent direction switching, which reduces the recovery time and stabilization time of the system and effectively shortens the overall acquisition time. Finally, the data collected at each angle is stored as a column, such as Figure 2 As shown in (b).

[0045] The interpolation module 102 is used for grid interpolation of K-space data. That is, due to the non-uniformity of radial sampling, in order to obtain complete K-space data, a grid interpolation method with a Gaussian kernel is used to map the collected sparse priors to a standard Cartesian coordinate system. Each radial sampling point is weighted by a Gaussian kernel function to ensure that the data is smoothly distributed on a standard grid, so as to ensure accuracy and consistency during subsequent image reconstruction.

[0046] Since radial sampling results in incomplete K-space sampling, this embodiment uses a kernel function interpolation method, such as Figure 3 As shown, the collected radial data is mapped to a standard Cartesian grid. Specifically: The interpolation module 102 performs polar coordinate to Cartesian coordinate conversion, and projects the acquired radial K-space data onto a standard Cartesian grid using the polar coordinate to Cartesian coordinate conversion formula (1), where: is the distance from the sampling point to the center of K space, n represents the nth piece of projection data collected, θ is the step size of the projection angle change for each data collection, FE represents the position of the horizontal coordinate in the standard Cartesian coordinate system with the center of K space as the origin, and PE represents the position of the vertical coordinate.

[0047] (1) The interpolation module 102 is based on the grid interpolation of the kernel function, and uses the Gaussian kernel function to weight each non-uniform sampling point so that it is smoothly distributed to the nearby grid points to ensure that the K space is fully covered. Figure 3 For each regular grid point, the weights of adjacent sampling points are accumulated to achieve complete K-space coverage and high-precision reconstruction, as shown in (a). Figure 3 As shown in (b).

[0048] The reconstruction module 103 is used to reconstruct the image based on the image prior constraints. That is, the prior constraints can effectively separate the structural information and noise characteristics of the image to achieve a clearer image restoration effect. By using appropriate prior information to adjust the reconstruction model, it can be adapted to the data and requirements collected at different angles, thereby improving the versatility of the algorithm and the quality of reconstruction.

[0049] The reconstruction module 103 uses the low-rank structure and sparse prior information of the image to reconstruct the image to improve the image's anti-noise ability and detail restoration effect. The low-rank components are used to capture the stable background information of the image, while the sparse components extract dynamic features. Specifically: The reconstruction module 103 establishes an optimization problem. During the reconstruction process, the K-space data is decomposed into low-rank and sparse components by solving a convex optimization problem with image prior constraints to minimize the reconstruction error. The optimization formula is as follows: (2) in, is the nuclear norm of the low-rank component, is the L1 norm of the sparse component, and are the weight parameters of low-rank and sparse terms, respectively, is the sampling operator, It is the K-space data obtained after interpolation by the interpolation module 102. Through this prior constraint model, the image restoration capability in a low-field imaging environment is effectively enhanced.

[0050] The reconstruction module 103 performs iterative optimization. Using the low-rank characteristics and sparse priors of the image, iterative optimization is performed to gradually adjust the weight parameters of the low-rank and sparse components. and , in order to achieve the best reconstruction effect.

[0051] The optimization module 104 is used to tune parameters using the whale optimization algorithm (WOA): WOA gradually optimizes the parameters of low-rank and sparse components by using a hybrid kernel function, and constrains the low-rank characteristics of the image and the sparse prior information to improve the effect of image reconstruction. This significantly improves the noise resistance and detail restoration effect of the reconstructed image, so that the final generated image has higher clarity and diagnostic value. Through this optimization strategy, the best reconstruction quality can be achieved in different imaging scenarios, further enhancing the clinical application value of the image.

[0052] Based on the use of image prior information, the optimization module 104 uses the Whale Optimization Algorithm (WOA) to further optimize the low-rank structure and coefficient components. Specifically: The optimization module 104 defines a hybrid kernel function, uses an improved Gaussian-polynomial hybrid kernel function, directly introduces the prior coefficients L and S into the WOA, and gradually adjusts the balance between the two during the optimization process to achieve the best reconstruction quality. The specific formula is as follows: (3) The optimization module 104 performs iterative solution, and gradually adjusts the kernel function parameters through WOA to achieve the optimal balance between low-rank and sparse components, thereby significantly improving image clarity and effectively suppressing artifacts. The reconstruction error is gradually reduced in multiple iterations to obtain a high-quality image reconstruction effect.

[0053] In other embodiments, for high motion artifacts or low signal-to-noise ratio environments, Total Variation (TV) regularization or deep learning-based reconstruction techniques may be more suitable, and the above methods can provide better detail enhancement and artifact suppression effects. In high-field or ultra-high-field magnetic resonance systems, the sampling density and reconstruction strategy can be further adjusted, such as reducing repeated sampling of low-frequency areas in high-field systems to optimize sampling efficiency. In addition, the resolution and accuracy of diffusion-weighted imaging can be further improved by utilizing the higher gradient intensity and larger b-value provided by the high-field system. The present application can also be applied to magnetic resonance angiography (MRA) or functional magnetic resonance imaging (fMRI). In particular, for imaging sequences that are sensitive to motion artifacts, increasing repeated sampling in low-frequency areas can help improve image consistency and enhance motion artifact suppression effects, thereby enhancing the potential for application in cardiovascular or neurological imaging.

[0054] The present invention comprehensively solves the problems of low signal-to-noise ratio, many motion artifacts, long sampling time, etc. in low-field MRI systems through innovative bidirectional radial sampling sequence design, Gaussian kernel-based grid interpolation K-space completion method, image reconstruction based on image prior information, and application of whale optimization algorithm, thereby improving the overall image quality of low-field diffusion imaging.

[0055] This application adopts a radial sampling mode that passes through the center of K space multiple times, which improves the acquisition speed and the acquisition density of low-frequency information, and fine-tunes the image prior constraints through the whale optimization algorithm (WOA), ultimately effectively improving the image clarity and artifact suppression capabilities. This application realizes diffusion imaging of low-field MRI systems, enhancing its practicality and reliability in clinical and scientific research applications.

[0056] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above-mentioned preferred embodiments are only used to illustrate the present invention and are not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A low-field magnetic resonance diffusion imaging reconstruction method, characterized in that: The method comprises the following steps: S1, collects K-space data of low-field magnetic resonance; S2, using the grid interpolation method of Gaussian kernel, maps the acquired K-space data into a standard Cartesian coordinate system, and weights each radial sampling point by the Gaussian kernel function to ensure that the data is smoothly distributed on the standard grid; S3, using the interpolated K-space data to reconstruct an image based on image prior constraints, where the image prior constraints include: a low-rank structure and sparse prior information of the image; S4, by using the hybrid kernel function to gradually optimize the parameters of low-rank and sparse components, and combining the low-rank structure and sparse prior information of the image to constrain, the effect of image reconstruction is improved.

2. The method according to claim 1, characterized in that The step S1 comprises: Step S11, establishing an adaptive K-space sampling model under low field strength conditions; Step S12, designing a low-field diffusion imaging sequence; Step S13: Acquire K-space data according to the designed sampling trajectory and time sequence.

3. The method according to claim 2, characterized in that The step S11 comprises: The signal acquisition trajectory is determined to be bidirectional radial, the central area of ​​the K space can be repeatedly and densely sampled, and the edge ensures sparsity to reduce the acquisition time, and finally obtains the K space sampling model.

4. The method according to claim 3, characterized in that The step S12 comprises: First, a 90-degree excitation pulse is applied, slice selection is achieved through G1, and the phase compensation gradient G2 is used to compensate for the phase error caused by the slice selection gradient. Then, gradients G3 and G13 control the acquisition direction, allowing continuous radial sampling at different angles. After passing through the center of K space, the above gradients continue to expand toward the periphery to achieve bidirectional acquisition. After the 180-degree focusing pulse, read gradients G5 and G15 are responsible for data acquisition to ensure signal coverage in the entire K space; at the same time, diffusion gradients are applied before G3 and G13 to directionally compensate for magnetic field inhomogeneity and capture anisotropic water molecule diffusion information to ensure reliable diffusion-weighted imaging signals under low fields. Gradient G4 was set after the 180-degree pulse to eliminate unwanted echoes and enhance stability, while destruction gradients G6, G7, and G8 were used to suppress secondary echoes, flow signals, and residual spin effects.

5. The method according to claim 4, characterized in that The step S2 comprises: Step S21: Project the collected radial K-space data onto a standard Cartesian grid using the polar coordinate to Cartesian coordinate conversion formula (1), where: is the distance from the sampling point to the center of K space, n represents the nth piece of projection data collected, θ is the step size of the projection angle change for each acquisition, FE represents the position of the abscissa in the standard Cartesian coordinate system with the center of K space as the origin, and PE represents the position of the ordinate; （1）； Step S22: Kernel-based grid interpolation uses a Gaussian kernel function to weight each non-uniform sampling point so that it is smoothly distributed to nearby grid points to ensure complete coverage of the K space.

6. The method according to claim 5, characterized in that The step S3 comprises: Step S31, establishing an optimization problem: during the reconstruction process, the K-space data is decomposed into low-rank and sparse components by solving a convex optimization problem with image prior constraints; Step S32, iterative optimization: utilizing the low-rank characteristics and sparse priors of the image, iterative optimization is performed to gradually adjust the weight parameters of the low-rank and sparse components to achieve the best reconstruction effect.

7. The method according to claim 6, characterized in that In the step S31: The optimization formula is as follows: （2） in, is the nuclear norm of the low-rank component, is the L1 norm of the sparse component, and are the weight parameters of low-rank and sparse terms, respectively, is the sampling operator, It is the K-space data obtained after interpolation in step S2.

8. The method according to claim 7, characterized in that The step S4 comprises: Step S41, define a hybrid kernel function, use an improved Gaussian-polynomial hybrid kernel function, directly introduce the prior coefficients L and S into the whale optimization algorithm, and gradually adjust the balance between the two during the optimization process to achieve the best reconstruction quality; Step S42, iterative solution, gradually iteratively adjust the kernel function parameters through the whale optimization algorithm, so that the low rank and sparse components achieve the optimal balance, thereby significantly improving the image clarity and effectively suppressing artifacts.

9. The method according to claim 8, characterized in that In the step S41: The specific formula of the mixed kernel function is as follows: （3）。 10. A low-field magnetic resonance diffusion imaging reconstruction system, characterized in that: The system includes an acquisition module, an interpolation module, a reconstruction module, and an optimization module, wherein: The acquisition module is used to acquire K-space data of low-field magnetic resonance; The interpolation module is used to map the collected K-space data into a standard Cartesian coordinate system by using a Gaussian kernel grid interpolation method, and weight each radial sampling point by a Gaussian kernel function to ensure that the data is smoothly distributed on a standard grid; The reconstruction module is used to use the interpolated K-space data to reconstruct the image based on the image prior constraints, and the image prior constraints include: low-rank structure and sparse prior information of the image; The optimization module is used to gradually optimize the parameters of low-rank and sparse components by using a hybrid kernel function, and to perform constraints in combination with the low-rank structure and sparse prior information of the image, thereby improving the effect of image reconstruction.

Images