A non-gated method and system for motion correction in four-dimensional hepatic blood flow magnetic resonance imaging

Abstract

This invention discloses a non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method and system, belonging to the field of medical magnetic resonance imaging technology, including the following steps: Step 1, acquiring four-dimensional blood flow spatial data using undersampling mode and simultaneously recording respiratory signals; Step 2, motion correction reconstruction. This invention employs the aforementioned non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method and system. Based on adaptive data screening and grouping, intra-group soft-gated weighting, and inter-group deformation field registration, it introduces an additional auxiliary variable based on MRI physical structure applied to the spatial readout level during motion correction reconstruction. This automatically and sequentially estimates and separates residual motion-damping data, preventing it from contaminating the data fidelity of image reconstruction. While maintaining clinically acceptable and predictable scan times, it improves motion artifact suppression, enhances small vessel visualization, and improves the accuracy and consistency of blood flow quantification.

Description

Technical Field

[0001] This invention relates to the field of medical magnetic resonance imaging technology, and in particular to a non-gated four-dimensional liver blood flow magnetic resonance imaging motion correction method and system. Background Technology

[0002] The clinical application of 4D flow MRI of the liver has long been severely limited by two factors: respiratory motion artifacts and excessively long scan times. Although advanced accelerated acquisition techniques have significantly shortened scan times, the suppression of respiratory motion artifacts most often relies on respiratory gating methods based on navigation echoes, which frequently lead to prolonged and unpredictable scan times. While ungated acquisition can improve scan efficiency to nearly 100%, the lack of effective motion suppression methods will result in decreased image quality and reduced accuracy of blood flow quantification.

[0003] Other studies have explored ungated 4D flow MRI with retrospective motion inhibition, including soft-gated, Spatial motion correction, image registration-based motion correction, and higher-dimensional methods such as five-dimensional flow (5D flow) are being explored. Most of these methods are developed for cardiovascular applications and have not yet been widely applied to liver 4D flow MRI. Due to the large amplitude of abdominal respiratory movements, coupled with the small diameter and slow blood flow velocity of the portal venous system, liver 4D flow MRI is particularly sensitive to motion artifacts, making accurate motion correction exceptionally difficult.

[0004] A fundamental flaw in existing motor correction and reconstruction schemes is that most of them assume that respiratory signals have been screened and grouped (binning). Spatial data is "clean" and consistent with the same state of motion. However, in actual free breathing data collection, irregular breathing patterns, baseline drift, occasional body movements, and errors in respiratory signal extraction itself can inevitably lead to imperfect binning during the grouping process: that is, some subjects who are actually in other states of motion... Spatial readout lines are incorrectly grouped into the current motion state. These incorrectly grouped readout lines constitute outliers relative to the rest of the data in their group. Conventional compressed sensing (CS) reconstruction uses a method based on... The data fidelity term of the norm is extremely sensitive to such outlier data. Outlier readouts introduce significant motion artifacts, blurred vascular boundaries, and distortion of blood flow phase signals in reconstructed images, thereby significantly reducing the accuracy and consistency of blood flow quantification (such as net flow rate and peak velocity). This problem is particularly prominent in liver imaging with large abdominal movements and complex motion states, as well as in cases of occasional body movement or exercise-induced stress (such as overload).

[0005] In its previous work, the applicant proposed an adaptive respiratory motion correction framework, AMOCO. This framework, through variable probability sampling combined with variable Poisson disk constraints (VPD sampling), adaptive data filtering and grouping based on modified z-scores, and motion correction reconstruction combining intra-group soft-gating weighting and inter-group deformation field registration, has been able to effectively suppress respiratory motion artifacts while maintaining predictable scan times. However, adaptive data filtering can only remove motion data with excessive displacement or overall deviation before grouping, and cannot identify and remove outlier readouts caused by incorrect grouping that remain within each motion state after grouping. Intra-group soft gating only applies amplitude attenuation weighting to data deviating from the reference respiratory position, and cannot specifically detect and suppress individual readouts severely disrupted by motion. Therefore, residual outliers still contaminate the reconstruction results through data fidelity terms, limiting further improvements in motion correction quality and blood flow quantification accuracy.

[0006] In summary, existing non-gated liver 4D Flow MRI motion correction techniques lack a technical solution that can automatically and sequentially detect and suppress residual motion-damaged data (outlier readouts) caused by misgrouping during motion correction reconstruction. This results in residual motion artifacts, unclear vascular visualization, and large errors in blood flow quantification in the reconstructed images, making it difficult to meet clinical requirements for reliability and reproducibility. Summary of the Invention

[0007] The purpose of this invention is to provide a non-gated liver 4D flow MRI motion correction method and system. In existing non-gated liver 4D flow MRI motion correction technology, due to the inevitable errors or inaccuracies in respiratory signal screening and grouping, outlier readout lines caused by motion still remain within each motion state. Conventional reconstruction cannot automatically detect and suppress these outlier readout lines, resulting in significant residual motion artifacts in the reconstructed images, unclear display of small blood vessels, and insufficient accuracy and consistency in blood flow quantification.

[0008] To achieve the above objectives, the present invention provides a non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method, comprising the following steps: Step 1: Acquire four-dimensional blood flow using undersampling mode Spatial data is collected and respiratory signals are recorded simultaneously. Based on the respiratory signals, a statistical screening criterion robust to outliers is used to remove motion-damaged data, and the remaining data is divided into multiple motion states. Step 2: Motion correction and reconstruction. Through intra-group weight constraints, inter-group deformation registration, readout line-level outlier error modeling, and ADMM optimization iteration, high-precision four-dimensional motion correction and reconstruction is achieved, and the corrected 4DFlow image is finally output.

[0009] Preferably, the statistical screening criterion for robust outliers in step one is the modified Z-score criterion, which removes data points that deviate too far from the median respiratory rate. The retained data is divided into multiple motion states based on the total respiratory displacement.

[0010] Preferably, step two includes the following steps: S21, For each four-dimensional blood flow within each motion state Spatial data points are assigned soft gating weights; S22, Four-dimensional blood flow in each motion state Spatial data are merged along the time dimension, and 3D compressed sensing reconstruction is performed for each motion state. A deep learning-based deformable image registration tool is used to estimate the inter-group deformation fields of the remaining motion states relative to the reference state. ; S23, Soft-gating weights within the fusion group Intergroup deformation field Establish a motion correction and reconstruction framework, and introduce Outlier auxiliary variables in spatial readout hierarchy Automatic detection and suppression of motion damage readout lines that remain in each motion state after grouping; S24. Solve using the alternating direction multiplier method (ADMM) iteratively. S25. After reconstruction, output the motion-corrected 4DFlow image and perform second-order background phase correction.

[0011] Preferably, in S23, the outlier auxiliary variable Apply a set of sparse constraints and compute along the interior of each readout line. The norm is used to obtain the energy of the readout line, and then applied along all readout line dimensions. Norms, encouraging outlier auxiliary variables The data is sparsely distributed at the readout line level, so that its non-zero terms correspond exactly to the entire readout line that has been disrupted by the motion, thus creating outlier auxiliary variables. With the image to be reconstructed Joint estimation is performed, and reconstruction is automatically performed using measurement data during iterations. Separate as outlier auxiliary variables With valid data Effective data is used in image reconstruction.

[0012] Preferably, the four-dimensional motion correction and reconstruction problem in S23 is formulated as the following optimization problem: in: This is the optimal solution obtained by solving an optimization problem. Intra-group soft-gating weights; encoding operators Grouping each sport undersampling operator, For the first Undersampling operator corresponding to each motion state For Fourier transform, For coil sensitivity, To deform the end-expiratory reference image to the first The inter-group deformation field of each motion state; For the collected Spatial data, As an outlier auxiliary variable, The image is a four-dimensional blood flow image to be reconstructed. To determine the strength of the outlier suppression regularization, represent The 2-norm constraint, To reconstruct the sparse transformation of the image, The options include , To regularize the weights for sparse transformation, For linear sparse transformation, or , The weights of the total variation regularization term for the image. Used to calculate the total variation.

[0013] Preferably, the grouping unit in S23 includes, but is not limited to, a single readout line, several adjacent readout lines, or a collection time period.

[0014] Preferably, in S24, the image to be reconstructed and the outlier auxiliary variables are updated alternately in each outer iteration; the image The update utilizes wavelet transform and total variational regularization, with outlier auxiliary variables. The update is achieved by applying a soft threshold to each readout line.

[0015] A motion correction system for non-gated four-dimensional liver blood flow magnetic resonance imaging includes an acquisition module, an adaptive screening module, a motion estimation module, and a motion correction and reconstruction module. The acquisition module uses an undersampling mode to acquire four-dimensional blood flow data. Spatial data is collected and respiratory signals are recorded simultaneously; an adaptive filtering module removes motion-disrupted data based on respiratory signals and divides the remaining data into multiple motion states; a motion estimation module assigns soft-gated weights to each motion state and estimates the inter-group deformation field; a motion correction and reconstruction module introduces applied pressure during the reconstruction that integrates the soft-gated weights and the inter-group deformation field. Outlier auxiliary variables of spatial readout lines or their grouping hierarchies, or group sparse or equivalent structured sparse constraints, are jointly estimated with the image to be reconstructed. This automatically detects and suppresses residual motion-damaging readout lines within each motion state and outputs a motion-corrected four-dimensional blood flow image.

[0016] Therefore, the present invention employs the above-mentioned non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method and system, which has the following beneficial effects: (1) Automatically suppress residual motion-damaged data to further improve motion artifact suppression effect: Outlier readouts that cannot be handled by adaptive data filtering and grouping and remain in each motion state after grouping are automatically detected and separated from the data fidelity items by grouping sparse outlier auxiliary variables, so that the reconstruction is no longer contaminated by them, which helps to further improve the clarity of blood vessels and the visualization of small blood vessels.

[0017] (2) Improve the accuracy and consistency of blood flow quantification: The motion-corrected images reconstructed by this invention can present clearer blood vessels, thereby obtaining more accurate quantitative results and improving measurement repeatability.

[0018] (3) It does not rely on the assumption of motion source and has a wide range of applications: The outlier auxiliary variable does not make any assumptions about the cause of outlier readouts. In addition to the misgrouping of respiratory motion, it can also suppress outlier readouts caused by other sources such as occasional body motion, exercise-induced stress, radio frequency interference, and blood flow artifacts. It can be extended to various applications such as free-running cardiovascular imaging.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the motion correction method for non-gated liver four-dimensional blood flow magnetic resonance imaging according to the present invention; Figure 2 This is a comparison image of the quality of the uncorrected and corrected images in this invention; Figure 3 This is a respiratory signal diagram of different subjects in this invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0023] Example Please see Figures 1-3 This invention provides a non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method, comprising the following steps: Step 1: Liver 4D Flow Imaging.

[0024] Four-dimensional blood flow was acquired using undersampling mode during free breathing. Spatial data, sampling mode is a variable probability Cartesian sampling (VPD): using four-point velocity encoding Cartesian 4DFlow. Continuous acquisition of sequences Spatial data, in the phase encoding plane With time direction The sampling probability of each phase encoding position is determined by a combination of spatial Gaussian weighting, Poisson disk spacing constraints between adjacent samples, and probability suppression along the time direction, to fully utilize the compressibility of 4DFlow data. Simultaneously with data acquisition, the subject's respiratory motion signals are recorded using a diaphragm navigator and / or respiratory bellows for subsequent data screening and motion correction.

[0025] VPD sampling can be replaced by other variable density / Poisson disk-type Cartesian or non-Cartesian undersampling modes; the modified Z-score screening criterion can be replaced by other statistical screening criteria that are robust to outliers.

[0026] After data collection, the recorded respiratory signals were further analyzed to screen out those severely affected by exercise and those exhibiting Cheyne-Stokes respiration. Figure 3The corresponding collected data: A modified z-score criterion robust to outliers was adopted to remove data points that deviated too far from the median respiratory level. The judgment threshold was adaptively adjusted according to the dispersion of the subject's respiratory signal, with a typical removal rate of about 10% to 30%. The retained data were then divided into several respiratory motion states based on their overall respiratory displacement range, which served as input for subsequent motion correction and reconstruction.

[0027] Step 2: Motion correction and reconstruction.

[0028] For each state of motion Assigning soft gating weights to spatial data points This is to suppress residual motion within the group.

[0029] Using the average respiratory amplitude as a reference position, the data for each motion state are merged along the time dimension, and three-dimensional compressed sensing reconstruction is performed on each motion state to obtain a three-dimensional image of that state. Using the three-dimensional image of the end-exhalation state as a reference, a deep learning-based deformable image registration tool is used to estimate the inter-group deformation fields of the other motion states relative to the reference state. Deformable registration based on deep learning can be replaced by other rigid or non-rigid image registration methods.

[0030] Intra-group soft-gating weights Intergroup deformation field Establish a motion correction and reconstruction framework, and introduce an auxiliary variable. Used for explicit modeling Outliers (motion disruption) remain in spatial measurement data. Given that MRI data is acquired in readout units—each readout completed within milliseconds, which is considered instantaneous relative to the timescale of respiratory motion—motion disruption often affects the entire readout rather than a single readout. Spatial sampling point.

[0031] For auxiliary variables Apply group sparsity constraints: first calculate along the interior of each readout line. The norm is used to obtain the energy of the readout line, and then applied along all readout line dimensions. Norms, thereby encouraging The readout lines are sparsely distributed at the readout line level, so that their non-zero terms correspond exactly to the entire readout line that was destroyed by the movement. With the image to be reconstructed Joint estimation is performed, and reconstruction is automatically performed using measurement data during iterations. Separation into outlier estimation With valid data Only valid data is used in image reconstruction.

[0032] The four-dimensional motion correction and reconstruction problem is formulated as the following optimization problem (solving the image) Outlier auxiliary variables ): in: This is the optimal solution obtained by solving an optimization problem. Intra-group soft-gating weights; encoding operators Grouping each sport undersampling operator, For the first The undersampling operator corresponding to each motion state is used to describe the actual spatial sampling location acquired under that motion state. For Fourier transform, For coil sensitivity, To deform the end-expiratory reference image to the first The inter-group deformation field of each motion state; For the collected Spatial data, The image is a four-dimensional blood flow image to be reconstructed. To determine the strength of the outlier suppression regularization, These are outlier auxiliary variables, representing readouts that were removed due to motion disruption. represent The 2-norm constraint, To reconstruct the sparse transformation of the image, The options include (in (for linear sparse transformation) For sparse transformation regularization weights, or (in (used to calculate the total variation) The weights are the total variation regularization terms for the image.

[0033] For sparse outlier terms, the grouping unit can be a single readout line, or a group of several adjacent readout lines or a collection time period (shot); in addition to ADMM, other convex optimization algorithms can be used to solve the outlier auxiliary variables.

[0034] The problem is solved iteratively using the alternating direction method of multipliers (ADMM): in each outer iteration, the original problem is decomposed into several subproblems that can be solved in a closed-form solution or by a small amount of gradient descent; where the image... The update utilizes wavelet transform and total variational regularization, with outlier auxiliary variables. The update is achieved by group soft-thresholding for each readout line, thereby automatically estimating and suppressing motion-damaged data on a readout line-by-readout basis.

[0035] like Figure 2 As shown, after reconstruction, a motion-corrected 4DFlow image is output, and second-order background phase correction is performed.

[0036] Respiratory signals can come from diaphragmatic navigation echoes, breathing bags, or other respiratory monitoring methods such as camera tracking, PilotTone, and millimeter-wave radar.

[0037] Taking an ungated liver 4D flow MRI on a 5.0T whole-body MRI system as an example, combined with Figure 1 The illustrated process details the implementation of this invention in a specific embodiment. This embodiment does not use exogenous contrast agents.

[0038] Step 1: Liver 4DFlow imaging, using Cartesian 4DFlow with four-point velocity encoding. The sequences were acquired continuously under free breathing conditions, without any prospective gating applied throughout. This was to fully exploit the 4DFlow data in the phase encoding plane. With time direction The compressibility of the upper part, and to preserve sufficient center for subsequent motion correction reconstruction. In terms of spatial information, this embodiment designs a variable probability Cartesian sampling (VPD): the probability of each phase encoding position being selected is jointly determined by spatial Gaussian weighting, Poisson disk spacing constraints between adjacent samples, and probability suppression along the time direction. The three factors work together to form a unified sampling probability model, which is described in detail below.

[0039] In terms of spatial distribution, A fully sampled elliptical region is reserved at the spatial center to ensure stable image contrast; outside this central region, up to the preset outer ellipse, the probability of the phase-encoded position being selected increases with its distance from the outer ellipse. The distance to the center of space gradually decreases as the distance increases; locations outside the outer ellipse are not sampled. This embodiment makes the center... Maintain high sampling density in the space, peripheral Spatial sampling is gradually made sparser according to distance.

[0040] The probability broadening in the two phase encoding directions is controlled by preset parameters; the central elliptical region of the fully sampled area is determined based on the central coverage ratio of the Gaussian weighted energy, and the central coverage ratio can be taken in the range of about 5% to 20%, preferably about 10%.

[0041] Based on the aforementioned spatial probability distribution, to prevent the selected sample points from excessively clustering within the weighted region or leaving excessively large voids, this embodiment further applies a minimum spacing constraint that varies with location between any two adjacent sample points: the closer they are... At the spatial center, the smaller the minimum allowable spacing, the denser the sampling; the closer to the periphery, the larger the allowable minimum spacing, and the sparser the sampling, thus obtaining a central coverage that is both dense and uniformly distributed.

[0042] In addition, in order to make To achieve balanced temporal coverage across different cardiac phases, this embodiment progressively reduces the resampling probability of selected peripheral locations in subsequent time frames. This temporal probability suppression makes the sampling locations more dispersed over time, avoiding repeated sampling concentrated in a few peripheral regions.

[0043] Through the combined use of spatial Gaussian weighting, Poisson disk minimum spacing constraint, and temporal direction probability suppression, the resulting sampling pattern has the following characteristics: First, after merging the data from all time frames, it can almost completely cover the entire time frame. Space; secondly, the center within a single state of motion. The space can still achieve sufficiently dense and uniform coverage (approximately 25% to 33% of the total sampled data), thus ensuring high-quality 3D reconstruction for each subsequent motion state.

[0044] The acquisition protocol consisted of the following steps: (1) ungated 4DFlow acquisition using VPD sampling mode, simultaneously recording navigation echo signals and respirator signals (recording navigation signals added approximately 15 seconds compared to not recording); (2) navigation-gated 4DFlow acquisition using VPD sampling mode, serving as a motion correction reference for image quality assessment; and (3) 2DFlow acquisition of the portal vein (PV), right portal vein (RPV), and left portal vein (LPV), serving as a quantitative reference for hemodynamic measurements. Subjects fasted for 2 hours prior to the scan. Cardiac gating was retrospectively applied using peripheral pulse (PPG) signals.

[0045] Because the breathing behaviors of different subjects vary greatly (e.g. Figure 3As shown, some subjects exhibited Cheyne-Stokes breathing with long exhalations and short inhalations, with the data concentrated in a narrow expiratory zone. Data scattered across other respiratory positions and spanning larger displacements offered limited help in reducing motion artifacts. Therefore, this data, severely affected by motion, needed to be adaptively removed before reconstruction. To address this, this embodiment uses a modified Z-score criterion to robustly discriminate respiratory signals point-by-point: the median of the respiratory signal is used as the primary respiratory level, and the absolute deviation of the median is used to measure the degree of dispersion, thereby determining whether each time point significantly deviates from the primary respiratory range.

[0046] When the respiratory signal at a given moment deviates from the median by more than an adaptive threshold, it is considered to have deviated too far from the normal respiratory level, and the corresponding imaging data is discarded. This threshold is automatically adjusted based on the overall fluctuation of the subject's respiratory signal: the more stable the breathing, the stricter the threshold; the greater the respiratory fluctuation, the more relaxed the threshold. For common regular breathing and Cheyne-Stokes respiration, the rejection rate is usually between 10% and 30%; if the rejection rate exceeds 50%, it is considered abnormally excessive breathing, and this method is not applicable to this data (at this point, excessive data loss will lead to central...). Insufficient spatial coverage).

[0047] The retained data after screening are divided into motion states based on their total respiratory displacement. The number of states is determined by rounding up the total respiratory displacement to the nearest preset displacement range (preferably about 3~8mm, about 5mm in this embodiment). Here, it is assumed that the soft-gating weight can absorb the residual displacement within the group within about two voxels. In this embodiment, the voxel size in the head-to-foot direction is 2.5mm, typically resulting in several motion states. If the breathing bag signal is used for screening, since its reading does not directly correspond to the actual respiratory displacement amplitude, the number of motion states is fixed at 3. In this embodiment, the data is divided into an average of 2.6±0.51 motion states.

[0048] Step 2: Motion correction and reconstruction.

[0049] Intra-group motion correction: For each motion state, each Spatial data points are assigned soft-gated weights, with the average respiratory amplitude within that motion state used as a reference position. Data points closer to the reference position have higher weights, while data points farther from the reference position have lower weights. This approach retains valid data within each group while reducing the impact of large residual respiratory motion data on the reconstruction.

[0050] The weight decay rate is controlled by scaling parameters and set according to different types of respiratory signals such as navigation echoes or breathing bags, so that the weight of data within the group is kept within a reasonable range and the effective sampling is not insufficient due to excessive weighting.

[0051] Inter-group motion correction: Data from each motion state are merged along the time dimension, and 3D compressed sensing reconstruction is performed for each motion state. Using the 3D image at the end-expiratory state as a reference, a deep learning-based deformable image registration tool (such as Nodo) is used to estimate the inter-group deformation fields of the remaining motion states relative to the reference state. In this embodiment, inter-group registration can significantly reduce residual motion between different respiratory states: motion state The root mean square error between state 1 and state 1 decreased from 0.024±0.006 to 0.018±0.003, and the structural similarity (SSIM) increased from 0.888±0.05 to 0.947±0.02.

[0052] Soft-gating weights within the fusion group Intergroup deformation field Based on the motion correction and reconstruction framework, further introduction Outlier auxiliary variables in spatial readout hierarchy Automatic detection and suppression of motion-damping readout lines remaining within each motion state after grouping. Reconstruction is described as the process of reconstructing the image. Outlier variables Joint optimization: The encoding operator consists of undersampling operations for each motion group, Fourier transform, coil sensitivity, and inter-group deformation field, used to describe the difference between the reference image and the actual image after motion deformation and sampling. Relationships between spatial data. (Collected) Spatial data is explained by both valid data and outlier auxiliary variables; the newly added group sparse terms are used for each... The energy of the outlier component corresponding to the spatial readout line is the basic unit, and its sparsity is controlled by the outlier suppression strength parameter, thereby encouraging the outlier auxiliary variable to be non-zero only on a few readout lines that are disrupted by motion.

[0053] The ADMM algorithm is used for iterative solution: the image is updated alternately in each outer layer iteration. Outlier auxiliary variables The coil sensitivity is estimated using a coil sensitivity estimation method (e.g., ESPIRiT). After reconstruction, a motion-corrected 4D Flow image is output (e.g., ...). Figure 2 As shown in the figure, second-order background phase correction is performed.

[0054] Through the above implementation method, a set of non-gated liver 4DFlow reconstruction results with significantly suppressed motion artifacts, clear display of small blood vessels, and high consistency between blood flow quantification and 2DFlow reference was obtained. All non-gated acquisitions were completed in about 5 minutes and 30 seconds, and the scanning time was predictable.

[0055] A cross-platform embodiment, demonstrating the feasibility of this invention across different field strengths and platforms, allows for ungated 4D flow acquisition using compressed sensing sampling sequences on a 3.0T system (such as MAGNETOM Vida, Siemens). Only the airbag signal is recorded for motion tracking, and the motion correction reconstruction of this invention is performed according to steps one and two described above. Results show that the reconstruction using the airbag signal significantly improves image quality compared to the uncorrected reconstruction and approaches the level of a navigation-gated reference, validating the application potential of this invention across different field strengths and different vendor platforms.

[0056] Therefore, this invention employs the aforementioned non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method and system to automatically suppress residual motion-damaging data, further enhancing the motion artifact suppression effect. For outlier readoutlines that remain within each motion state after grouping and cannot be handled by adaptive data filtering and grouping, this invention automatically detects and separates them from the data fidelity terms using grouped sparse outlier auxiliary variables, ensuring the reconstruction is no longer contaminated and contributing to further improved vascular clarity and small vessel visualization. It also improves the accuracy and consistency of blood flow quantification: the motion-corrected images reconstructed by this invention present clearer blood vessels, resulting in more accurate quantitative results and improved measurement repeatability. Furthermore, it maintains high scanning efficiency and predictable scanning time: this invention uses non-gated acquisition, achieving 100% scanning efficiency; non-gated liver acquisition is completed in approximately 5 minutes and 30 seconds, with predictable scanning time. Unreliant on motion source assumptions, with broad applicability: The outlier auxiliary variable makes no assumptions about the causes of outlier readouts. Besides respiratory motion misgrouping, it can also suppress outlier readouts caused by incidental body movement, exercise-induced stress, radiofrequency interference, blood flow artifacts, and other sources, making it applicable to various applications such as free-running cardiovascular imaging. Cross-field strength and cross-platform feasibility: This invention can use both navigation echo signals and respiratory bag signals widely used in various manufacturers' MRI equipment for motion tracking and correction. Preliminary validation has been achieved on 5T and 3T systems, demonstrating its potential for cross-field strength and cross-vendor platform applications.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A non-gated four-dimensional liver blood flow magnetic resonance imaging motion correction method, characterized in that, Includes the following steps: Step 1: Acquire four-dimensional blood flow using undersampling mode Spatial data is collected and respiratory signals are recorded simultaneously. Based on the respiratory signals, a statistical screening criterion robust to outliers is used to remove motion-damaged data, and the remaining data is divided into multiple motion states. Step 2: Motion correction and reconstruction. Through intra-group weight constraints, inter-group deformation registration, readout line-level outlier error modeling, and ADMM optimization iteration, high-precision four-dimensional motion correction and reconstruction is achieved, and the corrected 4DFlow image is finally output.

2. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 1, characterized in that: In step one, the statistical screening criterion for robust outliers is the modified Z-score criterion, which removes data points that deviate too far from the median respiratory rate. The retained data is divided into multiple motion states based on the total respiratory displacement.

3. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 1, characterized in that, Step two includes the following steps: S21, For each four-dimensional blood flow within each motion state Spatial data points are assigned soft gating weights; S22, Four-dimensional blood flow in each motion state Spatial data are merged along the time dimension, and 3D compressed sensing reconstruction is performed for each motion state. A deep learning-based deformable image registration tool is used to estimate the inter-group deformation fields of the remaining motion states relative to the reference state. ; S23, Soft-gating weights within the fusion group Intergroup deformation field Establish a motion correction and reconstruction framework, and introduce Outlier auxiliary variables in spatial readout hierarchy Automatic detection and suppression of motion damage readout lines that remain in each motion state after grouping; S24. Solve using the alternating direction multiplier method (ADMM) iteratively. S25. After reconstruction, output the motion-corrected 4DFlow image and perform second-order background phase correction.

4. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 3, characterized in that: S23 on outlier auxiliary variables Apply a set of sparse constraints and compute along the interior of each readout line. The norm is used to obtain the energy of the readout line, and then applied along all readout line dimensions. Norms, encouraging outlier auxiliary variables The data is sparsely distributed at the readout line level, so that its non-zero terms correspond exactly to the entire readout line that has been disrupted by the motion, thus creating outlier auxiliary variables. With the image to be reconstructed Joint estimation is performed, and reconstruction is automatically performed using measurement data during iterations. Separate as outlier auxiliary variables With valid data Effective data is used in image reconstruction.

5. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 4, characterized in that: The four-dimensional motion correction and reconstruction problem in S23 is formulated as the following optimization problem: in: This is the optimal solution obtained by solving an optimization problem. Intra-group soft-gating weights; encoding operators Grouping each sport undersampling operator, For the first Undersampling operator corresponding to each motion state For Fourier transform, For coil sensitivity, To deform the end-expiratory reference image to the 1st The inter-group deformation field of each motion state; For the collected Spatial data, As an outlier auxiliary variable, The image is a four-dimensional blood flow image to be reconstructed. To determine the strength of the outlier suppression regularization, represent The 2-norm constraint, To reconstruct the sparse transformation of the image, The options include , To regularize the weights for sparse transformation, For linear sparse transformation, or , The weights of the total variation regularization term for the image. Used to calculate the total variation.

6. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 5, characterized in that: In S23, the grouping unit is a single readout line, several adjacent readout lines, or a collection time period.

7. The non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to claim 3, characterized in that: In S24, the image to be reconstructed and outlier auxiliary variables are updated alternately in each outer iteration; the image The update utilizes wavelet transform and total variational regularization, with outlier auxiliary variables. The update is achieved by applying a soft threshold to each readout line.

8. A motion correction system for non-gated liver four-dimensional blood flow magnetic resonance imaging, applied to the non-gated liver four-dimensional blood flow magnetic resonance imaging motion correction method according to any one of claims 1-7, characterized in that: It includes an acquisition module, an adaptive filtering module, a motion estimation module, and a motion correction and reconstruction module. The acquisition module uses an undersampling mode to acquire four-dimensional blood flow. Spatial data and synchronous recording of respiratory signals; The adaptive screening module removes motion-damaged data based on respiratory signals and divides the remaining data into multiple motion states; the motion estimation module assigns soft-gated weights to each motion state and estimates the inter-group deformation field. The motion correction reconstruction module introduces an applied force in the reconstruction that integrates soft-gated weights and inter-group deformation fields. Outlier auxiliary variables of spatial readout lines or their grouping hierarchies, or group sparse or equivalent structured sparse constraints, are jointly estimated with the image to be reconstructed. This automatically detects and suppresses residual motion-damaging readout lines within each motion state and outputs a motion-corrected four-dimensional blood flow image.

Images