Non-linear mapping parallel imaging reconstruction method and device based on manifold

Abstract

The invention relates to the technical field of magnetic resonance imaging, in particular to a non-linear mapping parallel imaging reconstruction method and device based on manifold. The method comprises the following steps: acquiring K space data through multiple channels; setting a neighborhood size according to the collected K space data, and adaptively optimizing and adjusting key parameters of high-dimensional mapping; non-linearly related image information and noise components of the K space data are used as feature input and mapped to a higher-dimensional feature space through a popularity method, and a topological relation of the high-dimensional feature space is established; in a high-dimensional feature space, differences between neighborhood points are smoothed through manifold regression, a nonlinear model is established to perform fitting on signal and noise features to complement a down-sampled K space part, and reversible Fourier transform is performed based on reconstructed K space data to recover to an image space. According to the method, the precision and anti-noise performance of the reconstructed image are improved, the application range of the algorithm is expanded, and the problem of low signal-to-noise ratio in low-field-intensity magnetic resonance parallel imaging is solved.

Description

Technical Field

[0001] The present invention relates to the field of magnetic resonance imaging technology, and in particular to a manifold-based nonlinear mapping parallel imaging reconstruction method and device. Background Art

[0002] Magnetic resonance imaging (MRI) has been widely used in medical diagnosis due to its advantages of high resolution of soft tissues and no radiation hazard. However, low-field MRI has the potential for clinical and scientific research applications due to its low cost and less shielding requirements, but its low signal-to-noise ratio and long scanning time have become key factors restricting its application. In order to improve the practical applicability of low-field MRI, it is necessary to shorten the scanning time as much as possible while ensuring image quality. However, traditional parallel imaging methods are difficult to effectively complement undersampled data in low-field MRI, resulting in reduced image quality and increased artifacts, further affecting the accuracy of clinical diagnosis and research.

[0003] The main problem of existing parallel imaging technology is the low signal-to-noise ratio of the reconstructed image. This is due to the reduction in the number of samples and the pathological nature of the system matrix, which makes the original noise of the ACS part amplified during the inversion process, reduces the signal-to-noise ratio of the reconstructed image, and the quality is not ideal. Among them, the most representative GRAPPA (Generalized Autocalibrating Partially Parallel Acquisitions) reconstruction algorithm reconstructs complete K-space data from incompletely sampled K-space data through a linear model. However, the GRAPPA algorithm has the following disadvantages: 1. Limited reconstruction accuracy: Since it only relies on linear terms for data fitting, it cannot fully capture the nonlinear relationship in the data, resulting in low accuracy of the reconstructed image. 2. Poor noise resistance: The linear model is more sensitive to noise, and noise may significantly affect the quality of the reconstruction results. 3. Limited applicability: For complex MR data, the linear model is difficult to provide sufficient reconstruction quality, which limits its application in high-resolution and complex imaging tasks. 4. Insufficient data utilization: The high-order information in the acquired K-space data is not fully utilized, resulting in information loss during the reconstruction process. Summary of the invention

[0004] The embodiment of the present invention provides a manifold-based nonlinear mapping parallel imaging reconstruction method and device to at least solve the technical problem of slow speed of existing magnetic resonance imaging.

[0005] According to an embodiment of the present invention, a nonlinear mapping parallel imaging reconstruction method based on manifold is provided, comprising the following steps: S101: multi-channel acquisition of K-space data; S102: setting the neighborhood size according to the actually acquired K-space data, and adaptively optimizing and adjusting the key parameters of the high-dimensional mapping; S103: Using the nonlinearly related image information and noise components of the K-space data as feature input, mapping them to a higher-dimensional feature space through an epidemiological method, and establishing a topological relationship in the high-dimensional feature space; S104: In the high-dimensional feature space, the differences between data points in the neighborhood and the differences between center points in different fields are smoothed by manifold regression, and finally a nonlinear model that minimizes noise features is established to fit the K-space part that is complemented and downsampled, and the reversible Fourier transform based on the complemented K-space data is restored to the image space.

[0006] Furthermore, in step S101 , multiple receiving coils are used to collect K-space data in parallel in a low-field magnetic resonance system.

[0007] Furthermore, step S101 specifically includes: Establish a K-space model under low-field magnetic resonance conditions and determine the acquisition trajectory of echo signals in the K-space model; Determine the time series parameters of the low-field magnetic resonance scan according to the acquisition trajectory, and calculate the encoding gradient required to be applied by the magnetic resonance imaging system; Acquire K-space data for parallel imaging.

[0008] Furthermore, in step S102, the neighborhood size is determined according to actual data characteristics, and the key parameters of the high-dimensional mapping are adaptively optimized and adjusted, and the key parameter is the kernel width.

[0009] Furthermore, step S102 specifically includes: Analyze the K-space data, determine the size of each neighborhood divided by the high-dimensional mapping from the ACS partial data, and generate data analysis results; Select the initial kernel width based on the data analysis results; During the reconstruction process, the kernel width is dynamically adjusted through the error feedback mechanism for adaptive iterative optimization.

[0010] Furthermore, step S103 specifically includes: The nonlinearly related image information and noise components of K-space data are used as feature inputs, mapped to a higher-dimensional feature space through epidemiological methods, and the nonlinear relationship features of the high-dimensional space are obtained; the difference between the actual sampling value of the missing position of the ACS and the high-dimensional mapped feature value of the linear combination is minimized to select the nonlinear relationship features and conduct manifold modeling; Adjust the high-dimensional feature space to balance computational complexity and manifold model performance; Construct the topological relationship of high-dimensional feature space, and use the adjusted high-dimensional data to deeply explore the complex associations between each neighborhood kernel in manifold, and combine multi-level manifold feature analysis to reveal the nonlinear relationship between signals and noise, and between different tissues.

[0011] Furthermore, in step S104, smoothing the differences between data points in the neighborhood by kernel regression includes: using manifold regression to smooth the data in the neighborhood so that the signal strength is more consistent in the local area.

[0012] Furthermore, in step S104, after smoothing the differences between the data in the neighborhood by manifold regression, the method further includes: performing nonlinear enhancement on kernel center points in different neighborhoods by using a regression model to make the signal strength more consistent in the global region.

[0013] Further, in step S104, the nonlinear relationship between the signal and the noise is processed in the high-dimensional feature space by manifold regression, and finally a nonlinear model with minimized noise features is established to fit the K-space part of the complement downsampled K-space data, and the reversible Fourier transform based on the complemented K-space data is restored to the image space, including: In the interpolation and reconstruction process, the unsampled points in the K space are first interpolated using the manifold regression results obtained by adaptive iterative optimization of ACS data to complete the downsampled data; manifold regression fitting is performed in high-dimensional space to further reduce the interpolation error; then, the interpolated data is combined with the existing sampling points to generate complete K space data; finally, the reconstructed K space data is converted to image space through IFT.

[0014] According to another embodiment of the present invention, a nonlinear mapping parallel imaging reconstruction device based on manifold is provided, comprising: An acquisition unit, used for acquiring K-space data through multiple channels; A parameter optimization unit sets the neighborhood size according to the acquired K-space data and adaptively optimizes and adjusts the key parameters of the high-dimensional mapping; The mapping unit takes the nonlinearly related image information and noise components of the K-space data as feature inputs, maps them to a higher-dimensional feature space through an epidemiological method, and establishes a topological relationship in the high-dimensional feature space; The fitting recovery unit is used to smooth the differences between data points in the neighborhood and the differences between center points in different fields in the high-dimensional feature space through manifold regression, and finally establish a nonlinear model that minimizes noise features to fit the completed downsampled K-space part, and restore it to the image space based on the reversible Fourier transform of the completed K-space data.

[0015] A storage medium stores a program file capable of implementing any one of the above-mentioned manifold-based nonlinear mapping parallel imaging reconstruction methods.

[0016] A processor is used to run a program, wherein when the program is run, any one of the above-mentioned manifold-based nonlinear mapping parallel imaging reconstruction methods is executed.

[0017] The nonlinear mapping parallel imaging reconstruction method and device based on manifold in the embodiment of the present invention maps data into a high-dimensional space without explicitly calculating high-dimensional features. This method allows for a better separation of signal and noise, while being able to capture complex imaging features, thereby being able to more accurately complete the missing sampled K-space information and more accurately restore image details. By adding interaction terms or nonlinear features, the model can more carefully identify potential risk patterns, improve the accuracy and noise resistance of reconstructed images, expand the scope of application of the algorithm, solve the problem of low signal-to-noise ratio in low-field strength magnetic resonance parallel imaging, and significantly improve the quality and diagnostic value of reconstructed images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 It is a flow chart of the nonlinear mapping parallel imaging reconstruction method based on manifold of the present invention; Figure 2 This is a schematic diagram of the K-space data acquisition method, showing the way of scanning data. One part collects the data of the central part of the K-space, and the other part collects the data of the entire K-space with uniform downsampling. The echo chain calibration technology is used to correct the error between the two. Figure 3 It is the flow chart of low-field parallel imaging based on manifold; Figure 4 This is a comparison diagram of the GRAPPA method and the nonlinear modeling method based on manifold in water film imaging, where: Figure 4 a in the figure is the reference image reconstructed using the SOS method based on the full acquisition data. Figure 4 b and Figure 4 The d in the figure are the reconstructed image and its residual image by GRAPPA method, Figure 4 c and Figure 4 The e in the figure are the reconstructed image and its residual image by the manifold nonlinear modeling method; Figure 5 is the comparison result of the GRAPPA method and the nonlinear modeling method based on manifold in human brain imaging, where: Figure 5 a in the figure is the reference image reconstructed using the SOS method based on the full acquisition data. Figure 5 b and Figure 5 The c in the figure are the reconstructed image and its residual image by GRAPPA method, Figure 5 d and Figure 5 The e in the figure are the reconstructed image and its residual image by the manifold nonlinear modeling method; Figure 6 It is a module diagram of the nonlinear mapping parallel imaging reconstruction device based on manifold in the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] Example 1 According to an embodiment of the present invention, a nonlinear mapping parallel imaging reconstruction method based on manifold is provided. Figure 1 , including the following steps: S101: multi-channel acquisition of K-space data; S102: setting the neighborhood size according to the acquired K-space data, and adaptively optimizing and adjusting key parameters of high-dimensional mapping; S103: Using the nonlinearly related image information and noise components of the K-space data as feature input, mapping them to a higher-dimensional feature space through an epidemiological method, and establishing a topological relationship in the high-dimensional feature space; S104: In the high-dimensional feature space, the differences between data points in the neighborhood and the differences between center points in different fields are smoothed by manifold regression, and finally a nonlinear model that minimizes noise features is established to fit the K-space part that is complemented and downsampled, and the reversible Fourier transform based on the complemented K-space data is restored to the image space.

[0022] The nonlinear mapping parallel imaging reconstruction method based on manifold in the embodiment of the present invention maps data into a high-dimensional space without explicitly calculating high-dimensional features. This method allows for a better separation of signal and noise, while being able to capture complex imaging features, thereby being able to more accurately complete the missing sampled K-space information and more accurately restore image details. By adding interaction terms or nonlinear features, the model can more carefully identify potential risk patterns, improve the accuracy and noise resistance of reconstructed images, expand the scope of application of the algorithm, solve the problem of low signal-to-noise ratio in low-field strength magnetic resonance parallel imaging, and significantly improve the quality and diagnostic value of reconstructed images.

[0023] Specifically, in order to solve the problem of slow magnetic resonance imaging, parallel imaging technology is a major breakthrough in magnetic resonance imaging technology. This method can accelerate the magnetic resonance imaging (MRI) scanning process by using multiple radio frequency coils to receive signals simultaneously. These coils cover different spatial regions and can capture more spatial information, thereby reducing the required k-space (frequency domain space) sampling points. This method can significantly reduce scanning time, improve the temporal resolution of images, and improve patient comfort. However, reducing sampling will increase the complexity of image reconstruction, such as obvious aliasing artifacts, so an effective reconstruction algorithm is required to restore high-quality images.

[0024] In low-field magnetic resonance imaging, the signal-to-noise ratio (SNR) is low, which further limits the scanning speed and image quality. In this case, the manifold method can perform projection operations in high-dimensional space to separate the signal and noise more closely, while capturing complex imaging features, thereby more accurately restoring the details of the image and more accurately completing the missing sampled K-space information, significantly improving the quality and diagnostic value of the reconstructed image.

[0025] In view of the above shortcomings, the present invention proposes a nonlinear mapping parallel imaging reconstruction method based on manifolds, which is suitable for the method of low-field magnetic resonance image reconstruction, and aims to solve the problems of low signal-to-noise ratio and slow imaging speed in low fields. The present invention adopts multi-channel acquisition and nonlinear modeling based on manifolds to improve the clarity and contrast of the image by more accurately completing the K-space data. Specifically, the present invention maps the data into a high-dimensional space through this method without explicitly calculating the high-dimensional features. This method allows the signal and noise to be more separated, and at the same time can capture complex imaging features, so that the K-space information of the missing samples can be more accurately completed, and the details of the image can be more accurately restored. By adding interaction terms or nonlinear features, the model can more carefully identify potential risk patterns, improve the accuracy and noise resistance of the reconstructed image, expand the scope of application of the algorithm, solve the problem of low signal-to-noise ratio in low-field strength magnetic resonance parallel imaging, and significantly improve the quality and diagnostic value of the reconstructed image.

[0026] The basic contents of the technical solution of this method are as follows: K-space data is collected in a low-field magnetic resonance system. Multiple receiving coils are used to collect K-space data in parallel. The data collected by each coil is undersampled (that is, there are data points that are not collected), but the K-space data in the center of each channel is guaranteed to be fully collected. These data are called autocalibration signals (ACS lines). These ACS lines provide complete K-space information for nonlinear modeling.

[0027] Adaptive parameter optimization. In this process, adaptive optimization parameters (such as kernel width and neighborhood size) can be adjusted according to actual data characteristics to adapt to different imaging conditions and noise levels, further improving the flexibility and applicability of the model.

[0028] Based on manifold nonlinear modeling. Since the collected K-space ACS data contains nonlinearly related image and noise features, it is initially used as a feature and mapped to a higher dimension through the manifold method. By adjusting the mapped high-dimensional feature space, the noise is separated more and the noise interference is minimized. Finally, the topological relationship of the high-dimensional feature space is established, which helps to extract the real structural features of the image more clearly.

[0029] Manifold regression and image reconstruction. In high-dimensional feature space, manifold regression smoothes the differences between data points in the neighborhood and the differences between center points in different areas, thereby establishing a nonlinear model that minimizes noise features to fit the K-space part that completes the downsampled part. Finally, based on the reversible Fourier transform (IFT) of the reconstructed K-space data, this information is restored to the image space to generate a higher quality image.

[0030] See also Figure 2, the technical solution of the present invention is described in detail as follows: 1. Acquisition of K-space data in low-field MRI systems The schematic diagram of K-space data acquisition method is as follows Figure 3 As shown in the figure, this sampling method strictly follows the Nyquist sampling theorem for full sampling in the ACS part, and reduces the sampling in the part outside the ACS, ensuring that the low-resolution complete K-space information is provided for nonlinear modeling while shortening the sampling time. The specific steps are as follows: Step 1: Establish a K-space model under low-field magnetic resonance conditions and determine the acquisition trajectory of the echo signal in the model.

[0031] Specifically, the acquisition trajectory of the echo signal is as follows: all echo signals are collected in parallel along one coordinate direction, wherein the central area of ​​the K space is fully sampled (ACS part), strictly following the Nyquist sampling theorem to ensure that complete high-resolution data is obtained for reconstruction coefficient estimation, while providing an accurate benchmark for subsequent image reconstruction; the edge area of ​​the K space is downsampled to reduce the amount of data collection and increase the acquisition speed.

[0032] Step 2: Determine the time sequence of the ultra-low field magnetic resonance scan according to the acquisition trajectory, and calculate the encoding gradient required to be applied by the magnetic resonance imaging system. The specific steps include: calculating the scanning time sequence parameters and encoding gradient of the full sampling area (ACS part) to ensure the integrity of the data acquisition in the central area; calculating the scanning time sequence parameters and encoding gradient of the down-sampling area to ensure the uniformity and efficiency of the data acquisition in the edge area. In a low-field environment, the signal-to-noise ratio is low, so it is necessary to specially optimize the time sequence and encoding gradient to maximize the signal strength and data acquisition efficiency, such as the FSE sequence with a fast acquisition speed adopted in the present invention.

[0033] Step 3: Collect K-space data for parallel imaging. Specifically, the three-dimensional multi-layer K-space data collected in this embodiment includes full sampling data of the central area of ​​each layer and down-sampled data of the edge area: Full sampling of the central area (ACS part): Full sampling of the central part of the K-space is performed according to the Nyquist sampling theorem to ensure the acquisition of complete high-resolution data. This is especially important for low-field environments. Downsampling of the edge area: Uniform downsampling is performed on the edge of the K-space to reduce data acquisition time. First, dense phase encoding acquisition is performed on the adaptive calibration scan (ACS) part to obtain complete central K-space data. Then, the phase encoding interval is gradually increased to make the phase encoding steps sparse and expand the range of phase encoding, thereby acquiring a uniformly downsampled K-space data.

[0034] 2. Adaptive parameter optimization In the kernel method of manifold, kernel width determines the projection effect of data blocks in high-dimensional feature space, which directly affects the accuracy of signal and noise separation. For low signal-to-noise ratio environments, increasing kernel width can reduce the impact of noise, make similar signals more concentrated, and enhance signal consistency; while in high signal-to-noise ratio environments, moderately reducing kernel width helps capture more subtle local features and details. The specific process is as follows: Step 1: Data analysis: First, analyze the K-space data to determine the "kernel" size, that is, the size of each neighborhood divided from the ACS partial data by high-dimensional mapping.

[0035] Step 2: Initial kernel width setting. Select the initial kernel width based on the data analysis results. , determines the initial dimension size of the high-dimensional mapping of the data.

[0036] Step 3: Adaptive iterative optimization. In each iteration, use the current estimate To calculate , and then calculate the approximate (i.e. according to the current The value is calculated by empirical formula 2 Then, using the new estimate To calculate the next step This step updates the kernel width by the least squares method. , the specific formula is as follows: [1] Among them, A represents the matrix composed of the non-undersampled position points of ACS, and the size of matrix A is m×k, where m is the total number of "kernels" divided in ACS data, and k is the number of neighborhood points in each "kernel". Indicates that the kernel width is used The epidemiological method performs high-dimensional mapping on A. The combination of estimates of the linear representation of the high-dimensional mapping of each data point at the non-undersampled position of the ACS, with dimension ,in is the dimensionality of the new feature space, which is usually much higher than k.

[0037]

[0038] Formula 2 is the empirical formula for high-dimensional mapping of each sampling point in A, where Indicates that the target coil The K-space signal of the data at the missing position of the upper ACS part, represents the acquired undersampled signal, where is the coil index, t and h represent and The offset of the adjacent K-space data in the direction, R represents the undersampling rate (ORF, sampling rate factor), that is, the undersampling factor, and denote the coordinates along the frequency encoding and phase encoding directions, respectively. and is the basic sampling interval in K space, which is used to describe the distance between points in the frequency encoding and phase encoding directions, It is an integer offset along the phase encoding direction, which determines the position of the target point. n represents the offset of n frequency encoding intervals from the adjacent K-space data. The larger the n, the more nonlinear relationship between higher-dimensional K-space data can be extracted. The conventional empirical formula 2 generally sets n to 3 to reduce the amount of calculation.

[0039] 3. Nonlinear modeling based on manifold All existing GRAPPA derivative algorithms are based on the linear model in formula [3]. Due to the low signal-to-noise ratio of low-field magnetic resonance, the nonlinear correlation between signal and noise in ACS data is difficult to separate, resulting in model deviation.

[0040]

[0041]

[0042] in Indicates that the target coil The K-space signal that is not acquired on represents the acquired undersampled signal, where is the coil index, t and h represent and The offset of the adjacent K-space data in the direction, represents the combination coefficients of the linear model, which are used to linearly combine the adjacent sampled signals into the unsampled signals on the target coil, L represents the total number of coils, R represents the undersampling rate (ORF, sampling rate factor), that is, the undersampling factor, l traverses the index of all coils, and They represent the coordinates along the frequency encoding and phase encoding directions respectively, and r is the integer offset along the phase encoding direction. Formula [3] only performs a linear combination in one dimension, which does not completely separate the image from the noise and results in a lot of image information being lost.

[0043] In order to solve the common nonlinear and noise-like error problems in the GRAPPA reconstruction process, this paper proposes an innovative method based on manifolds to improve the reconstruction accuracy by identifying and adjusting the nonlinear relationship between image features. This method uses the statistical characteristics of the data and the trend distribution in the sample space to effectively suppress noise and artifacts, thereby achieving more accurate signal reconstruction.

[0044] Step 1: Feature selection and nonlinear modeling. First, the nonlinear relationship features mapped to high-dimensional space are selected and modeled in manifold. For example, a method in manifold (kernel method) is used to perform high-dimensional mapping to separate the signal and noise of the nonlinear relationship in order to identify the error source in the reconstruction process. A typical kernel (inhomogeneous polynomial kernel) is as follows: [4] in, and is a scalar, is the order of the polynomial (also the kernel width of the kernel function for non-homogeneous polynomial kernel functions), which helps to map nonlinear relationships in the data to a high-dimensional space so that they can be separated. Indicates the first extracted from the ACS partial data of all channels and a "kernel", .

[0045] The present invention uses nonlinear mapping Enter the space The data in (ACS un-undersampled position data, i.e., initial features) are mapped to a high-dimensional feature space (High-dimensional features that can separate nonlinear signals and noise), use formula 4 to The linear combination after nonlinear mapping in high dimension is expressed as ,but: [5] Based on the least squares method of formula 5, the coefficients of each dimension between the high-dimensional mapping of the non-undersampled position data matrix A to the undersampled position data matrix B are calculated, and then the parameters used by the manifold model are obtained by combining formulas [2] and [4]. Here, it specifically refers to the kernel function and These two parameters.

[0046] Improper parameter settings may result in very high or even infinite dimensions of the mapping, so it is necessary to select high-dimensional features. In this example of the inhomogeneous polynomial kernel, it is found that retaining dimensional features above the quadratic term is difficult to further reduce the manifold model error, so it is discarded.

[0047] Step 2: Adjustment of feature space. Assume that a non-homogeneous polynomial kernel is used and =1, =1, =2, then the nonlinear mapping of manifold It can be specifically expressed as:

[0048] Using the nonlinear mapping of the above form , which helps to effectively process nonlinear inseparable data in high-dimensional space, where for Specifically, the k elements contained in "kernel" are: This mapping form contains first-order and second-order terms to preserve the basic information and nonlinear characteristics of the original data in high-dimensional space, which is particularly suitable for observing characteristic trends in manifold modeling. Preserve the underlying manifold trend of the data, the second-order terms The square sum interaction of each feature is captured, which enables the mapped feature space to reveal the nonlinear manifold trend. The higher the mapping order, the more complete the separation of nonlinear related information, but at the same time the calculation is more complicated. Adjusting the mapping to a suitable feature space is the key to this step. There are three quadratic terms, see formula [2]. To further reduce the complexity of the calculation, only the first two quadratic terms can be retained without affecting the new energy of the manifold model.

[0049] Step 3: Establish the topological relationship of the high-dimensional feature space. When constructing the topological relationship of the high-dimensional feature space, the adjusted high-dimensional data is used to deeply explore its complex associations in manifold. This process combines multi-level manifold feature analysis to reveal the potential patterns of the image and the local structure of the data. Modeling the topological relationship of the high-dimensional feature space helps capture the nonlinear relationship between signal and noise and between different tissues. As shown in formula [7], each neighborhood "kernel" divided in the ACS partial data of multiple channels is correlated with each other. The nth "kernel" is marked out in the ACS data of all channels. ACS data for all channels except Other "kernel" ( ) relative to The integer offset further affects , Formula 7 represents the mutual influence between all the “kernels” of all channels, where The remaining neighbors Composition matrix and coefficient matrix Finally, formulas [2] [4] [5] and [7] are combined to further integrate the relationship between different neighborhood "kernels" of the same channel, as well as the relationship between different neighborhood "kernel" data of different channels, to further improve the accuracy of the manifold model.

[0050]

[0051] 4. Manifold regression and signal reconstruction A manifold regression model is constructed to handle the nonlinear relationship between signal and noise in a high-dimensional feature space. The specific steps are as follows: Step 1: Smoothing: Manifold regression is used to smooth the data in the neighborhood to make the signal strength more consistent in the local area.

[0052] Step 2: Nonlinear enhancement of the center point: The center points of different “kernels” are nonlinearly enhanced through the regression model to make the signal strength more consistent in the global area.

[0053] Step 3: Interpolation and reconstruction. In the interpolation and reconstruction process, the unsampled points in the K space are interpolated using the manifold regression results obtained by adaptive iterative optimization of ACS data to complete the downsampled data; manifold regression fitting is performed in high-dimensional space to further reduce the interpolation error; then, the interpolated data is combined with the existing sampling points to generate complete K space data; finally, the reconstructed K space data is converted to image space through IFT.

[0054] The key points and intended protection points of the present invention are: The core innovation of the present invention is to propose a nonlinear mapping parallel imaging reconstruction method based on manifolds, which provides an effective technical solution to improve image quality and accelerate the scanning process for the problems of low signal-to-noise ratio and slow imaging speed in low-field magnetic resonance imaging. The following are the key innovations of the present invention and the technical contents to be protected: Nonlinear mapping method based on manifold. By adopting the nonlinear mapping technology in manifold, K-space data is mapped to high-dimensional space, so as to separate the nonlinearly related image and noise features in the original data acquired by low-field magnetic resonance. This method can overcome the shortcomings of traditional linear methods under low signal-to-noise ratio and improve the accuracy of image reconstruction. The present invention takes the kernel method in manifold as an example, and the high-dimensional feature space mapping methods of manifold such as polynomial feature expansion and high-dimensional representation in deep learning are also applicable to the present invention.

[0055] Combination of multi-channel acquisition and manifold modeling. The present invention combines multi-channel parallel acquisition with nonlinear modeling based on manifolds, receives signals simultaneously through multi-channel radio frequency coils, reduces K-space sampling points, and accurately completes undersampled data through nonlinear modeling. This technical solution significantly improves the image quality of low-field magnetic resonance imaging, especially in terms of signal-to-noise ratio and detail expression of reconstructed images.

[0056] Effective separation of signal and noise. The original data collected by low-field magnetic resonance imaging is mapped in high dimensions using manifold methods, effectively separating nonlinearly related image and noise features into a linearly combinable form. Modeling is performed by removing the "kernel" of the noise term to avoid the adverse effects of noise amplification in traditional methods. By separating the noise features, the present invention can more accurately restore image details and significantly improve image clarity and contrast.

[0057] Accurately complete undersampled K-space data. The present invention uses epidemiological regression technology to accurately calculate the coefficients of high-dimensional image features, and combines data smoothing in the neighborhood and nonlinear enhancement of center point data to more accurately restore undersampled K-space data in low-field magnetic resonance imaging, reducing image artifacts caused by data loss in traditional methods. This technical solution effectively improves the reconstruction accuracy of low-field magnetic resonance imaging.

[0058] Enhanced noise immunity and adaptability. The present invention introduces a manifold reconstruction method to complement the downsampled K-space data, so that the model can maintain strong noise immunity when dealing with noise in low-field magnetic resonance imaging. This innovation significantly broadens the scope of application of this method in various magnetic resonance imaging tasks, and is also applicable to complex high-resolution imaging tasks.

[0059] Kernel function selection and optimization iterative process. In the kernel method of the present invention, the selection of kernel function is crucial to the performance of high-dimensional mapping. In order to ensure the best imaging quality and reconstruction effect, the present invention allows the selection of multiple kernel functions, including non-homogeneous polynomial kernels, Gaussian kernels (RBF kernels), and hybrid kernel functions. By evaluating the performance of different kernel functions and combining the requirements of specific tasks, the best kernel function is selected iteratively automatically or through an optimization algorithm (such as alternating least squares method, gradient descent method, etc.), and its related parameters are tuned.

[0060] Sequence scanning model. The sequence used in the present invention has a fast scanning speed and can collect the central part full sampling and uniformly downsampled K-space data at the same time. The present invention is divided into two scans, one for scanning the central part data and one for scanning the downsampled data, and finally calibrated by the echo chain. Including but not limited to fast spin echo (FSE), gradient echo (GRE), balanced steady-state free precession (bSSFP), etc., such MRI sequences are applicable to the present invention.

[0061] Compared with the traditional GRAPPA method, the present invention significantly improves image quality, data completion capability and computational efficiency by introducing manifold nonlinear modeling methods and combining the application environment of low-field magnetic resonance, enhances adaptability to complex imaging environments, improves signal-to-noise ratio and image stability, and has important application value and promotion prospects. The details are as follows: (1) Stronger nonlinear modeling capabilities. GRAPPA relies on linear models and cannot effectively capture nonlinear relationships in data. In contrast, the present invention uses nonlinear manifold mapping to more accurately capture complex imaging features and improve image reconstruction accuracy.

[0062] (2) Better noise resistance. GRAPPA tends to amplify noise in low signal-to-noise ratio data, resulting in a decrease in image quality. Through the manifold method, the present invention can effectively separate signals and noise, improve noise resistance, thereby reducing artifacts and improving image clarity.

[0063] (3) More accurate K-space data completion. GRAPPA uses linear interpolation to fill in missing K-space data, which has limited accuracy. The present invention can more accurately fill in missing K-space data and retain more imaging details through high-dimensional nonlinear mapping.

[0064] (4) Higher adaptability. The applicability of GRAPPA is limited, especially when processing complex MR data. The present invention supports a variety of kernel functions (such as Gaussian kernel, non-homogeneous polynomial kernel, etc.), which can be flexibly adjusted according to data characteristics to adapt to different imaging tasks.

[0065] (5) Higher computational efficiency. Although the computational complexity of nonlinear methods is relatively high, the present invention can maintain high computational efficiency while maintaining image quality through manifold mapping and high-dimensional modeling, especially performing better on large data sets.

[0066] (6) Optimized data utilization and information recovery. GRAPPA does not fully utilize the high-order information of K-space data, resulting in information loss. The present invention effectively utilizes all acquired K-space data through nonlinear mapping, improves information recovery, and thus enhances image quality.

[0067] The feasibility of the present invention has been proved through experiments, simulations and use. The present invention has carried out water film phantom and human brain tests on a self-developed zero liquid helium 0.23T normal conduction low-field magnetic resonance system. The scanning sequence uses an FSE sequence as the basis. The phase encoding position of the FSE sequence is adjusted according to the K-space acquisition template specified by the present invention, which proves the feasibility of the present invention. Figure 4The figure shows a comparison of the water film phantom imaging results of the GRAPPA technology and the present invention based on manifold nonlinear modeling, wherein a is a water film image reconstructed using the sum of squares method (SOS) based on full sampling data, which is used as a reference image; b and d are difference diagrams between an image reconstructed using the GRAPPA method based on full sampling of the central 48 rows and uniformly reducing the sampling of the rest of the data by one time and the reference image; c and e are difference diagrams between an image reconstructed using the manifold method based on full sampling of the central 48 rows and uniformly reducing the sampling of the rest of the data by one time and the reference image; by comparing the water film parallel imaging results of the two methods, it can be clearly seen that under the condition of increasing the same scanning speed, the performance of the manifold method is significantly better than that of the GRAPPA method, and the NMSE between its reconstructed image and the reference image is only 0.28, which is 32.7% lower than that of the GRAPPA method.

[0068] Figure 5 The figure shows a comparison of the human brain imaging results of the GRAPPA technology and the parallel imaging based on manifold nonlinear modeling of the present invention, wherein a is a human brain image reconstructed using the sum of squares method (SOS) based on full-sampled data, which is used as a reference image; b and c are difference diagrams between an image reconstructed using the GRAPPA method based on full sampling of the central 48 lines and uniformly downsampling the rest of the data and the reference image; d and e are difference diagrams between an image reconstructed using the manifold method based on full sampling of the central 48 lines and uniformly downsampling the rest of the data and the reference image; by comparing the human brain parallel imaging results of the two methods, it can be clearly seen that under the condition of improving the same scanning speed, the performance of the manifold method is significantly better than that of the GRAPPA method, and the NMSE between its reconstructed image and the reference image is only 0.43, which is 33% lower than that of the GRAPPA method.

[0069] Example 2 According to another embodiment of the present invention, a nonlinear mapping parallel imaging reconstruction device based on manifold is provided. Figure 6 ,include: An acquisition unit 201 is used for acquiring K-space data through multiple channels; A parameter optimization unit 202 sets the neighborhood size according to the actually acquired K-space data, and adaptively optimizes and adjusts the key parameters of the high-dimensional mapping; A mapping unit 203 uses K-space data (including nonlinearly related image information and noise components) as feature input, maps it to a higher-dimensional feature space through an epidemiological method, and establishes a topological relationship in the high-dimensional feature space; The fitting recovery unit 204 is used to smooth the differences between data points in the neighborhood and the differences between center points in different fields in the high-dimensional feature space through manifold regression, and finally establish a nonlinear model that minimizes noise features to fit the K-space part that complements the downsampled K-space, and restore it to the image space based on the reversible Fourier transform of the reconstructed K-space data.

[0070] The nonlinear mapping parallel imaging reconstruction method based on manifold in the embodiment of the present invention maps data into a high-dimensional space without explicitly calculating high-dimensional features. This method allows for a better separation of signal and noise, while being able to capture complex imaging features, thereby being able to more accurately complete the missing sampled K-space information and more accurately restore image details. By adding interaction terms or nonlinear features, the model can more carefully identify potential risk patterns, improve the accuracy and noise resistance of reconstructed images, expand the scope of application of the algorithm, solve the problem of low signal-to-noise ratio in low-field strength magnetic resonance parallel imaging, and significantly improve the quality and diagnostic value of reconstructed images.

[0071] Specifically, in order to solve the problem of slow magnetic resonance imaging, parallel imaging technology is a major breakthrough in magnetic resonance imaging technology. This method can accelerate the magnetic resonance imaging (MRI) scanning process by using multiple radio frequency coils to receive signals simultaneously. These coils cover different spatial regions and can capture more spatial information, thereby reducing the required k-space (frequency domain space) sampling points. This method can significantly reduce scanning time, improve the temporal resolution of images, and improve patient comfort. However, reducing sampling will increase the complexity of image reconstruction, such as obvious aliasing artifacts, so an effective reconstruction algorithm is required to restore high-quality images.

[0072] In low-field magnetic resonance imaging, the signal-to-noise ratio (SNR) is low, which further limits the scanning speed and image quality. In this case, the manifold method can perform projection operations in high-dimensional space to separate the signal and noise more closely, while capturing complex imaging features, thereby more accurately restoring the details of the image and more accurately completing the missing sampled K-space information, significantly improving the quality and diagnostic value of the reconstructed image.

[0073] In view of the above shortcomings, the present invention proposes a nonlinear mapping parallel imaging reconstruction method based on manifolds, which is suitable for the method of low-field magnetic resonance image reconstruction, and aims to solve the problems of low signal-to-noise ratio and slow imaging speed in low fields. The present invention adopts multi-channel acquisition and nonlinear modeling based on manifolds to improve the clarity and contrast of the image by more accurately completing the K-space data. Specifically, the present invention maps the data into a high-dimensional space through this method without explicitly calculating the high-dimensional features. This method allows the signal and noise to be more separated, and at the same time can capture complex imaging features, so that the K-space information of the missing samples can be more accurately completed, and the details of the image can be more accurately restored. By adding interaction terms or nonlinear features, the model can more carefully identify potential risk patterns, improve the accuracy and noise resistance of the reconstructed image, expand the scope of application of the algorithm, solve the problem of low signal-to-noise ratio in low-field strength magnetic resonance parallel imaging, and significantly improve the quality and diagnostic value of the reconstructed image.

[0074] The basic contents of the technical solution of this method are as follows: K-space data is collected in a low-field magnetic resonance system. Multiple receiving coils are used to collect K-space data in parallel. The data collected by each coil is undersampled (that is, there are data points that are not collected), but the K-space data in the center of each channel is guaranteed to be fully collected. These data are called autocalibration signals (ACS lines). These ACS lines provide complete K-space information for nonlinear modeling.

[0075] Adaptive parameter optimization. In this process, adaptive optimization parameters (such as kernel width and neighborhood size) can be adjusted according to actual data characteristics to adapt to different imaging conditions and noise levels, further improving the flexibility and applicability of the model.

[0076] Based on manifold nonlinear modeling. Since the collected K-space ACS data contains nonlinearly related image and noise features, it is initially used as a feature and mapped to a higher dimension through the manifold method. By adjusting the mapped high-dimensional feature space, the noise is separated more and the noise interference is minimized. Finally, the topological relationship of the high-dimensional feature space is established, which helps to extract the real structural features of the image more clearly.

[0077] Manifold regression and image reconstruction. In high-dimensional feature space, manifold regression smoothes the differences between data points in the neighborhood and the differences between center points in different areas, thereby establishing a nonlinear model that minimizes noise features to fit the K-space part that completes the downsampled part. Finally, based on the reversible Fourier transform (IFT) of the reconstructed K-space data, this information is restored to the image space to generate a higher quality image.

[0078] See also Figure 2, the technical solution of the present invention is described in detail as follows: 1. Acquisition of K-space data in low-field MRI systems The schematic diagram of K-space data acquisition method is as follows Figure 3 As shown in the figure, this sampling method strictly follows the Nyquist sampling theorem for full sampling in the ACS part, and reduces the sampling in the part outside the ACS, ensuring that the low-resolution complete K-space information is provided for nonlinear modeling while shortening the sampling time. The specific steps are as follows: Step 1: Establish a K-space model under low-field magnetic resonance conditions and determine the acquisition trajectory of the echo signal in the model.

[0079] Specifically, the acquisition trajectory of the echo signal is as follows: all echo signals are collected in parallel along one coordinate direction, wherein the central area of ​​the K space is fully sampled (ACS part), strictly following the Nyquist sampling theorem to ensure that complete high-resolution data is obtained for reconstruction coefficient estimation, while providing an accurate benchmark for subsequent image reconstruction; the edge area of ​​the K space is downsampled to reduce the amount of data collection and increase the acquisition speed.

[0080] Step 2: Determine the time sequence of the ultra-low field magnetic resonance scan according to the acquisition trajectory, and calculate the encoding gradient required to be applied by the magnetic resonance imaging system. The specific steps include: calculating the scanning time sequence parameters and encoding gradient of the full sampling area (ACS part) to ensure the integrity of the data acquisition in the central area; calculating the scanning time sequence parameters and encoding gradient of the down-sampling area to ensure the uniformity and efficiency of the data acquisition in the edge area. In a low-field environment, the signal-to-noise ratio is low, so it is necessary to specially optimize the time sequence and encoding gradient to maximize the signal strength and data acquisition efficiency, such as the FSE sequence with a fast acquisition speed adopted in the present invention.

[0081] Step 3: Collect K-space data for parallel imaging. Specifically, the three-dimensional multi-layer K-space data collected in this embodiment includes full sampling data of the central area of ​​each layer and down-sampled data of the edge area: Full sampling of the central area (ACS part): Full sampling of the central part of the K-space is performed according to the Nyquist sampling theorem to ensure the acquisition of complete high-resolution data. This is especially important for low-field environments. Downsampling of the edge area: Uniform downsampling is performed on the edge of the K-space to reduce data acquisition time. First, dense phase encoding acquisition is performed on the adaptive calibration scan (ACS) part to obtain complete central K-space data. Then, the phase encoding interval is gradually increased to make the phase encoding steps sparse and expand the range of phase encoding, thereby acquiring a uniformly downsampled K-space data.

[0082] 2. Adaptive parameter optimization In the kernel method of manifold, kernel width determines the projection effect of data blocks in high-dimensional feature space, which directly affects the accuracy of signal and noise separation. For low signal-to-noise ratio environments, increasing kernel width can reduce the impact of noise, make similar signals more concentrated, and enhance signal consistency; while in high signal-to-noise ratio environments, moderately reducing kernel width helps capture more subtle local features and details. The specific process is as follows: Step 1: Data analysis: First, analyze the K-space data to determine the "kernel" size, that is, the size of each neighborhood divided from the ACS partial data by high-dimensional mapping.

[0083] Step 2: Initial kernel width setting. Select the initial kernel width based on the data analysis results. , determines the initial dimension size of the high-dimensional mapping of the data.

[0084] Step 3: Adaptive iterative optimization. In each iteration, use the current estimate To calculate , and then calculate the approximate (i.e. according to the current The value is calculated by empirical formula 2 Then, using the new estimate To calculate the next step This step updates the kernel width by the least squares method. , the specific formula is as follows: [1] Where A represents the matrix composed of the non-undersampled position points of ACS, and the size of matrix A is m×k, where m is the total number of "kernels" divided in the ACS data, and k is the number of neighborhood points in each "kernel". Indicates that the kernel width is used The epidemiological method performs high-dimensional mapping on A. The combination of estimates of the linear representation of the high-dimensional mapping of each data point at the non-undersampled position of the ACS, with dimension ,in is the dimensionality of the new feature space, which is usually much higher than k.

[0085]

[0086] Formula 2 is the empirical formula for high-dimensional mapping of each sampling point in A, where Indicates that the target coil The K-space signal of the data at the missing position of the upper ACS part, represents the acquired undersampled signal, where is the coil index, t and h represent and The offset of the adjacent K-space data in the direction, R represents the undersampling rate (ORF, sampling rate factor), that is, the undersampling factor, and denote the coordinates along the frequency encoding and phase encoding directions, respectively. and is the basic sampling interval in K space, which is used to describe the distance between points in the frequency encoding and phase encoding directions respectively. r is the integer offset along the phase encoding direction, which determines the position of the target point. n represents the offset of n frequency encoding intervals from the adjacent K space data. The larger the n, the more nonlinear relationship between higher-dimensional K space data can be extracted. The conventional empirical formula 2 generally sets n to 3 to reduce the amount of calculation.

[0087] 3. Nonlinear modeling based on manifold All existing GRAPPA derivative algorithms are based on the linear model in formula [3]. Due to the low signal-to-noise ratio of low-field magnetic resonance, the nonlinear correlation between signal and noise in ACS data is difficult to separate, resulting in model deviation.

[0088]

[0089]

[0090] in Indicates that the target coil The K-space signal that is not acquired on represents the acquired undersampled signal, where is the coil index, t and h represent and The offset of the adjacent K-space data in the direction, represents the combination coefficients of the linear model, which are used to linearly combine the adjacent sampled signals into the unsampled signals on the target coil, L represents the total number of coils, R represents the undersampling rate (ORF, sampling rate factor), that is, the undersampling factor, l traverses the index of all coils, and They represent the coordinates along the frequency encoding and phase encoding directions respectively, and r is the integer offset along the phase encoding direction. Formula [3] only performs a linear combination in one dimension, which does not completely separate the image from the noise and results in a lot of image information being lost.

[0091] In order to solve the common nonlinear and noise-like error problems in the GRAPPA reconstruction process, this paper proposes an innovative method based on manifolds to improve the reconstruction accuracy by identifying and adjusting the nonlinear relationship between image features. This method uses the statistical characteristics of the data and the trend distribution in the sample space to effectively suppress noise and artifacts, thereby achieving more accurate signal reconstruction.

[0092] Step 1: Feature selection and nonlinear modeling. First, the nonlinear relationship features mapped to high-dimensional space are selected and modeled in manifold. For example, a method in manifold (kernel method) is used to perform high-dimensional mapping to separate the signal and noise of the nonlinear relationship in order to identify the error source in the reconstruction process. A typical kernel (inhomogeneous polynomial kernel) is as follows: [4] in, and is a scalar, is the order of the polynomial (and also the kernel width of the kernel function for non-homogeneous polynomial kernel functions), which helps to map nonlinear relationships in the data to a high-dimensional space so that they can be separated. Indicates the first extracted from the ACS partial data of all channels and a "kernel", .

[0093] The present invention uses nonlinear mapping Will Input Space The data in (ACS un-undersampled position data, i.e., initial features) are mapped to a high-dimensional feature space (High-dimensional features that can separate nonlinear signals and noise), use formula 4 to The linear combination after nonlinear mapping in high dimension is expressed as ,but: [5] Based on the least squares method of formula 5, the coefficients of each dimension between the high-dimensional mapping of the non-undersampled position data matrix A to the undersampled position data matrix B are calculated, and then the parameters used by the manifold model are obtained by combining formulas [2] and [4]. Here, it specifically refers to the kernel function and These two parameters.

[0094] Improper parameter settings may result in very high or even infinite dimensions of the mapping, so it is necessary to select high-dimensional features. In this example of the inhomogeneous polynomial kernel, it is found that retaining dimensional features above the quadratic term is difficult to further reduce the manifold model error, so it is discarded.

[0095] Step 2: Adjustment of feature space. Assume that a non-homogeneous polynomial kernel is used and =1, =1, =2, then the nonlinear mapping of manifold It can be specifically expressed as:

[0096] Using the nonlinear mapping of the above form , which helps to effectively process nonlinear inseparable data in high-dimensional space, where for Specifically, the k elements contained in "kernel" are: This mapping form contains first-order and second-order terms to preserve the basic information and nonlinear characteristics of the original data in high-dimensional space, which is particularly suitable for observing characteristic trends in manifold modeling. Preserve the underlying manifold trend of the data, the second-order terms The square sum interaction of each feature is captured, which enables the mapped feature space to reveal the nonlinear manifold trend. The higher the mapping order, the more complete the separation of nonlinear related information, but at the same time the calculation is more complicated. Adjusting the mapping to a suitable feature space is the key to this step. There are three quadratic terms, see formula [2]. To further reduce the complexity of the calculation, only the first two quadratic terms can be retained without affecting the new energy of the manifold model.

[0097] Step 3: Establish the topological relationship of the high-dimensional feature space. When constructing the topological relationship of the high-dimensional feature space, the adjusted high-dimensional data is used to deeply explore its complex associations in manifold. This process combines multi-level manifold feature analysis to reveal the potential patterns of the image and the local structure of the data. Modeling the topological relationship of the high-dimensional feature space helps capture the nonlinear relationship between signal and noise and between different tissues. As shown in formula [7], each neighborhood "kernel" divided in the ACS partial data of multiple channels is correlated with each other. The nth "kernel" is marked out in the ACS data of all channels. ACS data for all channels except Other "kernel" ( ) relative to The integer offset further affects , Formula 7 represents the mutual influence between all the “kernels” of all channels, where The remaining neighbors Composition matrix and coefficient matrix Finally, formulas [2] [4] [5] and [7] are combined to further integrate the relationship between different neighborhood "kernels" of the same channel, as well as the relationship between different neighborhood "kernel" data of different channels, to further improve the accuracy of the manifold model.

[0098]

[0099] 4. Manifold regression and signal reconstruction A manifold regression model is constructed to handle the nonlinear relationship between signal and noise in a high-dimensional feature space. The specific steps are as follows: Step 1: Smoothing: Manifold regression is used to smooth the data in the neighborhood to make the signal strength more consistent in the local area.

[0100] Step 2: Nonlinear enhancement of the center point: The center points of different “kernels” are nonlinearly enhanced through the regression model to make the signal strength more consistent in the global area.

[0101] Step 3: Interpolation and reconstruction. In the interpolation and reconstruction process, the unsampled points in the K space are interpolated using the manifold regression results obtained by adaptive iterative optimization of ACS data to complete the downsampled data; manifold regression fitting is performed in high-dimensional space to further reduce the interpolation error; then, the interpolated data is combined with the existing sampling points to generate complete K space data; finally, the reconstructed K space data is converted to image space through IFT.

[0102] The key points and intended protection points of the present invention are as follows: The core innovation of the present invention is to propose a nonlinear mapping parallel imaging reconstruction method based on manifolds, which provides an effective technical solution to improve image quality and accelerate the scanning process for the problems of low signal-to-noise ratio and slow imaging speed in low-field magnetic resonance imaging. The following are the key innovations of the present invention and the technical contents to be protected: Nonlinear mapping method based on manifolds. By adopting the nonlinear mapping technology in manifolds, K-space data is mapped to high-dimensional space, thereby achieving accurate modeling of complex data relationships in low-field magnetic resonance imaging. This method can capture nonlinear features in the data, overcome the shortcomings of traditional linear methods under low signal-to-noise ratios, and improve the accuracy of image reconstruction. The present invention takes the kernel method in manifolds as an example, and high-dimensional feature space mapping methods in manifolds such as polynomial feature expansion and high-dimensional representation in deep learning are also applicable to the present invention.

[0103] Nonlinear mapping method based on manifold. By adopting the nonlinear mapping technology in manifold, K-space data is mapped to high-dimensional space, so as to separate the nonlinearly related image and noise features in the original data acquired by low-field magnetic resonance. This method can overcome the shortcomings of traditional linear methods under low signal-to-noise ratio and improve the accuracy of image reconstruction. The present invention takes the kernel method in manifold as an example, and the high-dimensional feature space mapping methods of manifold such as polynomial feature expansion and high-dimensional representation in deep learning are also applicable to the present invention.

[0104] Combination of multi-channel acquisition and manifold modeling. The present invention combines multi-channel parallel acquisition with nonlinear modeling based on manifolds, receives signals simultaneously through multi-channel radio frequency coils, reduces K-space sampling points, and accurately completes undersampled data through nonlinear modeling. This technical solution significantly improves the image quality of low-field magnetic resonance imaging, especially in terms of signal-to-noise ratio and detail expression of reconstructed images.

[0105] Effective separation of signal and noise. The original data collected by low-field magnetic resonance imaging is mapped in high dimensions using manifold methods, effectively separating nonlinearly related image and noise features into a linearly combinable form. Modeling is performed by removing the "kernel" of the noise term to avoid the adverse effects of noise amplification in traditional methods. By separating the noise features, the present invention can more accurately restore image details and significantly improve image clarity and contrast.

[0106] Accurately complete undersampled K-space data. The present invention uses epidemiological regression technology to accurately calculate the coefficients of high-dimensional image features, and combines data smoothing in the neighborhood and nonlinear enhancement of center point data to more accurately restore undersampled K-space data in low-field magnetic resonance imaging, reducing image artifacts caused by data loss in traditional methods. This technical solution effectively improves the reconstruction accuracy of low-field magnetic resonance imaging.

[0107] Enhanced noise immunity and adaptability. The present invention introduces a manifold reconstruction method to complement the downsampled K-space data, so that the model can maintain strong noise immunity when dealing with noise in low-field magnetic resonance imaging. This innovation significantly broadens the scope of application of this method in various magnetic resonance imaging tasks, and is also applicable to complex high-resolution imaging tasks.

[0108] Kernel function selection and optimization iterative process. In the kernel method of the present invention, the selection of kernel function is crucial to the performance of high-dimensional mapping. In order to ensure the best imaging quality and reconstruction effect, the present invention allows the selection of multiple kernel functions, including non-homogeneous polynomial kernels, Gaussian kernels (RBF kernels), and hybrid kernel functions. By evaluating the performance of different kernel functions and combining the requirements of specific tasks, the best kernel function is selected iteratively automatically or through an optimization algorithm (such as alternating least squares method, gradient descent method, etc.), and its related parameters are tuned.

[0109] Sequence scanning model. The sequence used in the present invention has a fast scanning speed and can collect the central part full sampling and uniformly downsampled K-space data at the same time. The present invention is divided into two scans, one for scanning the central part data and one for scanning the downsampled data, and finally calibrated by the echo chain. Including but not limited to fast spin echo (FSE), gradient echo (GRE), balanced steady-state free precession (bSSFP), etc., such MRI sequences are applicable to the present invention.

[0110] Compared with the traditional GRAPPA method, the present invention significantly improves image quality, data completion capability and computational efficiency by introducing manifold nonlinear modeling methods and combining the application environment of low-field magnetic resonance, enhances adaptability to complex imaging environments, improves signal-to-noise ratio and image stability, and has important application value and promotion prospects. The details are as follows: (1) Stronger nonlinear modeling capabilities. GRAPPA relies on linear models and cannot effectively capture nonlinear relationships in data. In contrast, the present invention uses nonlinear manifold mapping to more accurately capture complex imaging features and improve image reconstruction accuracy.

[0111] (2) Better noise resistance. GRAPPA tends to amplify noise in low signal-to-noise ratio data, resulting in a decrease in image quality. Through the manifold method, the present invention can effectively separate signals and noise, improve noise resistance, thereby reducing artifacts and improving image clarity.

[0112] (3) More accurate K-space data completion. GRAPPA uses linear interpolation to fill in missing K-space data, which has limited accuracy. The present invention can more accurately fill in missing K-space data and retain more imaging details through high-dimensional nonlinear mapping.

[0113] (4) Higher adaptability. The applicability of GRAPPA is limited, especially when processing complex MR data. The present invention supports a variety of kernel functions (such as Gaussian kernel, non-homogeneous polynomial kernel, etc.), which can be flexibly adjusted according to data characteristics to adapt to different imaging tasks.

[0114] (5) Higher computational efficiency. Although the computational complexity of nonlinear methods is relatively high, the present invention can maintain high computational efficiency while maintaining image quality through manifold mapping and high-dimensional modeling, especially performing better on large data sets.

[0115] (6) Optimized data utilization and information recovery. GRAPPA does not fully utilize the high-order information of K-space data, resulting in information loss. The present invention effectively utilizes all acquired K-space data through nonlinear mapping, improves information recovery, and thus enhances image quality.

[0116] The feasibility of the present invention has been proved through experiments, simulations and use. The present invention has carried out water film phantom and human brain tests on a self-developed zero liquid helium 0.23T normal conduction low-field magnetic resonance system. The scanning sequence uses an FSE sequence as the basis. The phase encoding position of the FSE sequence is adjusted according to the K-space acquisition template specified by the present invention, which proves the feasibility of the present invention. Figure 4 The figure shows a comparison of the water film phantom imaging results of the GRAPPA technology and the present invention based on manifold nonlinear modeling, wherein a is a water film image reconstructed using the sum of squares method (SOS) based on full sampling data, which is used as a reference image; b and d are difference diagrams between an image reconstructed using the GRAPPA method based on full sampling of the central 48 rows and uniformly reducing the sampling of the rest of the data by one time and the reference image; c and e are difference diagrams between an image reconstructed using the manifold method based on full sampling of the central 48 rows and uniformly reducing the sampling of the rest of the data by one time and the reference image; by comparing the water film parallel imaging results of the two methods, it can be clearly seen that under the condition of increasing the same scanning speed, the performance of the manifold method is significantly better than that of the GRAPPA method, and the NMSE between its reconstructed image and the reference image is only 0.28, which is 32.7% lower than that of the GRAPPA method.

[0117] Figure 5 The figure shows a comparison of the human brain imaging results of the GRAPPA technology and the parallel imaging based on manifold nonlinear modeling of the present invention, wherein a is a human brain image reconstructed using the sum of squares method (SOS) based on full-sampled data, which is used as a reference image; b and c are difference diagrams between an image reconstructed using the GRAPPA method based on full sampling of the central 48 lines and uniformly downsampling the rest of the data and the reference image; d and e are difference diagrams between an image reconstructed using the manifold method based on full sampling of the central 48 lines and uniformly downsampling the rest of the data and the reference image; by comparing the human brain parallel imaging results of the two methods, it can be clearly seen that under the condition of improving the same scanning speed, the performance of the manifold method is significantly better than that of the GRAPPA method, and the NMSE between its reconstructed image and the reference image is only 0.43, which is 33% lower than that of the GRAPPA method.

[0118] Example 3 A storage medium stores a program file capable of implementing any one of the above-mentioned manifold-based nonlinear mapping parallel imaging reconstruction methods.

[0119] Example 4 A processor is used to run a program, wherein when the program is run, any one of the above-mentioned manifold-based nonlinear mapping parallel imaging reconstruction methods is executed.

[0120] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0121] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the system embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0123] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0124] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0125] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0126] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A nonlinear mapping parallel imaging reconstruction method based on manifold, characterized in that: The following steps are involved: S101: multi-channel acquisition of K-space data; S102: setting the neighborhood size according to the acquired K-space data, and adaptively optimizing and adjusting key parameters of high-dimensional mapping; S103: Using the nonlinearly related image information and noise components of the K-space data as feature input, mapping them to a higher-dimensional feature space through an epidemiological method, and establishing a topological relationship in the high-dimensional feature space; S104: In the high-dimensional feature space, the differences between data points in the neighborhood and the differences between center points in different fields are smoothed by manifold regression, and finally a nonlinear model that minimizes noise features is established to fit the K-space part that is complemented and downsampled, and the reversible Fourier transform based on the complemented K-space data is restored to the image space.

2. The nonlinear mapping parallel imaging reconstruction method based on manifold according to claim 1, characterized in that: In step S101 , multiple receiving coils are used to collect K-space data in parallel in a low-field magnetic resonance system.

3. The nonlinear mapping parallel imaging reconstruction method based on manifold according to claim 1, characterized in that: Step S101 specifically includes: Establish a K-space model under low-field magnetic resonance conditions and determine the acquisition trajectory of echo signals in the K-space model; Determine the time series parameters of the low-field magnetic resonance scan according to the acquisition trajectory, and calculate the encoding gradient required to be applied by the magnetic resonance imaging system; Acquire K-space data for parallel imaging.

4. The nonlinear mapping parallel imaging reconstruction method based on manifold according to claim 1, characterized in that: In step S102, the neighborhood size is determined according to actual data characteristics, and the key parameters of the high-dimensional mapping are adaptively optimized and adjusted, and the key parameter is the kernel width.

5. The nonlinear mapping parallel imaging reconstruction method based on manifold according to claim 1, characterized in that: Step S102 specifically includes: Analyze the K-space data, determine the size of each neighborhood divided by the high-dimensional mapping from the ACS partial data, and generate data analysis results; Select the initial kernel width based on the data analysis results; During the reconstruction process, the kernel width is dynamically adjusted through the error feedback mechanism for adaptive iterative optimization.

6. The manifold-based nonlinear mapping parallel imaging reconstruction method according to claim 1, characterized in that: Step S103 specifically includes: The nonlinearly related image information and noise components of K-space data are used as feature inputs, mapped to a higher-dimensional feature space through epidemiological methods, and the nonlinear relationship features of the high-dimensional space are obtained; the difference between the actual sampling value of the missing position of the ACS and the high-dimensional mapped feature value of the linear combination is minimized to select the nonlinear relationship features and conduct manifold modeling; Adjust the high-dimensional feature space to balance computational complexity and manifold model performance; Construct the topological relationship of high-dimensional feature space, and use the adjusted high-dimensional data to deeply explore the complex associations between each neighborhood kernel in manifold, and combine multi-level manifold feature analysis to reveal the nonlinear relationship between signals and noise, and between different tissues.

7. The manifold-based nonlinear mapping parallel imaging reconstruction method according to claim 1, characterized in that: In step S104, smoothing the differences between data points in the neighborhood by manifold regression includes: smoothing the data in the neighborhood by manifold regression to make the signal strength more consistent in the local area.

8. The manifold-based nonlinear mapping parallel imaging reconstruction method according to claim 1, characterized in that: In step S104, after smoothing the differences between the data in the neighborhood by manifold regression, the method further includes: performing nonlinear enhancement on kernel center points of different neighborhoods by using a regression model to make the signal strength more consistent in the global region.

9. The nonlinear mapping parallel imaging reconstruction method based on manifold according to claim 1, characterized in that: In step S104, the nonlinear relationship between the signal and the noise is processed in the high-dimensional feature space by manifold regression, and finally a nonlinear model with minimized noise features is established to fit the K-space part of the complement downsampled K-space data, and the reversible Fourier transform based on the complemented K-space data is restored to the image space, including: In the interpolation and reconstruction process, the unsampled points in the K space are first interpolated using the manifold regression results obtained by adaptive iterative optimization of ACS data to complete the downsampled data; manifold regression fitting is performed in high-dimensional space to further reduce the interpolation error; then, the interpolated data is combined with the existing sampling points to generate complete K space data; finally, the reconstructed K space data is converted to image space through IFT.

10. A nonlinear mapping parallel imaging reconstruction device based on manifold, characterized in that: include: An acquisition unit, used for acquiring K-space data through multiple channels; A parameter optimization unit sets the neighborhood size according to the acquired K-space data and adaptively optimizes and adjusts the key parameters of the high-dimensional mapping; The mapping unit takes the nonlinearly related image information and noise components of the K-space data as feature inputs, maps them to a higher-dimensional feature space through an epidemiological method, and establishes a topological relationship in the high-dimensional feature space; The fitting recovery unit is used to smooth the differences between data points in the neighborhood and the differences between center points in different fields in the high-dimensional feature space through manifold regression, and finally establish a nonlinear model that minimizes noise features to fit the completed downsampled K-space part, and restore it to the image space based on the reversible Fourier transform of the completed K-space data.

Images