阅读说明：本技术 基于共轭和层间信息的并行磁共振成像算法 (Parallel magnetic resonance imaging algorithm based on conjugation and interlayer information ) 是由 何汶静 杨汉丰 祝元仲 于 2020-12-10 设计创作，主要内容包括：本发明公开了一种基于共轭和层间信息的并行磁共振成像算法,包括:对成像物体的梯度正向高频部分和梯度负向高频部分进行R倍欠采样,并将中间低频部分作为校准区进行全采样,得到矩阵F；构建欠采样数据的预测矩阵A；利用本层及相邻层矩阵F的校准区构建权重计算矩阵B；构建权重计算矩阵B的列向量矩阵C；根据权重计算矩阵B与权重矩阵W的乘积得到列向量矩阵C,采用最小二乘法可拟合得到插值窗内权重矩阵W；构建欠采样行数据的预测矩阵D；根据预测矩阵D与插值窗内权重矩阵W的乘积,得到列向量E；填充获得重建矩阵H,经傅里叶变换后得到并行磁共振成像的图像。本发明具有成像速度快、预测源矩阵信息量丰富、信噪比较高等优点。(The invention discloses a conjugate and interlayer information-based parallel magnetic resonance imaging algorithm, which comprises the steps of carrying out R-time undersampling on a gradient positive high-frequency part and a gradient negative high-frequency part of an imaging object, and carrying out full sampling by taking a middle low-frequency part as a calibration area to obtain a matrix F; constructing a prediction matrix A of undersampled data; constructing a weight calculation matrix B by utilizing the calibration areas of the matrixes F in the current layer and the adjacent layer; constructing a column vector matrix C of the weight calculation matrix B; obtaining a column vector matrix C according to the product of the weight calculation matrix B and the weight matrix W, and obtaining the weight matrix W in the interpolation window by fitting by adopting a least square method; constructing a prediction matrix D of undersampled row data; obtaining a column vector E according to the product of the prediction matrix D and the weight matrix W in the interpolation window; and filling to obtain a reconstruction matrix H, and performing Fourier transform to obtain an image of parallel magnetic resonance imaging. The invention has the advantages of high imaging speed, rich information quantity of the predicted source matrix, high signal-to-noise ratio and the like.)

1. The parallel magnetic resonance imaging algorithm based on conjugation and interlayer information utilizes K space data received by a phased array coil, wherein the phased array coil is provided with L, and L is a positive integer greater than 1, and the parallel magnetic resonance imaging algorithm is characterized by comprising the following steps:

performing R-time undersampling on a gradient positive high-frequency part and a gradient negative high-frequency part of an imaging object, and performing full sampling by taking a middle low-frequency part as a calibration area to obtain a matrix F; the sampling scanning totals Z layers, and Z is a positive integer greater than 1;

constructing a prediction matrix A of the matrix F: filling ith row sampling data of a z +1 th layer adjacent to a high-frequency undersampled part in the matrix F into the ith row of the z-th layer, and filling conjugate symmetric data in the z-th layer with each other;

and (3) constructing a weight calculation matrix B by utilizing the calibration areas of the matrixes F in the layer and the adjacent layer: the weight calculation matrix B is a transformation combination of K space data of L phased array coils of the z-th layer at a position x (i, j, L) and interpolation window data of a position x (-i, j, L), and a transformation combination of K space data of L phased array coils of the z + 1-th layer at a position x (i, j, L) and interpolation window data of a position x (-i, j, L);

moving the position of x (i, j,1) in the calibration area to obtain any row of data of the weight calculation matrix B, and recording the data of the position x (i, j, l), wherein any row of data is one row to obtain a column vector matrix C;

obtaining a column vector matrix C according to the product of the weight calculation matrix B and the weight matrix W, and obtaining the weight matrix W in the interpolation window by fitting by adopting a least square method;

the non-sampled data in the matrix F is inquired and the position is recorded asPosition in the prediction matrix ASetting an interpolation window and forming a prediction matrix D of undersampled row data;

obtaining a column vector E according to the product of the prediction matrix D and the weight matrix W in the interpolation window;

will positionFilling the data to the corresponding position in the matrix F, removing the originally filled 0 to obtain a reconstructed matrix H, and obtaining the parallel magnetic resonance imaging image after Fourier transformation.

2. The conjugate and interlayer information based parallel magnetic resonance imaging algorithm of claim 1, wherein the matrix F is constructed as follows:

dividing K space data received by any phased array coil into two symmetrical parts of positive gradient and negative gradient;

sampling is started from a low-frequency part in the positive direction of the gradient, and in the z-th layer, if the gradient is in the positive direction, sampling is carried out on the q + z% R row; q satisfies q% R ═ 1;

if the gradient is negative, sampling the (q + z% R +1) line;

and filling the non-sampling rows with 0, and fully sampling the middle low-frequency part to obtain a matrix F.

3. The conjugate and interlayer information based parallel magnetic resonance imaging algorithm of claim 1, wherein R is 4 or 5 or 6.

Technical Field

The invention relates to the technical field of magnetic resonance imaging, in particular to a parallel magnetic resonance imaging algorithm based on conjugation and interlayer information.

Background

Parallel magnetic resonance imaging (parallel MRI, p-MRI) is a major technological breakthrough in the field of high-speed imaging and image reconstruction of medical images, and has been applied to various large magnetic resonance devices. In the parallel magnetic resonance imaging process, a multi-channel phased array coil acquires magnetic resonance K space signals, and each channel of the phased array coil contains information of adjacent channels, so that partial down-sampling information can be estimated through the information of the adjacent coils, and the magnetic resonance scanning speed is improved by reducing the phase encoding times.

At present, image reconstruction algorithms in the prior art are mainly classified into two types:

firstly, based on reconstruction of an image domain, the algorithm mainly utilizes coil sensitivity to carry out image reconstruction to obtain an artifact-free image, the representative algorithm is SENSE (sensitivity encoding) and a subsequently developed expansion method SC-SENSE (Self-calibration), PILS (local sensing) and the like, the algorithm requires a channel coil to have more accurate sensitivity, coil sensitivity information is calculated by collecting low-frequency signals, when the sensitivity is known, Chaari and the like combine compressed sensing with SENSE, regular sensing is combined, items are added, and a better reconstruction effect is obtained, but accurate estimation is often very difficult. For example, the invention has the patent application number of 201711246873.0, and is named as a parallel magnetic resonance imaging method and device based on adaptive joint sparse coding, and a computer readable medium.

Secondly, a coil-by-coil reconstruction technology based on K space acquires intermediate line data as a self-calibration signal (ACS), a final image is directly reconstructed in multiple channels, the coil sensitivity does not need to be estimated, weight coefficients of a multi-channel coil are calculated by undersampled K space data, the weight coefficients are used for fitting undersampled missing data, and the missing data are reconstructed into a diagnostic image, and a representative algorithm is AUTO-SMASH (simulated Acquisition of Spatial harmonics), GRAPPA (Generalized AUTO-calibrating Parallel Acquisition), VD-AUTO-SMASH algorithm and the like. For example, the invention patent in china with patent application number "201510216413.8" and the name "a parallel magnetic resonance imaging phase processing method" is obtained by: performing Fourier inverse transformation on K space data acquired by a multi-channel coil in parallel magnetic resonance imaging to obtain the amplitude and the phase of each coil image; constructing a reference coil image, and estimating the spatial sensitivity distribution of each coil of the multiple channels; performing two-dimensional Fourier transform on the spatial sensitivity of the reference coil image, and intercepting a middle matrix as a convolution kernel; constructing a K space data convolution model, and solving the joint weight W of the coil; obtaining a K space value of the virtual coil, and obtaining a virtual coil image through Fourier inverse transformation; phase unwrapping, and removing the phase of the virtual coil image background; the mask image is used to extract the phase of the region of interest.

At present, in the process of calculating the reconstruction coefficient matrix, the GRAPPA algorithm in the prior art increases the condition number of the equation coefficient matrix along with the increase of the acceleration multiple, reduces the available information amount, and increases the equation ill-condition degree. So that the noise in the process of inverting the matrix is amplified.

In addition, the GRAPPA algorithm in the prior art can well avoid the wrap-around artifact for 2-time acceleration, but the reconstruction effect is poor for the acceleration exceeding 2 times, the noise is obvious, and the wrap-around artifact still remains. The method is an urgent problem to be solved for application scenes sensitive to scanning speed, such as magnetic resonance three-dimensional acquisition, dynamic imaging, real-time imaging, planar echo imaging and the like.

In the chinese invention patent with patent application number "2020111409917.9" and named "an improved algorithm for parallel magnetic resonance imaging", although the parallel magnetic resonance imaging reconstruction quality is higher by 3 times or less, under the acceleration condition of 3 times or more, the blank information is too much, at this time, if only the intra-layer conjugate information is still used to be added into the interpolation window, although the imaging effect can be significantly improved, the problem of too little information still exists, the signal-to-noise ratio is insufficient, and the ideal imaging quality cannot be achieved.

Therefore, there is an urgent need to provide a parallel magnetic resonance imaging algorithm based on conjugate and interlayer information, which utilizes conjugate corresponding row data and adjacent layers to increase the data amount in the interpolation window, increase the information amount of the prediction source matrix, and improve the signal-to-noise ratio while accelerating.

Disclosure of Invention

In view of the above problems, the present invention aims to provide a parallel magnetic resonance imaging algorithm based on conjugate and interlayer information, and the technical solution adopted by the present invention is as follows:

the parallel magnetic resonance imaging algorithm based on conjugation and interlayer information utilizes K space data received by a phased array coil, wherein the number of the phased array coil is L, and L is a positive integer greater than 1, and the parallel magnetic resonance imaging algorithm comprises the following steps:

performing R-time undersampling on a gradient positive high-frequency part and a gradient negative high-frequency part of an imaging object, and performing full sampling by taking a middle low-frequency part as a calibration area to obtain a matrix F; the sampling scanning totals Z layers, and Z is a positive integer greater than 1;

constructing a prediction matrix A of the matrix F: filling ith row sampling data of the adjacent z +1 th layer in the matrix F into the z-th layer, and filling conjugate symmetric data in the z-th layer;

and (3) constructing a weight calculation matrix B by utilizing the calibration areas of the matrixes F of the layer and the adjacent layers: the weight calculation matrix B is a transformation combination of K space data of L phased array coils of the z-th layer at a position x (i, j, L) and interpolation window data of a position x (-i, j, L), and a transformation combination of K space data of L phased array coils of the z + 1-th layer at a position x (i, j, L) and interpolation window data of a position x (-i, j, L);

moving the position of x (i, j,1) in the calibration area to obtain any row of data of the weight calculation matrix B, and recording the data of the position x (i, j, l), wherein any row of data is one row to obtain a column vector matrix C;

obtaining a column vector matrix C according to the product of the weight calculation matrix B and the weight matrix W, and obtaining the weight matrix W in the interpolation window by fitting by adopting a least square method;

the non-sampled data in the matrix F is inquired and the position is recorded asPosition in the prediction matrix ASetting an interpolation window and forming a prediction matrix D of undersampled row data;

obtaining a column vector E according to the product of the prediction matrix D and the weight matrix W in the interpolation window;

will positionFilling the corresponding position in the matrix F in the data, removing the originally filled 0 to obtain a reconstructed matrix H, and performing Fourier transform to obtain a parallel magnetic resonance imaging image.

Further, the construction process of the matrix F is as follows:

dividing K space data received by any phased array coil into two symmetrical parts of positive gradient and negative gradient;

sampling is started from a low-frequency part in the positive direction of the gradient, and in the z-th layer, if the gradient is in the positive direction, sampling is carried out on the q + z% R row; q satisfies q% R ═ 1;

if the gradient is negative, sampling the (q + z% R +1) line;

and filling the non-sampling rows with 0, and fully sampling the middle low-frequency part to obtain a matrix F.

Preferably, said R is 4 or 5 or 6.

Compared with the prior art, the invention has the following beneficial effects:

(1) the invention can increase the data in the interpolation window, not only utilizes the conjugate information in the layer, but also adds the information of the adjacent layer for fitting, thereby increasing the information quantity, increasing the signal-to-noise ratio and reducing the convolution artifact; for example, if the information of the adjacent layer 1 is added, 3 rows of fitting data can be added, and if the data of the upper and lower adjacent layers 2 is used, 5 rows can be added;

(2) the invention adopts undersampling to obtain high-frequency sampling data of positive direction and negative direction of the gradient, and adopts full sampling to obtain the middle low-frequency part, thereby reducing the sampling data, improving the imaging speed and simultaneously ensuring the sufficient information content;

(3) on the basis of conjugate symmetry, the method can obtain more useful information under the condition of down-sampling, and effectively improves the signal-to-noise ratio;

(4) the invention utilizes the calibration areas of the matrix F of the layer and the adjacent layer to obtain the weight matrix in the interpolation window, applies the weight matrix in the interpolation window to the under-sampling area to obtain the data around the non-sampled signal, and reconstructs the under-sampling point by using the prediction source matrix with larger information quantity;

in conclusion, the method has the advantages of rich information quantity of the prediction source matrix, high signal-to-noise ratio and the like, and has high practical value and popularization value in the technical field of magnetic resonance imaging.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention, and therefore should not be considered as limiting the scope of protection, and it is obvious for those skilled in the art that other related drawings can be obtained according to these drawings without inventive efforts.

Fig. 1 is a schematic structural diagram of a matrix F according to the present invention.

Fig. 2 is a schematic structural diagram of a prediction matrix a according to the present invention.

Detailed Description

To further clarify the objects, technical solutions and advantages of the present application, the present invention will be further described with reference to the accompanying drawings and examples, and embodiments of the present invention include, but are not limited to, the following examples. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Examples

As shown in fig. 1 to 2, the present embodiment provides a parallel magnetic resonance imaging algorithm based on conjugate and interlayer information, which utilizes K-space data received by a phased array coil, where the number of the phased array coil is L, and the parallel magnetic resonance imaging algorithm of the present embodiment includes the following steps:

firstly, performing R (4, 5, 6) times undersampling on a gradient positive high-frequency part and a gradient negative high-frequency part of an imaging object by using a magnetic resonance equipment imaging system, and performing full sampling by taking a middle low-frequency part (16 rows) as a calibration area to obtain a matrix F;

in the embodiment, the K-space data received by any phased array coil is divided into two symmetrical parts of positive gradient and negative gradient; sampling is carried out from a low-frequency part in the positive direction of the gradient, and in the cross section of the z-th layer, if the gradient is in the positive direction, sampling is carried out on the q + z% R row; q satisfies q% R ═ 1; if the gradient is negative, sampling- (q + z% R + 1); and filling the non-sampling rows with 0, and fully sampling the middle low-frequency part to obtain a matrix F.

Secondly, constructing a prediction matrix A of the matrix F: in the undersampled area, filling ith row of sampling data of the adjacent z +1 th layer in the matrix F into the z-th layer, and filling conjugate symmetric data in the z-th layer with each other; since it is ensured in the first step that when the mth row is the sampled data, the-mth row is necessarily the undersampled data, the intralayer conjugate symmetric row data are interleaved, and the existing data are not covered. At this time, the data in the interpolation window is changed from 2 lines of the original algorithm to 5 lines.

Thirdly, constructing a weight calculation matrix B in the calibration area of the matrix F of the layer and the adjacent layer: the weight calculation matrix B is a transformation combination of K space data of L phased array coils of the z-th layer at a position x (i, j, L) (namely the K space position is an ith row, a jth column and an ith coil, wherein i, j and L are positive integers larger than 1) and interpolation window data of a position x (-i, j, L), and a transformation combination of K space data of L phased array coils of the z + 1-th layer at a position x (i, j, L) and interpolation window data of a position x (-i, j, L); the positions shifted by x (i, j,1) in the calibration area, and a row of data is obtained at each position to form a matrix B.

Fourthly, moving the position x (i, j,1) in the calibration area to obtain any row of data of the weight calculation matrix B, and recording the data of the position x (i, j, l), wherein any row of data is one row to obtain a column vector matrix C;

fifthly, obtaining a column vector matrix C according to the product of the weight calculation matrix B and the weight matrix W, and obtaining the weight matrix W in the interpolation window by fitting by adopting a least square method;

sixthly, inquiring the data which are not sampled in the matrix F and recording the position of the data asPosition in the prediction matrix ASetting an interpolation window and forming a prediction matrix D of undersampled row data;

seventhly, obtaining a column vector E according to the product of the prediction matrix D and the weight matrix W in the interpolation window;

the eighth step of comparing the positionFilling the corresponding position in the matrix F in the data, removing the originally filled 0 to obtain a reconstructed matrix H, and performing Fourier transform to obtain a parallel magnetic resonance imaging image.

The above-mentioned embodiments are only preferred embodiments of the present invention, and do not limit the scope of the present invention, but all the modifications made by the principles of the present invention and the non-inventive efforts based on the above-mentioned embodiments shall fall within the scope of the present invention.