CN108523890B - Local Field Estimation Method Based on Spatial Distribution of Magnetic Resonance Dipole Field - Google Patents

Abstract

A local field estimation method based on magnetic resonance dipole field space distribution relates to nuclear magnetic resonance quantitative susceptibility imaging. Number of windings based on phase map; based on the standard deviation σ (r) of the gaussian convolution kernel modulated by the amplitude, gradient and number of wraps of the phase map; dividing the region of interest into a high susceptibility change region and a uniform susceptibility region based on the spatial information; the method can effectively remove the background fields of the strong magnetic field in different imaging directions and accurately estimate the local field, simultaneously reserves the integrity of brain tissues, and has an effect obviously superior to that of the existing method.

Description

Local Field Estimation Method Based on Spatial Distribution of Magnetic Resonance Dipole Field

technical field

The invention relates to nuclear magnetic resonance quantitative magnetic susceptibility imaging, in particular to a local field estimation method based on the spatial distribution of magnetic resonance dipole fields.

Background technique

Magnetic susceptibility is an inherent property of matter, which can reflect the degree of magnetization of biological tissues under the action of an external magnetic field. It is one of the main signal sources of Magnetic Resonance Imaging (MRI). Since the magnetic susceptibility distribution of biological tissues is affected by factors such as iron content, calcification, and blood oxygen saturation, the magnetic susceptibility varies between different tissues of the human body, and between normal tissues and diseased tissues. There is iron deposition in the gray matter of the human brain, which is paramagnetic, and the magnetic susceptibility value is positive; the oxygenated hemoglobin in the blood vessels is diamagnetic, and the deoxyhemoglobin is paramagnetic; the main component of the white matter is myelin, which is diamagnetic , the magnetic susceptibility value is negative.

Susceptibility Weighted Imaging (SWI) and R2* images are commonly used in MRI to detect the magnetic susceptibility of tissues, but these methods have low sensitivity and specificity, and SWI only qualitatively measures the change of magnetic susceptibility, while R2* images Susceptible to intra-voxel spin phase dispersion at strong gradient edges. Quantitative Susceptibility Mapping (QSM) utilizes phase information that is often neglected in the past, through phase unwinding and preprocessing to remove the background field, to obtain the changing characteristics of the local magnetic field, and obtain tissue through complex field-to-source inversion calculations. The quantitative magnetic susceptibility map of ^[1,2] . QSM technology can effectively overcome the above-mentioned defects of SWI and R2*, and has been widely used in the research of cerebral hemorrhage, multiple sclerosis and various degenerative neurological diseases, and is of great significance for the clinical diagnosis of these diseases ^[3-6] , Complementary with traditional MRI imaging technology, showing high clinical value and application prospects.

QSM technology is a research hotspot in the field of medical imaging. One of the main problems it faces is how to remove the influence of the background field outside the volume of interest (VOI) on the internal field of the VOI, so as to obtain accurate local field estimation . The accuracy of the local field estimation directly affects the accuracy of the QSM. According to the physical principle of the background field, there are mainly two important local field estimation models: one is the Projection onto DipoleField (PDF) ^[7] , which assumes that the local field and the background field in the VOI The interior is mutually orthogonal, the inner product of the two is zero, and the local field can be obtained by the optimal solution of the equation; the second is the Sophisticated Harmonic Artifact Reduction for Phase data (SHARP) method. ^[8] , that is, assuming that the background field satisfies the harmonic function property in the VOI, the local field can be obtained by solving the normalized spherical mean convolution kernel. Due to the spherical mean convolution kernel, the voxel in the VOI edge region of SHARP method is convolved with invalid signals outside the VOI, resulting in blurring of the edge region equal to the diameter of the convolution kernel, so these blurs are removed in the QSM reconstruction. Because SHARP has the defect of corroding the edge of brain tissue, researchers have proposed corresponding improvement measures such as the LBV method ^[9] to solve the boundary problem of the Laplace equation, and the V-SHARP method ^[10] with variable convolution kernel radius. a certain effect.

Because the strong magnetic susceptibility value is very high at the junction of the brain tissue and the sinus or skull, and the spatial distribution of the dipole field is related to the geometric structure of the tissue and the imaging direction, with the change of the imaging angle of the human brain, the strong magnetic susceptibility generated in this area is very high. The dipole field produces strong interference to the surrounding brain tissue in different directions, which has a serious impact on the correct estimation of the local field ^[11] . If the above methods are used, serious phase residues will be left on the local field maps of these regions, and serious magnetic susceptibility artifacts will be left in the subsequent QSM reconstruction, covering up the details of normal tissue structures. In order to eliminate such effects, the above-mentioned methods all need to use a mask to remove the brain tissue in this area to reduce the phase residual, thereby reducing the QSM artifact. However, the use of different orientation masks will lead to serious brain tissue loss, which will prevent us from obtaining complete brain diagnostic information, which is not conducive to the clinical application of QSM technology. Therefore, when removing the background field of the strong magnetic susceptibility region under multi-directional imaging and accurately estimating the local field, the above methods all have shortcomings, or the brain tissue is seriously missing, or the phase residual is serious, all of which need to be further improved.

references:

[1] Shmueli, K., de Zwart, J.A., van Gelderen, P., Li, T.Q., Dodd, S.J., Duyn, J.H., 2009. Magnetic susceptibility mapping of brain tissue in vivo using MRIphase data.Magn.Reson.Med .62, 1510-1522.

[2] Haacke, E.M., Liu, S.F., Buch, S., Zheng, W.L., Wu, D.M., Ye, Y.Q., 2015. Quantitative susceptibility mapping: current status and future directions. Magn.Reson.Imag.33,1-25 .

[3] He, N. Ling, H. Ding, B. Huang, J. Zhang, Y. Zhang, Z. Liu, C. Chen, K. Yan, F, 2015. Region-Specific Disturbed Iron Distribution in Early Idiopathic Parkinson's Disease Measured by Quantitative Susceptibility Mapping,HumanBrain Mapping,36,4407-4420.

[4] Wang, Y., Liu, T., 2015. Quantitative susceptibility mapping (QSM): decoding MRI data for a tissue magnetic biomarker. Magn.Reson.Med.73,82-101.

[5] Bergen, J., Hua, J., Lim, I.A.L, Jones, C.K., Margolis, R.L., Ross, C.A., P.C.M. van Zijl, Li, X, 2016. Quantitative susceptibility mapping suggests alteredbrain iron in premanifest Huntington disease. Am. J. Neuroradiol. 37, 789-796.

[6] Lotfipour A K, Wharton S, Schwarz S T, et al, 2012. High resolution magnetic susceptibility mapping of the substantia nigra in Parkinson'sdisease. [J]. Journal of Magnetic Resonance Imaging Jmri, 35, 48–55.

[7] Liu, T., Khalidov, I., de Rochefort, L., Spincemaille, P., Liu, J., Tsiouris, A.J., 2011. A novel background field removal method for MRI using projection onto diplole fields (PDF) .NMR Biomed. 24, 1129-1136.

[8] Schweser, F., Deistung, A., Lehr, B.W., Reichenbach, J.R., 2011. Quantitative imaging of intrinsic magnetic tissue properties using MRIsignal phase: an approach to in vivo brain iron metabolism? NeuroImage. 54, 2789-2807.

[9] Zhou, D., Liu, T., Spincemaille, P., Wang, Y., 2014. Background fieldremoval by solving the Laplacian boundary value problem. NMR Biomed. 27, 312-319.

[10] Wu, B., Li, W., Guidon, A., Liu, C.L., 2012. Whole brain susceptibilitymapping using compressed sensing. Magn.Reson.Med.67,137-147.

[11] Bao, L. Li, X. Cai, C. Chen, Z., P. C. M. V. Zijl, 2016. Quantitative Susceptibility Mapping Using Structural Feature Based Collaborative Reconstruction (SFCR) in the Human Brain, IEEE Transactions on Medical Imaging. 35, 2040-2050 .

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a local field estimation method based on the spatial distribution of the magnetic resonance dipole field.

The present invention comprises the following steps:

1) Based on the winding number of the phase diagram, the model is:

为解缠绕后的相位图，

为缠绕相位图，W[·]为相位缠绕算子；In the model, K(r) is the winding number of the phase diagram, γ is the gyromagnetic ratio, TE is the echo time, B ^Δ is the total field value of the dipole field,

For the unwrapped phase diagram,

is the winding phase diagram, W[ ] is the phase winding operator;

2) Based on the standard deviation σ(r) of the Gaussian convolution kernel modulated by the amplitude, gradient and winding number of the phase map, the model is:

为总场图的梯度，ε为大于零的实数，K(r)∈[0,1]为归一化的相位缠绕数，由此建立的与逐个体素空间信息相关的卷积核表示为：In the model, C is the positive adjustment constant, B ^Δ is the total field map,

is the gradient of the total field map, ε is a real number greater than zero, K(r)∈[0,1] is the normalized phase wrapping number, and the convolution kernel related to the voxel-by-voxel spatial information thus established is expressed as :

where (x, y, z) are the spatial coordinates of the points in the Gaussian kernel;

3) The region of interest is divided into regions with high magnetic susceptibility variation and uniform magnetic susceptibility regions based on spatial information. The model is:

为σ(r)倒数，记为

为区域分割阈值，定义为

其中，

与σ_max分别表示

的均值和最大值；δ为单位冲击函数，B_int为局部场，

为临时变量，

表示卷积运算；通过

将感兴趣区域分为高磁化率变化区域D₁和均匀磁化率区域D₂，且分别与高斯卷积核和球均值卷积作卷积运算，将背景场去除，得到局部场。in the model,

is the reciprocal of σ(r), denoted as

is the region segmentation threshold, defined as

in,

and σ _max , respectively

The mean and maximum value of ; δ is the unit shock function, B _int is the local field,

is a temporary variable,

represents a convolution operation; by

The region of interest is divided into high magnetic susceptibility variation region D ₁ and uniform magnetic susceptibility region D ₂ , and convolution operations are performed with Gaussian convolution kernel and spherical mean convolution respectively, and the background field is removed to obtain the local field.

In step 1), the winding number of the phase map can include the evolution results of different field map changes at different times, which can be used for local field estimation at any echo time.

In step 2), the spatial information composed of the amplitude, gradient and winding number of the phase map can modulate the standard deviation of the Gaussian convolution kernel by voxel, and when the value of the spatial information is larger, it indicates that there is a strong background field in the area, Then a smaller Gaussian standard deviation is obtained, that is, the weight of the corresponding Gaussian core sphere is high and the surrounding point values decay rapidly, which can better suppress the background field and obtain an accurate local field estimate.

In step 3), the region of interest is divided into a high magnetic susceptibility change region and a uniform magnetic susceptibility change region based on the spatial information, which can speed up the operation speed of the algorithm. The spatial information considers the field image components in different imaging directions, and can be used Local field estimation in multiple directions.

The present invention utilizes the spatial information of the phase map, that is, the amplitude, gradient and normalized winding number of the phase map are introduced to modulate the standard deviation of the Gaussian convolution kernel by voxel, and then the weights of the sphere center point and other points. This method can effectively The background fields of regions with strong magnetic susceptibility under different imaging directions are removed and the local fields are accurately estimated while preserving the integrity of brain tissue, which is significantly better than existing methods.

Description of drawings

Figure 1 is a graph of human brain data in different imaging directions. (a) Unwrapped phase map, (b) spatial information value map, (c) local field map, (d) magnetic susceptibility map.

Figure 2 is the experimental graph of the simulation data. (a) True magnetic susceptibility map, full field map and local field map in direction 1, (b) true magnetic susceptibility map, full field map and local field map in direction 2, (c) True magnetic susceptibility map, full field map and local field map in direction 3 Field map and local field map, (d) True susceptibility map for direction 4, full field map and local field map.

Figure 3 shows the local field estimation results from the simulation data. (a) Local field map calculated by three methods of PDF, R-SHARP and iRSHARP in direction 1, (b) local field map calculated by three methods of PDF, R-SHARP and iRSHARP in direction 2, (c) direction 3 Local field map calculated by three methods PDF, R-SHARP and iRSHARP, (d) Direction 4 Local field map calculated by three methods PDF, R-SHARP and iRSHARP.

Figure 4 shows the estimation results of the local field of the human brain under different imaging directions. (a) Phase map after unwrapping in 4 directions, (b) local field estimation result by PDF method, (c) local field estimation result by R-SHARP method, (d) local field estimation result by iRSHARP method.

Figure 5 shows the reconstruction results of human brain magnetic susceptibility under different imaging directions. (a) Magnetic susceptibility results for local field reconstructions estimated from PDF, (b) susceptibility results for local field reconstructions estimated from R-SHARP, (C) susceptibility results for local field reconstructions estimated from iRSHARP.

Detailed ways

The following embodiments will further illustrate the present invention in conjunction with the accompanying drawings.

Figure 1 presents a graph of human brain data in different imaging directions. The invention can effectively remove the background field, obtain an accurate local field, and at the same time protect the integrity of the brain tissue. The specific implementation process of the method is as follows:

1) First, carry out numerical simulation experiments. Create a 128ⅹ128ⅹ64 matrix, embed a large ellipsoid in the matrix to simulate the human brain, and then generate 5 small ellipsoids in the large ellipsoid to simulate the sinuses, blood vessels, substantia nigra, globus pallidus and caudate nucleus respectively. The magnetic susceptibility is 9.4ppm, 0.3ppm, 0.16ppm, 0.1ppm and 0.05ppm. The magnetic susceptibility of the remaining parts of the brain area is set to 0ppm, and an ellipsoid with a magnetic susceptibility value of 9.4ppm is added outside the VOI for simulation. Other background field sources, and Gaussian standard noise with SNR=40 is added to the simulation data. The sinuses were rotated 10° to the left, 20° to the right, and 10° to the back to simulate different imaging orientations, as shown in Figure 2.

2) Secondly, choose the PDF (Projection onto Dipole Field) method proposed by Liu, T. in 2011 and the R-SHARP (Sophisticated Harmonic Artifact Reduction for Phase Data Using Region Adaptive Kernel) method proposed by Fang, J.S. in 2017 as a comparison. It shows that the local field estimation algorithm based on the spatial distribution of the magnetic resonance dipole field proposed by the present invention can effectively remove the background field of the magnetic resonance magnetic susceptibility region under multi-directional imaging, and obtain an accurate local field. On the local field map (Fig. 3), as indicated by the arrows, the PDF method has strong phase residuals around the simulated sinuses that will produce severe susceptibility artifacts; R-SHARP in the first direction , the background field can be removed well, but there are small phase residuals in the other three directions, as shown in the enlarged view of Fig. 3; the iRSHARP method has no phase residuals in all four directions, and the running time is long Less than the R-SHARP method.

In order to verify the local field estimation of this method, the root mean square error (RMSE) is used to quantitatively evaluate the processing results of the three methods. Among them, the root mean square error is defined as:

Among them, B _int represents the calculated local field map, B ₀ represents the real local field, and n is the number of calculation points.

It can be seen from Table 1 that the iRSHARP delocalized field estimation is better than the other two methods, and the running time is shorter than that of R-SHARP.

Table 1: RMSE and runtime comparison of simulation graph experiments

method direction 1 direction 2 Direction 3 Direction 4 time(s) PDF 0.225 0.201 0.202 0.223 37.8 R-SHARP 0.145 0.145 0.145 0.146 122.1 iRSHARP 0.144 0.144 0.144 0.144 58.9

3) Finally, the human brain experimental data is carried out to further verify the feasibility of the method of the present invention in practical application. The experimental data was collected on a Philips 7T, 32-channel human body imager. The three-dimensional gradient echo sequence was used. The imaging parameters were TR/TE1/ΔTE=45ms, 2ms, 2ms, 8 echoes, and the imaging field of view was 220mm×220mm×110mm , the layer thickness is 1mm, and the data matrix is 224×224×110. Figure 4 lists the local field estimation results of PDF, R-SHARP and iRSHARP under different imaging directions. PDF leaves obvious directional phase residues in the brain tissue above the sinuses, and also exists in the limbic regions of the brain tissue Strong phase residual; the R-SHARP method can remove the background field well in direction 1, but there are weak phase residuals in the other three directions; the iRSHARP method can be very strong in all four directions. Nicely remove the background field. The resulting local field is subjected to susceptibility inversion, as shown in Figure 5. In the brain tissue surrounding the sinuses (shown by arrows), the residual phase left by the PDF method produces severe susceptibility artifacts and overwrites other regions ; The weak phase remaining in directions 2, 3, and 4 of the R-SHARP method also produces susceptibility artifacts that are observable to the naked eye; while the iRSHARP method removes the phase residual information well in all directions, and the reconstructed magnetization The brain tissue structure details can be distinguished on the rate map.

In MRI quantitative magnetic susceptibility imaging, the distribution of the dipole field of the brain tissue is related to the geometric structure of the tissue and the angle between the tissue and the main magnetic field. Therefore, the strong magnetic susceptibility areas such as the sinuses, the junction of the tissue and the skull will follow the head and the main magnetic field. The change of the imaging angle produces strong background field interference in different directions to the surrounding brain tissue, thus affecting the accurate estimation of the local field. The local field estimation algorithm based on the spatial distribution of the magnetic resonance dipole field uses the spatial information synthesized by the gradient value of the phase map, the phase amplitude and the normalized phase winding array to detect the strong magnetic susceptibility region in the brain tissue. The spatial information contains the evolution results of the dipole field at different times and in different directions. Therefore, the central weight of the Gaussian convolution kernel is modulated by these spatial information, which can effectively eliminate the background field of the strong magnetic susceptibility region in different directions and retain it accurately. Local field information generated by tissue, thereby effectively suppressing susceptibility artifacts. Therefore, in the multi-directional human brain data experiment, this method does not need to remove the strong magnetic susceptibility tissue around the sinuses and can ensure the integrity of the human brain tissue. The effect is significantly better than other existing methods, and it has potential clinical application value.

Claims (4)

1. Based on the local field estimation algorithm of magnetic resonance dipole field spatial distribution, it is characterized by comprising the following steps: 1) The winding number based on the phase diagram, whose model is,

In the model, K(r) is the winding number of the phase map, γ is the gyromagnetic ratio, TE is the echo time, B ^Δ is the total field map of the dipole field,

For the unwrapped phase diagram,

is the winding phase map, W[ ] is the phase winding operator, and r is the spatial coordinate of the voxel in the total field map of the dipole field; 2) Based on the standard deviation σ(r) of the Gaussian convolution kernel modulated by the amplitude, gradient and winding number of the phase map, the model is:

In the model, C is the positive adjustment constant, B ^Δ is the total field map of the dipole field,

is the gradient of the total field map of the dipole field, ε is a real number greater than zero, and K _N (r)∈[0,1] is the normalized phase map winding number; the thus established voxel-by-voxel spatial information correlation The convolution kernel is expressed as:

where σ(r) is the standard deviation of the Gaussian convolution kernel; 3) The region of interest is divided into regions with high magnetic susceptibility variation and uniform magnetic susceptibility regions based on spatial information. The model is:

in the model,

is the reciprocal of σ(r), denoted as

is the region segmentation threshold, defined as

in

and σ _max represent the total volume of all voxels in the dipole field map, respectively.

The mean and maximum value of ; δ is the unit shock function, B _int is the local field,

is a temporary variable,

represents a convolution operation; by

The region of interest is divided into high magnetic susceptibility variation region D ₁ and uniform magnetic susceptibility region D ₂ , and convolution operations are performed with Gaussian convolution kernel and spherical mean convolution respectively, and the background field is removed to obtain the local field. 2. the local field estimation algorithm based on magnetic resonance dipole field spatial distribution as claimed in claim 1, is characterized in that in step 1), the winding number of described phase map comprises the evolution result of different field map changes under different time , for local field estimation at any echo time. 3. the local field estimation algorithm based on magnetic resonance dipole field spatial distribution as claimed in claim 1, is characterized in that in step 2) in, the spatial information that the amplitude, gradient and winding number of described phase map are formed is body by body. Prime modulates the standard deviation of the Gaussian convolution kernel. When the spatial information value is larger, it means that there is a strong background field in the area, and a smaller Gaussian standard deviation is obtained, that is, the weight of the corresponding Gaussian kernel sphere is high and the surrounding point values decay rapidly. , to better suppress the background field and obtain an accurate local field estimate. 4. the local field estimation algorithm based on magnetic resonance dipole field spatial distribution as claimed in claim 1, is characterized in that in step 3), described based on spatial information, the region of interest is divided into high magnetic susceptibility variation region and uniform magnetization The rate change area is to speed up the operation speed of the algorithm, and the spatial information considers the field image components under different imaging directions, which is used for multi-directional local field estimation.

Images