A diffusion relaxation spectroscopy imaging method for measuring brain microstructure

Abstract

The present invention discloses a diffusion relaxation spectroscopy imaging method for measuring brain microstructure, comprising: acquiring diffusion magnetic resonance data of a target object at multiple echo times and multiple b values; establishing a diffusion relaxation spectroscopy imaging model for multi-tissue brain microstructure; obtaining diffusion relaxation times and diffusion sensitivity coefficient spectra of multi-tissue brain microstructure through a multi-parameter joint optimization method based on spherical averaging technique; and characterizing the obtained relaxation times, brain tissue volume fractions, and distribution functions to obtain brain tissue microstructure parameters. The present invention can provide tissue composition-specific non-T2-weighted intraneurite fraction and free water fraction, and solves the TE-dependence of traditional composition fractions.

Description

Technical Field

[0001] The present invention belongs to the technical fields of magnetic resonance imaging and medical image processing, and particularly relates to a diffusion relaxation spectroscopy imaging method for measuring brain microstructural organization. Background Art

[0002] The purpose of measuring brain microstructural organization is to achieve specificity for neuronal microstructures to generate unique clinical biomarkers for non-invasive imaging. Since biological tissues are very complex, microscopic structure modeling usually requires simulating the tissue structure based on specific assumptions to obtain the target indicators of interest. For microscopic parameter-specific imaging, an effective way to achieve this goal is to increase the measurement dimension of data to better distinguish tissue microcomponents, such as by utilizing the synergistic effects between diffusion and T1, T2 relaxation, and spectroscopy imaging. Related progress includes relaxation-diffusion related imaging techniques, which have been applied to animal and ex vivo brains. Recently, this multi-dimensional imaging method has been combined with microscopic structure modeling to improve the estimation of tissue microstructure parameters for human brain imaging. However, multi-dimensional imaging further increases the difficulties of data acquisition and model fitting while providing specific parameters.

[0003] In addition, under the existing diffusion magnetic resonance data acquisition scheme, the microscopic structure parameters are not sensitive to the changes in diffusion signals, which limits the specificity of the model parameters. Among them, common multi-component models only consider the differences in diffusion characteristics between different components, but do not consider the differences in other characteristics such as relaxation time, and diffusion magnetic resonance data are often acquired with a single TE. This results in the quantitative indicators derived from the model being interfered by T1 and T2 weighting. When there are T1 or T2 weighting differences between components, the model cannot distinguish the relaxation and diffusion characteristics between components. Between the intra- and extra-neurite components in tissue, the difference in T2 is much smaller, but it has also been shown to cause the derived parameters of some diffusion models in white matter to have echo-time dependence, which will reduce the specificity of the model-derived quantitative indicators and the interpretability of research results, especially for studying the processes of brain development and maturation accompanied by significant changes in T2 relaxation time.

[0004] In summary, the characteristics of the brain tissue microstructure lie in the heterogeneity between the diffusion rate and the transverse relaxation rate. Diffusion magnetic resonance imaging and its higher-order models can non-invasively detect the brain tissue microstructure. Among them, the signal differences among the water molecules in the restricted, hindered, and free components can be characterized by the higher-order diffusion model, so as to estimate the relative proportions of the cell bodies, axon fibers, and free components within the voxel, and thus be used to detect the developmental, degenerative, and disease-driven changes in the brain tissue microstructure. However, the accurate characterization of tissue components is affected not only by the tissue-dependent diffusion coefficient but also by the transverse relaxation rate. Explicitly considering the relaxation-diffusion coupling relationship may significantly improve the characterization of the brain tissue microstructure. Standard diffusion magnetic resonance imaging techniques with a single echo time mainly provide information about the diffusion rate, while relaxation-diffusion magnetic resonance imaging involves multiple echo times and multiple diffusion weighting intensities, and can detect the coupling relationship of tissue characteristics between relaxation and diffusion rates. Therefore, it is of great significance to propose a diffusion relaxation spectroscopy imaging method for measuring the microstructure of the brain. Summary of the Invention

[0005] In order to overcome the challenges of the existing technology, the present invention provides a diffusion relaxation spectroscopy imaging method, which detects the relaxation-diffusion coupling relationship and quantitatively measures the brain tissue microstructure parameters, and solves the TE dependence of the traditional component fraction.

[0006] The technical solution for achieving the purpose of the present invention is as follows:

[0007] A diffusion relaxation spectroscopy imaging method for measuring the microstructure of the brain includes the following steps:

[0008] Step 1: Perform diffusion-weighted imaging scanning on the brain through a magnetic resonance device, and perform sampling according to the settings of different b values and different echo times to obtain diffusion magnetic resonance imaging images with multiple b values and multiple echo times;

[0009] Step 2: Fit the diffusion magnetic resonance imaging images with multiple b values and multiple echo times based on the spherical mean diffusion relaxation spectroscopy imaging model to obtain relaxation times specific to the brain tissue and dependent on the b value, as well as the brain tissue volume fraction;

[0010] Step 3: Fit the diffusion magnetic resonance images with multiple b values and multiple echo times based on the diffusion relaxation spectroscopy imaging model and the relaxation times obtained in Step 2 to obtain a fiber direction distribution function specific to the brain tissue without T2 weighting;

[0011] Step 4: Characterize the relaxation times, brain tissue volume fractions obtained in Step 2, and the distribution function obtained in Step 3 to obtain brain tissue microstructure parameters.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can provide tissue composition-specific non-T2 weighted neurite fraction and free water fraction, solving the TE-dependence of traditional composition fractions; at the same time, it also provides the T2 relaxation times inside and outside the neurite that cannot be distinguished only by relaxation methods; meanwhile, through theoretical derivation and experimental verification, we show that these specific parameters can be estimated and achieved from the scanning equipment in the clinical environment; the present invention improves the tissue parameter specificity and quantitative characteristics, and will bring new insights to the research based on diffusion magnetic resonance imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flowchart of the diffusion relaxation spectrum imaging method for measuring brain microstructure according to the present invention;

[0014] Figure 2 is the effect diagram of the method according to the present invention on real human brain data (TE = 75, 85, 95, 105, 115, 125, 135 ms, b = 400, 800, 1600, 3200 s / mm 2 2), where Figure (a) is the effect diagram of a healthy sample, and Figure (b) is the effect diagram of a glioma patient sample;

[0015] Figure 3 is the relationship analysis diagram of the relaxation time and b value calculated by the method according to the present invention on real human brain data;

[0016] Figure 4 is the effect diagram of the method according to the present invention on different glioma patients. (a) shows the relaxation time effect diagram of the restricted brain tissue in the white matter lesion; (b) is the estimated diagram of the neurite morphology by the traditional single TE method; (c) is the estimated diagram of the neurite morphology by the method according to the present invention; (d) is the effect diagram of the calculated axon radius by the method according to the present invention;

[0017] Figure 5 is the fiber direction distribution calculated by the method according to the present invention on healthy human brain data, showing the visual comparison of the fiber direction distribution with and without explicitly considering relaxation. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1

[0020] This embodiment proposes a diffusion relaxation spectroscopy imaging method, which acquires multi-b value (diffusion sensitivity coefficient) diffusion magnetic resonance data of a target object at a first echo time; while keeping the diffusion time and the scan repetition time unchanged, repeatedly acquires the above multi-b value diffusion magnetic resonance data for different echo times (TE); according to the acquired multi-b value diffusion magnetic resonance data of multiple echoes, a unified strategy is proposed to estimate (1) the T2 relaxation time specific to brain tissue; (2) the multi-scale brain microstructure tissue parameters without T2 weighting; (3) the multi-scale fiber orientation distribution function without T2 weighting.

[0021] Combined with Figure 1 , the diffusion relaxation spectroscopy imaging method is specifically described as follows:

[0022] A diffusion relaxation spectroscopy imaging method for measuring brain microstructure includes the following steps:

[0023] Step 1: Perform diffusion-weighted imaging scan on the brain of a subject by a magnetic resonance device, and sample according to the settings of different b values and different echo times to obtain diffusion magnetic resonance imaging images with multi-b values and multi-echo times;

[0024] Step 2: Fit the diffusion magnetic resonance imaging images with multi-b values and multi-echo times based on the spherical mean diffusion relaxation spectroscopy imaging model to obtain the relaxation time specific to brain tissue and b value-dependent, as well as the brain tissue volume fraction;

[0025] Step 3: Fit the diffusion magnetic resonance images with multi-b values and multi-echo times based on the diffusion relaxation spectroscopy imaging model and the relaxation time obtained in Step 2 to obtain a fiber orientation distribution function without T2 weighting and specific to brain tissue;

[0026] Step 4: Characterize the relaxation time, brain tissue volume fraction obtained in Step 2, and the distribution function obtained in Step 3 to obtain brain tissue microstructure parameters.

[0027] Further, the diffusion magnetic resonance imaging data with multi-b values and multi-echo times in Step 1 is obtained through the following steps:

[0028] Step 1.1: Set a multi-b value gradient table, and as the b value increases, the number of gradient directions increases by an order of magnitude of 2;

[0029] Step 1.2: Under the condition that the diffusion time and the scan repetition time are both kept fixed, perform diffusion magnetic resonance imaging with a single echo time (TE) of multi-b values on the subject to obtain full-sampling magnetic resonance images corresponding to each b value;

[0030] Step 1.3: Increase the echo time (TE) in equal steps, repeatedly execute Step 1.2, and obtain diffusion magnetic resonance imaging data with multi-b values and multi-echo times.

[0031] Further, the spherical mean diffusion relaxation spectrum imaging model is the spherical averaging result of the diffusion relaxation spectrum imaging model.

[0032] Further, in step 2, the spherical mean diffusion relaxation spectrum imaging model excludes the influence of the fiber direction distribution on the relaxation time.

[0033] Further, in step 2, the spherical mean diffusion relaxation spectrum imaging model characterizes the microscopic tissue environment of the brain as the restricted, hindered, and free diffusion parts within the brain tissue, and distinguishes the restricted and hindered parts according to a geometric curvature.

[0034] Further, in step 2, each diffusion part in the spherical mean diffusion relaxation spectrum imaging model is a weighted combination of the spherical mean signals of a series of spin packets, and its corresponding volume fraction is calculated, so as to be able to detect the relaxation-diffusion coupling across a series of diffusion scales.

[0035] Further, the spherical mean diffusion relaxation spectrum imaging model in step 2 is the spherical averaging result of the diffusion relaxation spectrum imaging model in step 3; when the relaxation time is obtained by solving in step 2, step 3 will generate a tissue-specific multi-scale fiber direction distribution function.

[0036] Further, the microscopic structure parameters of the brain tissue in step 4 are obtained through the following steps:

[0037] Step 4.1: The direction-invariant microscopic structure parameters without T2 relaxation influence are calculated by the tissue-specific volume fraction calculated in step 2, including axon shape index, intra-axon volume fraction, nerve density, microscopic anisotropy index, axial diffusivity of axons, radial diffusivity of axons, etc.;

[0038] Step 4.2: The direction-variable microscopic structure parameters without T2 relaxation influence are calculated by the tissue-specific fiber direction distribution function calculated in step 3, including directional restricted diffusivity, directional hindered diffusivity, etc.

[0039] The method of the present invention is called diffusion relaxation spectrum imaging (RDSI). This method can directly estimate the proportion of non-T2 weighted brain microstructural tissue fractions and the T2 relaxation time specific to brain tissue, and has the following characteristics:

[0040] The spherical mean diffusion relaxation spectrum imaging model in the method of the present invention excludes the influence of the fiber direction distribution on the relaxation time;

[0041] In the spherical mean diffusion relaxation spectrum imaging model of the method of the present invention, each diffusion part is a weighted combination of the spherical mean signals of a series of spin packets, and its corresponding volume fraction is calculated, so as to be able to detect relaxation-diffusion coupling spanning a series of diffusion scales;

[0042] The diffusion relaxation spectrum model in the method of the present invention reveals a multi-scale fiber direction distribution with tissue specificity;

[0043] Using the tissue-specific volume fraction calculated by the present invention, it is possible to reveal direction-invariant microstructure parameters unaffected by T2 relaxation, including axon morphology index, intra-axon volume fraction, nerve density, microscopic anisotropy index, axial diffusivity of axons, radial diffusivity of axons, etc.;

[0044] Using the tissue-specific fiber direction distribution function calculated by the present invention, it is possible to reveal direction-variable microstructure parameters unaffected by T2 relaxation, including restricted diffusion rate in the direction, hindered diffusion rate in the direction, etc.

[0045] Example 2

[0046] Based on Example 1, the specific designs of the diffusion relaxation spectrum imaging model and the spherical mean diffusion relaxation spectrum imaging model include:

[0047] The specific form of the diffusion relaxation spectrum imaging model is:

[0048] The diffusion relaxation spectrum imaging model models the diffusion magnetic resonance signal S(τ, b, g) acquired according to a specific echo time τ, diffusion gradient direction g, and b value as And it is extended to a multi-tissue model as:

[0049]

[0050] Among them, S r (b, h), S h (b, g) and S f (b) respectively represent the diffusion signals of the restricted, hindered, and free diffusion parts in the brain tissue; the apparent relaxation rate r(b)=1 / T2(b) dependent on the gradient strength b is estimated by repeatedly acquiring diffusion signals at multiple different echo times, and is expressed as:

[0051]

[0052]

[0053] Among them, R(b, g, D r ), R(b, g, D h ), R(b, g, D f ) are related to the apparent diffusion coefficient D r , Dh , D f -related brain tissue-specific response functions, based on the multi-scale response functions and the apparent diffusion coefficient, a tissue-specific multi-scale fiber orientation distribution function is generated, represented as spherical harmonic coefficients f(D r ), f(D h ), f(D f ); the operator A maps the spherical harmonic coefficients back to the fiber orientation distribution function.

[0054] To effectively solve the above model, the present invention simplifies the above equation through spherical averaging technology, eliminates the influence of the fiber orientation distribution on the calculation of the T2 relaxation time, and thus solves the above model step by step. The spherical average diffusion relaxation spectroscopy imaging model is specifically:

[0055]

[0056]

[0057] where w(D r ), w(D h ), w(D f ) are the volume fractions corresponding to the restricted, hindered, and free diffusion parts within the brain tissue, k(b, D r ), k(b, D h ), k(b, D f ) are the spherical averages of the response functions R(b, g, D r ), R(b, g, D h ), R(b, g, D f ), and are calculated through the following equation

[0058]

[0059]

[0060]

[0061] In the present invention, the apparent diffusion coefficient D r , D h , D f can be parametrically represented by the parallel diffusion coefficient λ || and the perpendicular diffusion coefficient λ ⊥ in the corresponding sets of restricted Λ r , hindered Λ h , free diffusion Λ f in the brain tissue, and the restricted Λ r and the hindered Λ hportion. Thus, the spherical average signal can be regarded as a weighted combination of spherical average signals from a series of spin packets, which enables us to detect relaxation-diffusion coupling across a series of diffusion scales. The average diffusion coefficient λ || and the perpendicular diffusion coefficient λ ⊥ range from [0e-33e-3], and the geometric curvature φ is π 2 / 4.

[0062] To solve this model, the present invention first solves the spherical average coefficient in the relaxation mode, then decomposes the relaxation term from the spherical average coefficient in the relaxation mode and solves the relaxation rate. Finally, the estimated relaxation rate is used to solve the fiber direction distribution function, as follows:

[0063] We rewrite the above model as

[0064]

[0065] where is the spherical average signal of multiple echoes with multiple b-values, K is the spherical average matrix of the response function, E is the relaxation matrix, W is the volume fraction matrix, is the Kronecker product operator. To solve for X, we transform the above equation into the following problem for solution

[0066]

[0067] After solving for X, E and W can be decomposed by minimizing the following constrained nonlinear multivariable problem

[0068]

[0069] After solving for E, the estimation model of the fiber direction distribution function can be rewritten as a strictly convex quadratic programming problem for solution

[0070]

[0071] After solving for W, the axonal morphology indices (including Mean NR, Std.NR, and Cov.NR) can be calculated as follows:

[0072]

[0073] where ∈>0 is the pulse scale and depends only on the pulse width δ and diffusion time Δ of the diffusion gradient,

[0074]

[0075]

[0076] The subject is a human or an experimental animal, and the experimental animal is a mouse, a monkey, a dog or a pig.

[0077] Obtained through theoretical derivation and experimental verification Figures 2 to 5 , Figure 2 is the effect diagram of the method according to the present invention on real human brain data (TE = 75, 85, 95, 105, 115, 125, 135 ms, b = 400, 800, 1600, 3200 s / mm 2 ), the left side is a healthy sample, and the right side is a glioma patient sample. Figure 3 is the relationship analysis diagram of the relaxation time and b value calculated by the method according to the present invention on real human brain data, showing that at higher b values, the relaxation times of more voxels in the restricted brain tissue are between 100 and 200 ms, especially for high-grade gliomas. Figure 4 is the effect diagram of the method according to the present invention on different glioma patients. (a) shows the relaxation time of the restricted brain tissue in the white matter lesion, indicating that the relaxation time of glioma is longer than that of normal white matter tissue; (b-c) shows that the method of the present invention (c) is more sensitive to the estimation of neurite morphology than the traditional single TE method (b); (c-d) is the axon radius calculated by the method according to the present invention, showing that the method of the present invention is very sensitive to the description of the tumor range and the standard of the microenvironment within the tumor. Figure 5 is the fiber direction distribution calculated by the method according to the present invention on real healthy human brain data, respectively showing the visual comparison of the fiber direction distribution when relaxation is explicitly considered and when relaxation is not considered. Through Figures 2 - 5 , it shows that these specific parameters can be estimated and realized from the scanning equipment in the clinical environment, verifying that the present invention solves the TE dependence of the traditional component fraction; at the same time, it also provides the T2 relaxation times inside and outside the neurite that cannot be distinguished only by the relaxation method.

Claims

1. A diffusion relaxation spectroscopy imaging method for measuring brain microstructure, characterized in that It includes the following steps: Step 1: Perform diffusion-weighted imaging scan on the brain using a magnetic resonance device, and sample according to the settings of different b-values and different echo times to obtain diffusion magnetic resonance imaging images with multiple b-values and multiple echo times; Step 2: Fit the diffusion magnetic resonance imaging images with multiple b-values and multiple echo times based on the spherical mean diffusion relaxation spectrum imaging model to obtain relaxation times with brain tissue specificity and b-value dependence, as well as brain tissue volume fractions; Step 3: Fit the diffusion magnetic resonance images with multiple b-values and multiple echo times based on the diffusion relaxation spectrum imaging model and the relaxation times obtained in Step 2 to obtain a fiber direction distribution function without T2 weighting and with brain tissue specificity; Step 4: Characterize the relaxation times, brain tissue volume fractions obtained in Step 2, and the distribution function obtained in Step 3 to obtain brain tissue microstructure parameters; The specific form of the diffusion relaxation spectrum imaging model is: The diffusion relaxation spectrum imaging model models the diffusion magnetic resonance signal S(τ, b, g) acquired according to a specific echo time τ, diffusion gradient direction g, and b value as And it is extended to a multi-tissue model as: where S r (b,g), S h (b,g) and Sx(b) represent the diffusion signals of the restricted, hindered, and freely diffusing parts in the brain tissue, respectively; the gradient strength b-dependent apparent relaxation rate r(b) = 1 / T2(b) is estimated by repeatedly acquiring diffusion signals at multiple different echo times and is expressed as: where R(b, g, D r ), R(b, g, D h ), R(b, g, D f ) are response functions specific to brain tissue related to the apparent diffusion coefficient D r , D h , D f . Based on the multi-scale response function and the apparent diffusion coefficient, a multi-scale fiber orientation distribution function specific to tissue is generated, which is represented as spherical harmonic coefficients f(D r ), f(D h ), f(D f ) in the spherical harmonic space; the operator maps the spherical harmonic coefficients back to the fiber orientation distribution function; The specific form of the spherical mean diffusion relaxation spectrum imaging model is: where w(D r ), w(D h ), w(D f ) are the volume fractions corresponding to the restricted, hindered, and free diffusion parts in the brain tissue, and k(b, D r ), k(b, D h ), k(b, D f ) are the spherical averages of the response functions R(b, g, D r ), R(b, g, D h ), R(b, g, D f ), which are calculated specifically through the following equations λ || is the parallel diffusion coefficient, λ ⊥ is the vertical diffusion coefficient.

2. The diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 1, characterized in that The specific content of Step 1 includes: Step 1.1: Set a multi-b-value gradient table. As the b-value increases, the number of gradient directions increases by an order of magnitude of 2; Step 1.2: Under the condition that both the diffusion time and the scan repetition time are kept fixed, perform diffusion magnetic resonance imaging with a single echo time for multiple b-values to obtain full-sampling magnetic resonance images corresponding to each b-value; Step 1.3: Increase the echo time in equal steps, and repeat Step 1.2 to obtain diffusion magnetic resonance imaging data with multiple b-values and multiple echo times.

3. A diffusion relaxation spectrum imaging method for measuring brain microstructure according to claim 1, characterized in that, The spherical mean diffusion relaxation spectrum imaging model is the spherical averaging result of the diffusion relaxation spectrum imaging model.

4. A diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 3, characterized in that, The spherical mean diffusion relaxation spectrum imaging model excludes the influence of fiber direction distribution on the relaxation time, characterizes the microscopic tissue environment of the brain as the restricted, blocked, and free diffusion parts within the brain tissue, and distinguishes the restricted and blocked parts according to a geometric curvature; each diffusion part is a weighted combination of the spherical mean signals of a series of spin packets, and its corresponding volume fraction is calculated to detect the relaxation-diffusion coupling across a series of diffusion scales.

5. A diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 1, characterized in that, The apparent diffusion coefficient D r , D h , D f is parameterized by the parallel diffusion coefficient λ || and the perpendicular diffusion coefficient λ ⊥ in the corresponding sets of restricted Λ r , blocked Λ h , and free diffusion Λ f , and the restricted Λ r and blocked Λ h parts are distinguished according to a geometric curvature φ.

6. The diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 5, characterized in that The parallel diffusion coefficient λ || and the vertical diffusion coefficient λ ⊥ range from [0e-3 3e-3], and the geometric curvature φ is π 2 / 4.

7. A diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 1, characterized in that, The diffusion relaxation spectrum imaging model and the spherical mean diffusion relaxation spectrum imaging model are solved in the following way: First, solve the spherical mean coefficients of the diffusion relaxation spectrum imaging model, then decompose the relaxation terms from the spherical mean coefficients and solve the relaxation rates, and finally use the estimated relaxation rates to solve the fiber direction distribution function, specifically as follows: The model equation of the spherical mean diffusion relaxation spectrum imaging model is rewritten as: where is the spherical mean signal of multiple echoes with multiple b - values, K is the spherical mean matrix of the response function, E is the relaxation matrix, W is the volume fraction matrix, is the Kronecker product operator. To solve for X, the above equation is transformed into the following problem for solution: After solving for X, E and W are decomposed by minimizing the following constrained non-linear multivariate problem: After solving for E, the estimation model of the fiber direction distribution function is rewritten as a strictly convex quadratic programming problem for solution:

8. A diffusion relaxation spectroscopy imaging method for measuring brain microstructure according to claim 7, characterized in that, The specific content of Step 4 includes: Step 4.1: The direction-invariant microstructure parameters without T2 relaxation effects are calculated using the tissue-specific volume fractions calculated in Step 2, including axon morphology index, intra-axon volume fraction, nerve density, microscopic anisotropy index, axial diffusivity of axons, and radial diffusivity of axons; Step 4.

2. The direction variability microstructure parameters without T2 relaxation effects are calculated using the tissue-specific fiber orientation distribution function obtained in Step 3. The axon shape index includes the mean axon radius Mean NR, the internal deviation index Std.NR, and the relative axon radius Cov.NR, specifically: where ∈>0 is the pulse scale and depends only on the pulse width δ and diffusion time Δ of the diffusion gradient.

Images