A method for evaluating key fiber tracts in early mild cognitive impairment

Abstract

A method for evaluating key fiber bundles in early mild cognitive impairment. The abnormal diffusion information of brain white matter caused by diseases such as Alzheimer's disease, which is currently the most studied, is a prominent manifestation of brain plasticity. Its impact on fibers includes changes in brain positioning and corresponding indicators. Fiber tracking can extract diffusion information in fibers. In addition to providing fiber direction, it can also display the diffusion activity of water molecules in brain cells and the network structure of the brain, providing relevant references for researchers to find brain changes caused by diseases. The present invention optimizes and improves the more mature fiber automatic quantification technology currently available, and at the same time adjusts the parameters of the Mahalanobis distance formula to clear and screen the fiber bundles initially obtained, which can more comprehensively and accurately extract the whole-brain fiber bundles. The verification of key fiber bundles with significant differences between the early mild cognitive impairment patient group and the healthy control group can provide certain reference value and help for the assessment and prediction of early cognitive impairment.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing, and in particular relates to a method for evaluating key white matter fiber bundles in early mild cognitive impairment based on brain medical images. Background Art

[0002] As a non-invasive in vivo method, diffusion magnetic resonance imaging (diffusion MRI) is widely used in brain research. It has played a significant role in obtaining clinical information about neuronal fiber structure and assisting in understanding functional connectivity between different brain regions. Based on dMRI data, researchers typically use diffusion tensor imaging (DTI) to extract brain diffusion information and compare differences in diffusion information between control and control groups to study brain plasticity. Diffusion information includes fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). FA measures the anisotropy index of water molecules and closely reflects the integrity of white matter axonal membranes and the directionality of water molecule translation. MD, the mean diffusivity, reflects the overall diffusion level of water molecules and diffusion resistance. It is generally the opposite of FA and reflects the overall magnitude of water diffusion by indicating cell expansion and cell density. AD and RD diffusivities are considered in vivo surrogate markers of myelin and axonal damage, respectively. These indicators are closely related to the maturation process, lesions and higher-level cognitive functions of the brain's microstructure. These indicators can effectively reflect the intrinsic changes in the nerve fiber bundles of an individual's brain under different influences, which is the manifestation of brain plasticity.

[0003] Taking disease-induced brain plasticity as an example, abnormal white matter diffusion patterns in diseases such as Alzheimer's disease, a prominent manifestation of brain plasticity, are currently under extensive study. These patterns affect fiber orientation and alterations in corresponding parameters. Fiber tracking can extract diffusion information from fibers. Besides providing fiber orientation, it can also reveal the diffusion activity of water molecules within brain cells and the network structure of the brain, providing researchers with insights into disease-induced brain changes. For example, Weiler et al. found fiber bundles with significantly reduced FA values ​​in the white matter of Alzheimer's patients. Furthermore, Wang et al., conducting a cross-group analysis of 30 Parkinson's disease patients and 28 healthy controls, found significant differences in localized patterns in 13 hemispheric fiber bundles and 8 commissural fiber bundles, including the cingulate fasciculus, inferior fronto-occipital fasciculus, corpus callosum, uncinate fasciculus, and superior longitudinal fasciculus. Yogesh et al. conducted quantitative and statistical studies on neural fibers in patients with schizophrenia and found that more than half of the studied fiber bundles exhibited differences in relevant diffusion parameters. Summary of the Invention

[0004] To overcome the shortcomings of the existing technology, the present invention optimizes and improves the currently mature automatic fiber quantification technology. At the same time, the parameters of the Mahalanobis distance formula are adjusted to clear and filter the initially obtained fiber bundles. This can more comprehensively and accurately extract the fiber bundles of the whole brain. The verification of key fiber bundles with significant differences between the early mild cognitive impairment patient group and the healthy control group can provide certain reference value and help for the assessment and prediction of early cognitive impairment. Specifically, it includes:

[0005] Step (1): preprocessing the diffusion tensor imaging images of patients with early mild cognitive impairment and healthy subjects, including data format conversion and data resampling;

[0006] Step (2): using the preprocessing results of step (1) as input, complete the whole-brain fiber tract tracking;

[0007] Step (3): Extract the tracked fiber bundles and remove the erroneous fiber bundles to obtain a preliminary whole-brain fiber bundle set, and then use the Mahalanobis distance formula to obtain the optimal de-redundant fiber bundle set;

[0008] Step (4): Four indices, namely, partial anisotropy index, mean diffusivity, axial diffusivity, and radial diffusivity, were used to measure brain plasticity, and the indices were calculated for the cleared fiber bundles.

[0009] Step (5): Repeat steps (2) to (4) for the two groups of image data of patients with early mild cognitive impairment and healthy subjects, and then perform inter-group comparison analysis on the indicators obtained from the two groups of data;

[0010] Step (6): Evaluate the inter-group difference fiber bundle groups obtained in step (5), and calculate their ROC curve areas to obtain evaluation reference results.

[0011] Furthermore, the step (1) specifically includes:

[0012] Step (1.1) Use dcm2niigui software to convert the original DICOM medical data format into .nii, bvec, and bval files as input files for fiber tract tracking. The .nii file contains image data, and the bvec and bval files contain the eigenvectors and eigenvalues ​​of the image data, respectively.

[0013] Step (1.2) resamples the voxels of the original image to a size of 2×2×2mm 3 .

[0014] Furthermore, the step (2) specifically includes: using a deterministic tracking algorithm to complete whole-brain fiber bundle tracking, that is, given one or more starting points, searching for a new tracking direction according to a set calculation method, continuously propagating forward until the fiber bundle is stopped from tracking when the termination condition is reached, and finally obtaining a fiber streamline trajectory; wherein the termination condition is set to the following six: (1) FA≤0.2 or angle≥45°; (2) FA≤0.25 or angle≥45°; (3) FA≤0.3 or angle≥45°; (4) FA≤0.2 or angle≥50°; (5) FA≤0.2 or angle≥55°; (6) FA≤0.2 or angle≥60°, wherein FA is anisotropy index and angle is fiber bundle bending angle; six deterministic whole-brain fiber bundle trackings are performed on the same individual according to the six termination conditions to obtain multiple groups of whole-brain fiber bundle sets.

[0015] Furthermore, the step (3) specifically includes:

[0016] Step (3.1) first calculates the mean of each whole-brain fiber bundle set, then removes the fiber bundles that are less than 10% of the mean and greater than 90% of the mean in the fiber bundle set to obtain a preliminary whole-brain fiber bundle set;

[0017] Step (3.2) uses the starting and ending regions of the fiber bundles in the MNI standard space as regions of interest (ROIs), and screens out fiber bundles that pass through both ROIs from the preliminary whole-brain fiber bundle set;

[0018] Step (3.3) averages the binary masks of the relevant brain fiber bundle positions of multiple normal brain data in the standard coordinate system to obtain a probability map, wherein the probability map is the JHU white matter fiber bundle map. After comparison with the probability map, the fiber bundles passing through the low-probability position will be removed;

[0019] In step (3.4), the Mahalanobis distance formula is used to remove fiber bundles. The specific parameters for obtaining the optimal de-redundant fiber bundle set are: the distance threshold is set to 5, the length threshold is set to 5, the minimum number of fiber bundles in the cluster is 20, and the number of cycles is 5.

[0020] Furthermore, step (5) specifically includes: resampling each fiber in each fiber bundle set to 100 equidistant nodes to quantify the diffusion characteristics of the central part of the fiber bundle, and then using one-way analysis of variance in SPSS software to analyze the four index values ​​of the two groups of subjects, and performing FDR correction. When p < 0.05, the difference is considered statistically significant, and finally a partial fiber bundle group with significant differences is obtained.

[0021] Furthermore, step (6) specifically includes: extracting the four index values ​​of each fiber bundle group of each subject in each group to form four pairs of key fiber bundle index groups, and then performing receiver operating characteristic curve (ROC) on all groups in the four groups, calculating the area under the ROC curve of the four groups, that is, the AUC value, which ranges from [0, 1], and taking the fiber bundles and characteristic indicators corresponding to all groups with AUC values ​​greater than 0.85 as the final evaluation reference.

[0022] The beneficial effects of the present invention are:

[0023] 1. Based on the automatic fiber tract tracking and extraction algorithm, the tracking and extraction process parameters have been optimized, which increases the number of fiber tracts obtained while making the extraction results more reliable. Without affecting the execution speed, it can more comprehensively track fiber tracts throughout the brain.

[0024] 2. The use of fiber bundle ROI and probability maps helps to show the unique physiological structural characteristics of medical images. The tracking and extraction of fiber bundles is more stable. Combined with SPSS software to analyze fiber bundle differences, it can effectively show the statistical differences of certain key fiber bundles in different groups of subjects. The screening and evaluation of the sensitivity and specificity of multiple key fiber bundles and characteristic indicators have certain reference value and help in the auxiliary diagnosis of Alzheimer's disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is an implementation flow chart of the present invention;

[0026] Figure 2 is a schematic diagram of the fiber bundles extracted by the tracking algorithm of the present invention;

[0027] Figure 3 is a schematic diagram of the fiber bundle system interface of the present invention;

[0028] Figure 4 This is a schematic diagram of the ROC curve of a key fiber bundle

[0029] Figure 5 AUC value diagram for a key fiber bundle. DETAILED DESCRIPTION

[0030] The following describes the specific embodiments of the present invention in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that the following description is only used to explain the present invention and is not intended to limit the present invention.

[0031] The details are as follows:

[0032] Step (1) is to preprocess the diffusion tensor imaging images, including images of patients with early mild cognitive impairment and images of healthy people, and specifically includes the following steps:

[0033] Step (1.1) Use dcm2niigui software to convert the original DICOM medical data format into .nii, bvec, and bval files, where the .nii file is the image data, and the bvec and bval files contain the eigenvectors ε1, ε2, ε3 and eigenvalues ​​λ1, λ2, λ3 of the image data, respectively.

[0034] The basic idea of ​​the current diffusion tensor imaging model is to use the Stejskal-Tanner signal conversion model to estimate the diffusion tensor obtained by DT-MRI:

[0035]

[0036] Where b is the diffusion sensitivity coefficient, D is the diffusion coefficient matrix, g is the magnetic resonance diffusion gradient direction, S(g) is the observation signal of each gradient, and S0 is the signal in the direction without magnetic resonance diffusion gradient.

[0037] This method uses DWI images from at least six different directions to calculate the diffusion tensor matrix D (a 3×3 positive symmetric matrix) of any image, thereby describing the motion information of water molecules in three dimensions:

[0038]

[0039] Step (1.2) resamples the voxels of the original image to a size of 2×2×2mm 3 , which is convenient for tracking and processing.

[0040] Step (2) uses the preprocessing results in step (1) as input to complete whole-brain fiber tract tracking, which specifically includes the following steps:

[0041] Step (2.1) uses a deterministic tracking algorithm to complete whole-brain fiber tract tracking. The principle of the deterministic tracking method is that since white matter fiber tracts are continuously and smoothly distributed in the brain, they can be represented by a three-dimensional spatial curve r(s), where s is the arc length. Given that diffusion MRI data reflects the fiber structure information contained in the voxel, the tangent at the arc length s should be consistent with the voxel direction obtained by fiber modeling, such as the main characteristic direction of the diffusion tensor or the extreme value direction obtained by the spherical deconvolution model, that is:

[0042]

[0043] The above equation can be converted into the following integral form for solution:

[0044]

[0045] In the actual solution process, Equation (4) is usually discretized and approximated. For example, for the i-th position r i , in this direction v(r i ) and take a step of △t to get the position of the next point on the curve:

[0046] r i+1 =r i +v(r i )·△t (5)

[0047] Given one or more starting points, a new tracking direction is searched according to the set calculation method, and it is continuously propagated forward until certain termination conditions are reached and the fiber bundle is stopped from being tracked, and finally a fiber streamline trajectory is obtained.

[0048] In step (2.2), the termination conditions are set to the following six types: (1) FA≤0.2 or angle≥45°; (2) FA≤0.25 or angle≥45°; (3) FA≤0.3 or angle≥45°; (4) FA≤0.2 or angle≥50°; (5) FA≤0.2 or angle≥55°; (6) FA≤0.2 or angle≥60°. Five deterministic whole-brain fiber tract tracings are performed on the same individual to obtain multiple sets of whole-brain fiber tract sets {fiber ij ,i=1,2,...,18,j=1,2,...,5}, subscript i represents 18 fiber bundles, and subscript j represents 5 tracking times.

[0049] Step (3) is the process of extracting the tracked fiber bundles and removing the erroneous fiber bundles, which specifically includes the following steps:

[0050] Step (3.1) is to perform superposition, averaging and numerical comparison of multiple groups of whole-brain fiber bundles. First, the mean of each fiber bundle is calculated, and then the fiber bundle {fiberij , i=1,2,...,18, j=1,2,...,5} contains some fiber bundles with values ​​less than 90% and greater than 90% of the mean, and the preliminary whole-brain fiber bundle set is obtained, which is recorded as {fiber i ,i=1,2,...,18}.

[0051] In step (3.2), the starting and ending regions of the fiber bundles in the MNI standard space are used as regions of interest (ROIs). The fiber bundles that pass through both ROIs are screened out from the preliminary whole-brain fiber bundle set. Only the fiber bundles that pass through both ROIs are the target fiber bundles. The positions of the starting and ending regions of the fiber bundles are determined by the anatomical positions.

[0052] Step (3.3) removes erroneous fibers based on the probability maps of each key fiber bundle. To match the segmented nerve fibers of each individual, we define the start and end ROIs and probability maps of the key fiber bundles in MNI space. When segmenting the key fibers of each individual, the ROIs and probability maps are aligned from MNI space to the individual's DTI space. To address the problem of individual differences, a nonlinear registration method is used to extract features from the individual image and the target image to obtain feature points. By performing a similarity measurement, matching feature point pairs are found. The image space coordinate transformation parameters are obtained from the matching feature point pairs, and finally the coordinate transformation parameters are used for image registration.

[0053] In the process of removing erroneous fibers, each fiber object in the individual is compared with the standard probability atlas. The probability atlas uses a twelve-mode affine transformation to co-register the DTI images of 28 subjects to the template JHU-DTI and the template MNIICBM152. The resulting affine transformation matrix is ​​then applied to the fiber bundles of the subject to transform it into the template. The binary masks of the relevant brain fiber bundle positions of the 28 subjects in the standard coordinate system are averaged to generate the probability map. The probability map needs to be transformed into the individual brain space when used. After the atlas comparison, the fibers passing through the low probability positions will be removed.

[0054] In step (3.4), the Mahalanobis distance formula is used to remove fiber bundles. The specific parameters for obtaining the optimal de-redundant fiber bundle set are: the distance threshold is set to 5, the length threshold is set to 5, the minimum number of fiber bundles in the cluster is 20, and the number of cycles is 5.

[0055] Step (4) is to calculate the indicators of the correct fiber bundles after clearance. Four indicators are used to measure brain plasticity, namely fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD). The tensor model is used to calculate the eigenvalues ​​λ1, λ2 and λ3 of the tensor within the voxel. The calculation formulas and meanings of the four indicators are as follows:

[0056] The fractional anisotropy index (FA) is the ratio of the anisotropic component of water molecules to the entire diffusion tensor, and its range of variation is 0 to 1. 0 represents unrestricted diffusion, such as the FA value of cerebrospinal fluid close to 0; for very regular and directional tissues, the FA value is greater than 0, such as the FA value of brain white matter fibers close to 1.

[0057]

[0058] To comprehensively assess the diffusion state of a tissue voxel or region, the mean diffusivity (MD) must eliminate the influence of anisotropic diffusion and be expressed as a constant parameter, meaning that its variation is independent of the direction of diffusion. MD reflects the overall molecular diffusion level (the size of the mean ellipsoid) and the overall diffusion resistance. MD only indicates the magnitude of diffusion and is independent of the direction of diffusion. A larger MD indicates a greater number of free water molecules within the tissue.

[0059]

[0060] Axial Diffusivity (AD) represents the main diffusion direction and is defined as:

[0061] AD=λ1 (8);

[0062] Radial Diffusivity (RD) represents the mean of the other two secondary directions and is defined as:

[0063]

[0064] Step (5) is to repeat steps (2) to (4) for the two groups of data, namely, the images of patients with early mild cognitive impairment and the images of healthy people, and then perform a comparative analysis of the differences between the two groups of data. Each fiber in each key fiber bundle is resampled to 100 equidistant nodes to quantify the diffusion characteristics of the central part of the fiber bundle. Then, the four index values ​​of the two groups of subjects are analyzed using a one-way analysis of variance (ANOVA) in SPSS software, and an FDR correction is performed. A p<0.05 is considered to be statistically significant, and finally, a group of fiber bundles with significant differences is obtained.

[0065] Step (6) is to evaluate the inter-group difference fiber bundle group obtained in step (5) to obtain the evaluation reference method. The four index values ​​(FA, MD, etc.) of each key fiber bundle of each subject in each group are extracted to form four pairs of key fiber bundle index groups, namely group 1 (FA_MCI and FA_HC), group 2 (MD_MCI and MD_HC), group 3 (AD_MCI and AD_HC) and group 4 (RD_MCI and RD_HC). Each group is further divided into multiple groups according to all the extracted key fiber bundles. For example, group 1 can be divided into FA_MCI_fiber i ,i=1,2,...,m and FA_HC_fiber i ,i=1,2,...,m, where m is the number of fiber bundles with significant differences in step 5. Groups 1 to 4 contain the same number of subgroups, determined by the number of key fiber bundles. A receiver operating characteristic (ROC) curve was then constructed for all subgroups in the four groups. The horizontal axis of the curve represents the false positive rate (i.e., 1-specificity) and the vertical axis represents the sensitivity (true positive rate). The area under the ROC curve (AUC) for the four groups was then calculated, ranging from [0,1]. The fiber bundles and characteristic indicators corresponding to all subgroups with an AUC greater than 0.85 were used as the final evaluation reference.

[0066] The above content has provided a detailed introduction to the present invention, but the description of the specific implementation methods is only used to explain the method of the present invention and its core ideas, so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation methods. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concepts of the present invention are protected.

Claims

1. A method for assessing key fiber bundles in early mild cognitive impairment, characterized by: The following steps are involved: Step (1): preprocessing the diffusion tensor imaging images of patients with early mild cognitive impairment and healthy subjects, including data format conversion and data resampling; Step (2): using the preprocessing results of step (1) as input, complete the whole-brain fiber tract tracking; Step (3): Extract the tracked fiber bundles and remove the erroneous fiber bundles to obtain a preliminary whole-brain fiber bundle set, and then use the Mahalanobis distance formula to obtain the optimal de-redundant fiber bundle set; Step (4): Four indices, namely, partial anisotropy index, mean diffusivity, axial diffusivity, and radial diffusivity, were used to measure brain plasticity, and the indices were calculated for the cleared fiber bundles. Step (5): Repeat steps (2) to (4) for the two groups of image data of patients with early mild cognitive impairment and healthy subjects, and then perform inter-group comparison analysis on the indicators obtained from the two groups of data; Step (6): Evaluate the inter-group differential fiber bundle groups obtained in step (5), and calculate their ROC curve areas to obtain evaluation reference results; The step (3) specifically includes: Step (3.1) First, the whole-brain fiber bundle set is calculated to obtain the mean value, and then the fiber bundles with a value less than 10% and greater than 90% of the mean value are eliminated to obtain a preliminary whole-brain fiber bundle set; Step (3.2) uses the starting and ending regions of the fiber bundles in the MNI standard space as regions of interest (ROIs), and screens out fiber bundles that pass through both ROIs from the preliminary whole-brain fiber bundle set; Step (3.3) averages the binary masks of the relevant brain fiber bundle positions of multiple normal brain data in the standard coordinate system to obtain a probability map, wherein the probability map is the JHU white matter fiber bundle map. After comparison with the probability map, the fiber bundles passing through the low-probability position will be removed; In step (3.4), the Mahalanobis distance formula is used to remove fiber bundles. The specific parameters for obtaining the optimal de-redundant fiber bundle set are: the distance threshold is set to 5, the length threshold is set to 5, the minimum number of fiber bundles in the cluster is 20, and the number of cycles is 5.

2. The method for evaluating key fiber bundles in early mild cognitive impairment according to claim 1, wherein: The step (1) specifically includes: Step (1.1) Use dcm2niigui software to convert the original DICOM medical data format into .nii, bvec, and bval files as input files for fiber tract tracking. The .nii file contains image data, and the bvec and bval files contain the eigenvectors and eigenvalues ​​of the image data, respectively. Step (1.2) resamples the voxels of the original image to a size of 2×2×2mm 3 .

3. The method for evaluating key fiber bundles in early mild cognitive impairment according to claim 1, wherein: The step (2) specifically includes: using a deterministic tracking algorithm to complete whole-brain fiber bundle tracking, that is, given one or more starting points, searching for a new tracking direction according to a set calculation method, and continuously propagating forward until the fiber bundle is stopped from tracking when the termination condition is reached, and finally obtaining a fiber streamline trajectory; wherein the termination condition is set to the following six: (1) FA≤0.2 or angle≥45°; (2) FA≤0.25 or angle≥45°; (3) FA≤0.3 or angle≥45°; (4) FA≤0.2 or angle≥50°; (5) FA≤0.2 or angle≥55°; (6) FA≤0.2 or angle≥60°, wherein FA is anisotropy index and angle is fiber bundle bending angle; six deterministic whole-brain fiber bundle trackings are performed on the same individual according to the six termination conditions respectively, to obtain multiple groups of whole-brain fiber bundle sets.

4. The method for evaluating key fiber bundles in early mild cognitive impairment according to claim 1, wherein: The step (5) specifically includes: resampling each fiber in each fiber bundle set to 100 equidistant nodes to quantify the diffusion characteristics of the central part of the fiber bundle, and then using one-way analysis of variance in SPSS software to analyze the four index values ​​of the two groups of subjects, and performing FDR correction. When p<0.05, the difference is considered statistically significant, and finally a partial fiber bundle group with significant differences is obtained.

5. The method for evaluating key fiber bundles in early mild cognitive impairment according to claim 1, wherein: The step (6) specifically includes: extracting the four index values ​​of each fiber bundle group of each subject to form four pairs of key fiber bundle index groups, and then performing receiver operating characteristic curve (ROC) on the four groups, calculating the area under the ROC curve of the four groups, i.e., the AUC value, which ranges from [0, 1], and taking the fiber bundles and characteristic indicators corresponding to all groups with AUC values ​​greater than 0.85 as the final evaluation reference.

Images