Brain primary motion area grey matter layering method and system based on blood vessel space occupation

Abstract

The invention provides a brain primary motion area grey matter layering method and system based on vascular space occupancy, and the method comprises the steps: S1, scanning a BOLD image when a finger carries out the closing motion, and carrying out the positioning of a motion area based on the BOLD image, and obtaining an individualized motion area; s2, adjusting the position of the BOLD image based on the individualized motion area, and collecting a VASO image based on the adjusted image; s3, performing noise reduction on the VASO image by using a NORDIC noise reduction technology to obtain a noise-reduced image; step S4, preprocessing the noise-reduced image to obtain a preprocessed image, and step S5, performing up-sampling on the preprocessed image and calculating signal features, and layering the brain primary motion area based on the signal features.

Description

Technical Field

[0001] The present invention relates to the technical field of individualized brain motor area mapping stratification, and particularly relates to a method and system for gray matter stratification of the primary motor area of the brain based on vascular space occupancy. Background Art

[0002] The regions of the human cerebral cortex range from the superficial layer close to the cerebrospinal fluid to the deep layer close to the white matter, and are usually divided into six layers. Each layer is composed of specific types of neurons and nerve fibers and has unique functions. The stratified structure of gray matter provides an efficient tissue and information processing framework for the brain, enabling neurons in different regions and layers to cooperate to complete complex functions. This precise structure is the basis for the efficient operation of the brain. At the same time, the state activity of laminar flow can reveal the input and output activities of the cerebral cortex.

[0003] With the continuous development of medical imaging technology, functional magnetic resonance imaging (fMRI), as an advanced neuroimaging method, can non-invasively and non-destructively measure brain activity. Traditional fMRI imaging relies on the BOLD (blood oxygenation level-dependent) signal, but the traditional BOLD signal is limited by its spatial specificity at high resolutions because it often mainly focuses on large veins towards the meningal surface and is difficult to accurately reflect the actual activity of neurons in the cerebral cortex. To solve this problem, VASO imaging technology has emerged. VASO imaging has less deviation in shallow depths, provides more quantitative signals, has higher local specificity in the distribution of neural activities across cortical layers and cortical surface localization, and can more clearly observe the interlayer activities of cerebral gray matter.

[0004] However, VASO imaging also has many problems. First, to improve the signal-to-noise ratio of imaging, VASO imaging is usually limited by the number of layers and can only cover part of the brain area, and it is not necessarily possible to accurately locate the target area when collecting VASO images. Second, VASO images require a high signal-to-noise ratio. With the continuous improvement of high field strength and high spatio-temporal resolution, the acquisition of thermal noise also increases. Third, there are individual differences in brain structures among different subjects, and VASO is a single-slice analysis, making it difficult to accurately locate the individualized target area. Therefore, there is an urgent need for a method based on vascular space occupancy for gray matter stratification of the primary motor area of the brain to obtain the interlayer state of gray matter in the individualized brain motor area. Summary of the Invention

[0005] The present invention aims to solve the deficiencies of the prior art and provides a method based on vascular space occupancy for gray matter stratification of the primary motor area of the brain, including the following steps:

[0006] Step S1: Scan the BOLD images when the finger makes a closing movement and locate the position of the movement area based on the BOLD images to obtain an individualized motor area;

[0007] Step S2: Adjust the position of the BOLD images based on the individualized motor area, and acquire VASO images based on the adjusted images;

[0008] Step S3: Denoise the VASO images using the NORDIC denoising technique to obtain denoised images;

[0009] Step S4: Preprocess the denoised images to obtain preprocessed images;

[0010] Step S5: Upsample the preprocessed images and calculate signal features, and stratify the primary motor area of the brain based on the signal features.

[0011] Optionally, in step S1, the acquisition parameters of the BOLO images are: using single-shot GRE echo planar imaging (EPI) images, TR / TE = 3000 / 26.6 ms, flip angle = 90°, FOV = 100×50 mm 2 , matrix size = 128×64, resolution = 0.8×0.8×0.8 mm 3 , slice = 30 layers.

[0012] Optionally, in step S1, the BOLD images when the finger makes a closing movement specifically include:

[0013] Use the left index finger and left thumb to make closing and finger-opening movement forms, collect finger movements according to a preset tapping frequency, and collect the primary motor cortex images on both sides of the central sulcus of the right hemisphere while collecting the left finger movements.

[0014] Optionally, the content of using the NORDIC denoising technique to denoise the VASO images to obtain denoised images specifically includes:

[0015] Use MP-PCA in NORDIC to estimate the thermal noise level and g-factor map of the VASO images, and obtain a first processed image with uniform thermal noise based on the thermal noise level and g-factor map;

[0016] Select a local voxel fragment from the first processed image, and construct a Casorati matrix based on the local voxel fragment;

[0017] Perform principal component analysis on the Casorati matrix using singular value decomposition to obtain an analysis result;

[0018] Screen local voxel fragments based on the analysis results, average the local voxel fragments, and perform inverse normalization processing on the VASO image and the scaled g-factor submap to obtain a denoised image.

[0019] The present invention also discloses a gray matter stratification system for the primary motor area of the brain based on vascular space occupancy, and the system includes:

[0020] A data acquisition module, configured to scan the BOLD image when the finger makes a closing movement and locate the position of the movement area based on the BOLD image to obtain an individualized movement area;

[0021] A position adjustment module, configured to adjust the position of the BOLD image based on the individualized movement area and acquire a VASO image based on the adjusted image;

[0022] An image denoising module, configured to denoise the VASO image using the NORDIC denoising technology to obtain a denoised image;

[0023] A preprocessing module, configured to preprocess the denoised image to obtain a preprocessed image;

[0024] A stratification module, configured to upsample the preprocessed image and calculate signal features, and stratify the primary motor area of the brain based on the signal features.

[0025] Optionally, in the data acquisition module, the acquisition parameters of the BOLO image are: using a single-shot GRE echo planar imaging (EPI) image, TR / TE = 3000 / 26.6 ms, flip angle = 90°, FOV = 100×50 mm 2 , matrix size = 128×64, resolution = 0.8×0.8×0.8 mm 3 , slice = 30 layers.

[0026] Optionally, in the data acquisition module, the left index finger and the left thumb are used to make closing and finger-opening movement forms, the finger movement is collected according to a preset tapping frequency, and the primary motor cortex images on both sides of the central sulcus of the right hemisphere are collected while collecting the left finger movement.

[0027] Optionally, the image denoising module specifically includes:

[0028] Estimate the thermal noise level and g-factor map of the VASO image using MP-PCA in NORDIC, and obtain a first processed image with uniform thermal noise based on the thermal noise level and g-factor map;

[0029] Select a local voxel fragment from the first processed image and construct a Casorati matrix based on the local voxel fragment;

[0030] Perform principal component analysis on the Casorati matrix using singular value decomposition to obtain the analysis result;

[0031] Based on the analysis result, screen local voxel fragments, average the local voxel fragments, and perform inverse normalization processing on the VASO image and the scaled g-factor mapping subgraph to obtain a denoised image.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. Individualized stratification

[0034] Although there is a general functional localization pattern in the primary motor area of the cerebral cortex, there are subtle differences in the brain structure and functional localization of each individual in detail. By scanning the individualized motor area (such as the BOLD image during finger closing movement), the position of the motor area during finger closing can be adjusted according to the specific situation of each individual, so as to achieve more accurate individualized gray matter stratification of the primary motor area.

[0035] 2. Improve image quality and accuracy

[0036] During the acquisition of high-resolution VASO images, noise may exist, which affects the accuracy of activation analysis. The present invention adopts the NORDIC denoising technology to perform denoising processing on the VASO image, effectively removing the interference signals in the image. On the one hand, it can improve the image quality, and on the other hand, it can enhance the accuracy of subsequent stratification. Especially when extracting signals and gray matter stratification in small areas, the error can be significantly reduced.

[0037] 3. Combine multi-modal imaging (BOLD and VASO)

[0038] Traditional stratification methods may rely solely on a certain type of imaging technology, such as BOLD stratification or VASO stratification. Although the BOLD sequence can cover a wide range of brain regions, it is vulnerable to the influence of large veins and cannot accurately stratify. Due to the limitations of the imaging sequence, although VASO can overcome the influence of large veins, generally the field of view is small and the number of layers is small. To solve these problems, we adopt the combination of multi-modal imaging technology. BOLD provides wide brain region coverage and is suitable for the preliminary localization and preliminary analysis of brain regions. After localizing to the individualized activation region, the VASO sequence is used for experimental acquisition and stratification. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0040] Figure 1 This is an exemplary diagram of the finger closing experimental paradigm in the embodiments of the present invention;

[0041] Figure 2 This is a schematic diagram of the bold acquisition area and the individualized motor area in the embodiments of the present invention;

[0042] Figure 3 This is a schematic diagram of the VASO acquisition area in the embodiments of the present invention;

[0043] Figure 4 This is a visualization diagram in which the gray matter in the embodiments of the present invention is equally divided into 11 layers;

[0044] Figure 5 This is an exemplary diagram of the layering result applied in the embodiments of the present invention;

[0045] Figure 6 This is a method step diagram of the gray matter layering method for the primary motor area of the brain using the method based on vascular space occupancy in the embodiments of the present invention. Detailed implementation manners

[0046] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0047] Embodiment 1

[0048] A gray matter layering method for the primary motor area of the brain using a method based on vascular space occupancy, as Figure 6 shown, the method includes:

[0049] Step S1, scanning the BOLD (Blood Oxygen Level Dependent) image when the finger makes a closing movement and positioning the position of the movement area based on the BOLD image to obtain an individualized motor area.

[0050] Scanning the BOLD positioning image when the finger makes a closing movement block paradigm; the parameters of BOLD are as follows: using a single-shot GRE echo planar imaging (EPI) image, TR / TE = 3000 / 26.6 ms, flip angle = 90°, FOV = 100×50 mm 2 , matrix size = 128×64, resolution = 0.8×0.8×0.8 mm 3 , slice = 30 layers.

[0051] The task is as follows: using a block paradigm, the subject makes movements of closing the index finger and thumb and opening the finger (as Figure 1As shown). This task requires finger movements of muscle stretching and muscle flexion. The tapping frequency is self-defined, about once every 3 seconds. The tapping time is locked to the scanner trigger in units of 10 TRs (TR of BOLD = 3S), and visual cues about which finger to tap and for how long are included. Each tapping task runs for 10 TRs, and the rest runs for 10 TRs, repeating 3 times. The motor task is completed with the left hand while collecting BOLD images including the primary motor area of the right hemisphere, as Figure 2 shown. Specifically, the BOLD acquisition machine area is the area including the primary motor area of the right hemisphere. In any experiment of this study, the right hand is not used.

[0052] Based on the localization image of the scanned BOLD, analyze the activation area in a timely manner to locate the position L of the individualized motor area ROI ; time correction, motion correction, modeling and activation calculation, visualization of the activation area. It can be understood that the previous step serves as the input for the next step and is executed sequentially, and finally the position L of the individualized motor area is obtained ROI .

[0053] Time correction: The BOLD data in this example is scanned layer by layer in ascending order, and the acquisition time of each layer is slightly different. However, in subsequent analyses, it is assumed that all layers are acquired simultaneously, so time correction is required.

[0054] Head motion correction: To avoid errors in subsequent data analysis caused by the displacement of the subject's head movement during scanning, the BOLD images need to be rigidly registered to obtain the head motion parameter ε.

[0055] Modeling and activation calculation: Model according to the experimental paradigm of finger closing to obtain the design matrix x, and use the generalized linear model (GLM) of the following formula to analyze the relationship between voxels and the paradigm to obtain the β value, that is, the estimated effect size.

[0056] Y = Xβ + ε

[0057] where Y is the collected BOLD signal, X is the design matrix, β is the model parameter, and ε is the head motion error.

[0058] Visualization of the activation area: Set a threshold for the statistical significance level of the β value to screen out significantly activated voxels, as Figure 2 shown, to obtain the position L of the individualized motor area ROI .

[0059] Step S2: Adjust the position of the BOLD image based on the individualized motor area, and collect VASO images based on the adjusted image.

[0060] As Figure 3 shown, based on the individualized motor area LROI Adjust the position of the acquired image. The imaging slice needs to enclose this area and be perpendicular to the surface of the right hemisphere motor cortex. The VASO acquisition parameters are as follows: The VASO sequence is implemented based on 2D-EPI. Use a 5T scanner to obtain concurrent measurements of thin-slice selective slab inversion of VASO and BOLD signals. The parameters are: TR pair / TE = 4700 / 27 ms, TI-1 / TI-2 = 1300 / 3700 ms, resolution = 0.8×0.8×1.5 mm 3 , GRAPPA = 3, slices = 5 layers.

[0061] The task is as follows: Adopt a block paradigm. The subjects perform movements of closing the index finger and thumb and opening the fingers (as Figure 1 shown), repeating 10 times (TR of VASO = 4.7 s).

[0062] Step S3, use the NORDIC noise reduction technology to reduce the noise of the VASO image and obtain a denoised image.

[0063] Based on the acquired VASO image, use the NORDIC noise reduction technology to improve the quality of the image data. The goal of NORDIC (Noise Reduction with Distribution Corrected) is to focus on removing components in the time series that cannot be distinguished from Gaussian distributed noise estimation by algorithms, while retaining non-white noise sources such as physiological effects, signal drift, or head movement, as well as the signal of interest.

[0064] In NORDIC noise reduction, use MP-PCA (Pastur principal component analysis), that is, mixed probability principal component analysis to estimate the thermal noise level and the g-factor map. Then use the g-factor map to normalize the data and scale it according to the estimated thermal noise level to ensure uniform thermal noise throughout the image.

[0065] Select a local voxel patch and construct a Casorati matrix Y m×n , where m is the number of row vectors and n represents the time series of each voxel within the local voxel patch.

[0066] Run principal component analysis on the Casorati matrix using singular value decomposition (SVD). SVD decomposes Y m×n into three new matrices, such as Y m×n = USV. Where V contains the principal components (i.e., the patterns in the time series that best explain the data variation), S is the singular value, representing the variance size explained by each principal component, and U is the loading matrix, representing the weight of each voxel on these principal components.

[0067] Identifying the singular value threshold: The core idea of NORDIC is to identify which principal components reflect thermal noise and which reflect the signal of interest. The signals of interest usually appear redundantly in multiple voxels, so they will have larger singular values, while thermal noise has strong correlation only in a single voxel, manifested as smaller singular values. NORDIC determines a threshold through Monte Carlo Simulation to remove the principal components corresponding to small singular values.

[0068] Finally, NORDIC averages the denoised local voxel fragments and performs inverse normalization on the image with the scaled g-factor map.

[0069] Step S4: Preprocess the denoised image to obtain a preprocessed image.

[0070] Perform data preprocessing on the denoised image, mainly including motion correction, removing interference outside the scanning area through manual masking, and replacing non-steady-state images to improve data stability; eliminating the signal interference of large blood vessels by dividing the blood-inactivated image by the non-inactivated BOLD image, and aligning the time series through upsampling to ensure the time consistency of the image; calculating the signal change of the preprocessed data;

[0071] Head motion correction: 1) To avoid motion estimation error caused by variable distortion outside the field of view, first define a ROI area that needs to include the entire L ROI . 2) The SS-SI (Slice-selective slab-inversion) VASO sequence acquires interleaved images with and without blood contamination. Therefore, correct both images. 3) Ideally, the motion parameters of BOLD and VASO should be very similar. After head motion correction, it is necessary to check whether the motion parameters are too large. 4) Perform quality assessment on the imaging data, calculate quality indicators such as signal-to-noise ratio (tSNR), mean, skewness, and kurtosis to ensure the stability of the data processing effect.

[0072] BOLD contrast correction: The VASO signal reflects the change in cerebral blood volume (CBV), but does not directly measure the absolute CBV. Instead, it indirectly estimates the relative change in CBV through inversion preparation and signal normalization.

[0073] There is a T2* weight in any EPI reading. BOLD contamination will cause uneven signal distribution on the surface and in the deep layer, thus affecting the hierarchical resolution of VASO. After BOLD correction, the resolution of VASO data on different cortical layers is more uniform, and it can more accurately reflect the real activities of the brain hierarchy. Therefore, dynamic segmentation of BOLD decay is used for correction.

[0074] 1) This step requires separating the blood - contaminated images from the non - contaminated images.

[0075] 2) Since the BOLD and VASO images are acquired interleaved, time - series upsampling and relative shifting are needed to make the respective contrasts refer to the same time point.

[0076] 3) Partitioning is performed in the LAYNII program LN_BOCO. First, the trials are averaged, and then BOLD correction is carried out.

[0077] The gradient - echo signal S of a voxel can be expressed as:

[0078]

[0079] where M z represents the Z magnetization within the voxel at the excitation moment, which depends on the relative proton density within the voxel, the inversion time, and the relative volume distribution of the components (such as gray matter, white matter, blood, etc.) within the voxel. T 2 * is the transverse relaxation time in the gradient - echo acquisition, reflecting the attenuation characteristics of the signal. TE is the echo time.

[0080] Assume that a voxel contains only blood and gray matter (no white matter and cerebrospinal fluid), and under a specific "signal - nulling condition" (nc), only gray matter generates a signal. The signal from gray matter can be expressed as:

[0081]

[0082] where M z,GM,nc is the magnetization of gray matter under the nulling condition. is the transverse relaxation time of gray matter.

[0083] VASO is based on the following assumptions: The M z,GM,nc of gray matter changes during brain activity, and this change is proportional to 1 - CBV. Since the transverse relaxation time changes with brain activity, VASO normalizes the "signal under the nulling condition" (S nc ) to the "signal when blood is not nulled" (S nn ) to eliminate the influence caused by the change of . Therefore, the signal when blood is not nulled can be expressed as:

[0084]

[0085] where M z,par,nn represents the sum of the z - magnetizations of blood and brain parenchyma (gray matter). Represents the transverse relaxation time of the brain parenchyma.

[0086] When acquiring images with non-zeroed blood, the z magnetization intensities of blood and gray matter, as well as their proton densities, are very similar in this state. Under this condition, the value of M z,par,nn is not affected by brain activity and can be regarded as constant. Assuming M z，par，nn ≈ const, the relationship between the BOLD signal and relative CBV is finally obtained:

[0087]

[0088] Calculate the signal change of this task: To avoid biases in the variable noise magnitude and hemodynamic response function at different cortical depths, deconvolution or inferential statistical models are not used to measure activation. VASO is a quantitative measurement that is proportional to physical units (ml per 100ml), which means the units can be directly interpreted without the need to convert to percentage signal change. The signal change is calculated as follows:

[0089] First, for ease of explanation, the VASO signal v of each subject at each time point t t is converted from negative contrast to positive contrast as follows:

[0090] V t = V t * (-100)

[0091] Subsequently, calculate the average signal of 10 runs, with the resting state denoted as the task state denoted as To ensure that the activation or resting state has reached a stable state, we discard the first two and the last two time points and calculate the average of the resting state and the average of the task state

[0092]

[0093]

[0094] Finally, calculate the signal transformation

[0095]

[0096] Step S5: Upsample the preprocessed image and calculate signal features, and stratify the primary motor area of the brain based on the signal features.

[0097] Generate a smooth boundary through upsampling based on the processed data and calculate the signal features of each layer, and stratify the L ROI region;

[0098] 1) Upsampling: For the 0.7 - 0.8 mm resolution dataset, upsample by 4 - 5 times.

[0099] 2) Manually demarcate the boundary of gray matter: In the finger tapping task, an obvious activation area can be seen in the motor cortex. We select the position L of the individualized motor area ROI , and manually draw the gray matter - white matter boundary and gray matter - cerebrospinal fluid boundary of L ROI .

[0100] 3) Calculate the cortical depth of ROI: Use LAYNII to calculate 11 equally deep layers, as Figure 4 shown. Finally, extract the activation data based on the calculated cortical depth:

[0101] 1) Obtain the voxel value of each in the 11 ROIs to extract the activity values of different layers.

[0102] 2) Obtain the mean value, standard deviation, and number of voxels of each layer and visualize them, as Figure 5 shown.

[0103] Example 2

[0104] A method based on vascular space occupancy for the gray matter stratification system of the primary motor area of the brain, specifically:

[0105] A data acquisition module, configured to scan the BOLD image when the finger makes a closing movement and locate the position of the movement area based on the BOLD image to obtain an individualized motor area.

[0106] Scan the BOLD localization image during the finger closing movement block paradigm; the parameters of BOLD are as follows: use a single - shot GRE echo - planar imaging (EPI) image, TR / TE = 3000 / 26.6 ms, flip angle = 90°, FOV = 100×50 mm 2 , matrix size = 128×64, resolution = 0.8×0.8×0.8 mm 3 , slice = 30 layers.

[0107] The task is as follows: Adopt the block paradigm, and the subjects perform movements of closing the index finger and thumb and opening the fingers (such as Figure 1As shown). This task requires finger movements of muscle stretching and muscle flexion. The tapping frequency is self-defined, about once every 3 seconds. The tapping time is locked to the scanner trigger in units of 10 TRs (TR of BOLD = 3S), and visual cues including which finger to tap and how long to tap are included. Each tapping task runs for 10 TRs, and the rest runs for 10 TRs, repeating 3 times. The motor task is completed with the left hand while collecting BOLD images including the primary motor area of the right hemisphere, as Figure 2 shown. In any experiment of this study, the right hand was not used.

[0108] Based on the localization image of the scanned BOLD, analyze the activation area in a timely manner to locate the position L of the individualized motor area ROI ; time correction, motion correction, modeling and activation calculation, visualization of the activation area. It can be understood that the previous step is used as the input of the next step and is executed in sequence to finally obtain the position L of the individualized motor area ROI .

[0109] Time correction: The BOLD data in this example is scanned layer by layer in ascending order, and the acquisition time of each layer is slightly different. However, in subsequent analyses, it is assumed that all layers are acquired simultaneously, so time correction is required.

[0110] Head motion correction: To avoid errors in subsequent data analysis caused by the displacement of the subject's head movement during scanning, the BOLD images need to be rigidly registered to obtain the head motion parameter ε.

[0111] Modeling and activation calculation: Model according to the experimental paradigm of finger closing to obtain the design matrix x, and use the generalized linear model (GLM) of the following formula to analyze the relationship between voxels and the paradigm to obtain the β value, that is, the estimated effect size.

[0112] Y = Xβ + ε

[0113] where Y is the collected BOLD signal, X is the design matrix, β is the model parameter, and ε is the head motion error.

[0114] Visualization of the activation area: Set the threshold of the statistical significance level for the β value to screen out significantly activated voxels, as Figure 2 shown, to obtain the position L of the individualized motor area ROI .

[0115] A position adjustment module for adjusting the position of the BOLD image based on the individualized motor area and collecting VASO images based on the adjusted image.

[0116] Based on the individualized motor area L ROIAdjust the position of the acquired image so that the imaging slice wraps around this area and is perpendicular to the surface of the right hemisphere motor cortex; the VASO acquisition parameters are as follows: The VASO sequence is implemented based on 2D-EPI. Use a 5T scanner to obtain concurrent measurements of thin-slice selective slab inversion of VASO and BOLD signals, with parameters: TR pair / TE = 4700 / 27 ms, TI-1 / TI-2 = 1300 / 3700 ms, resolution = 0.8×0.8×1.5 mm 3 , GRAPPA = 3, slices = 5 layers.

[0117] The task is as follows: Adopt a block paradigm, and the subjects perform movements of closing the index finger and thumb and opening the fingers (as Figure 1 shown), repeating 10 times (TR of VASO = 4.7 s).

[0118] An image denoising module, which is used to denoise the VASO image using NORDIC denoising technology to obtain a denoised image.

[0119] Based on the acquired VASO image, use NORDIC denoising technology to improve the quality of the image data; the goal of NORDIC (Noise Reduction with Distribution Corrected) is to focus on removing components in the time series that cannot be distinguished from Gaussian distributed noise estimation by algorithms, while retaining non-white noise sources such as physiological effects, signal drift, or head movement, as well as the signal of interest.

[0120] In NORDIC denoising, use MP-PCA (Pastur principal component analysis), that is, mixed probability principal component analysis to estimate the thermal noise level and the g-factor map. Then use the g-factor map to normalize the data and scale it according to the estimated thermal noise level to ensure uniform thermal noise throughout the image.

[0121] Select a local voxel patch and construct a Casorati matrix Y m×n , where there are m row vectors, and n represents the time series of each voxel within the local voxel patch.

[0122] Run principal component analysis on the Casorati matrix using singular value decomposition (SVD). SVD decomposes Y m×n into three new matrices, such as Y m×n = USV. Where V contains the principal components (i.e., the patterns in the time series that best explain the data variation), S is the singular value, representing the variance size explained by each principal component, and U is the loading matrix, representing the weights of each voxel on these principal components.

[0123] Identifying the singular value threshold: The core idea of NORDIC is to identify which principal components reflect thermal noise and which reflect the signals of interest. Signals of interest usually redundantly appear in multiple voxels, so they will have larger singular values, while thermal noise only has strong correlation in a single voxel, showing smaller singular values. NORDIC determines a threshold through Monte Carlo Simulation to remove the principal components corresponding to small singular values.

[0124] Finally, NORDIC averages the denoised local voxel fragments and performs inverse normalization on the image with the scaled g-factor map.

[0125] Preprocessing module, used to preprocess the denoised image to obtain a preprocessed image.

[0126] Perform data preprocessing on the denoised image, mainly including motion correction. Remove interference outside the scanning area through manual masking, and at the same time replace non-steady-state images to improve data stability; eliminate the signal interference of large blood vessels by dividing the blood-inactivated image by the non-inactivated BOLD image, and align the time series through upsampling to ensure the time consistency of the image; calculate the signal change of the preprocessed data;

[0127] Head motion correction: 1) To avoid motion estimation errors caused by variable distortions outside the field of view, first define an ROI area that needs to include the entire L ROI . 2) The SS-SI (Slice-selective slab-inversion) VASO sequence acquires interleaved images with and without blood contamination. Therefore, correct both images. 3) Ideally, the motion parameters of BOLD and VASO should be very similar. After head motion correction, it is necessary to check whether the motion parameters are too large. 4) Perform quality assessment on the imaging data, calculate quality indicators such as signal-to-noise ratio (tSNR), mean, skewness, and kurtosis to ensure the stability of the data processing effect.

[0128] BOLD contrast correction: The VASO signal reflects the change in cerebral blood volume (CBV), but does not directly measure the absolute CBV. Instead, it indirectly estimates the relative change in CBV through inversion preparation and signal normalization.

[0129] There is a T2* weight in any EPI reading. BOLD contamination will cause uneven signal distribution on the surface and in the deep layer, thus affecting the hierarchical resolution of VASO. After BOLD correction, the resolution of VASO data on different cortical levels is more uniform, and it can more accurately reflect the true activities of the brain hierarchy. Therefore, dynamic segmentation of BOLD decay is used for correction.

[0130] 1) This step requires separating the blood - contaminated images from the non - contaminated images.

[0131] 2) Since BOLD and VASO images are acquired alternately, time - series upsampling and relative shifting are needed to make the respective contrasts refer to the same time point.

[0132] 3) Division is performed in the LAYNII program LN_BOCO. First, the trials are averaged, and then BOLD correction is carried out.

[0133] The gradient - echo signal S of a voxel can be expressed as:

[0134]

[0135] where M z represents the Z - magnetization intensity within the voxel at the excitation moment, which depends on the relative proton density within the voxel, the inversion time, and the relative volume distribution of each component (such as gray matter, white matter, blood, etc.) within the voxel. is the transverse relaxation time in the gradient - echo acquisition, reflecting the attenuation characteristics of the signal. TE is the echo time.

[0136] Assume that a certain voxel only contains blood and gray matter (no white matter and cerebrospinal fluid), and under a specific "signal - nulling condition" (nc), only gray matter generates a signal. The signal from gray matter can be expressed as:

[0137]

[0138] where M z,GM,nc is the magnetization intensity of gray matter under the nulling condition. is the transverse relaxation time of gray matter.

[0139] VASO is based on the following assumptions: The M z,GM,nc of gray matter changes during brain activity, and this change is proportional to 1 - CBV. Since the transverse relaxation time changes with brain activity, VASO eliminates the influence caused by the change of nc by normalizing the "signal under the nulling condition" (S nn ). Therefore, the signal when blood is not nulled can be expressed as:

[0140]

[0141] where M z,par,nn represents the sum of the z - magnetization intensities of blood and brain parenchyma (gray matter). represents the transverse relaxation time of brain parenchyma.

[0142] When acquiring an image where the blood has not returned to zero, the z magnetization intensities of the blood and gray matter, as well as their proton densities, are very similar in this state. Under such conditions, the value of M z,par,nn is not affected by brain activity and can be regarded as constant. We assume M z，par，nn ≈ const, and finally obtain the relationship between the BOLD signal and relative CBV:

[0143]

[0144] Calculate the signal change for this task: To avoid biases in the variable noise magnitude and hemodynamic response function at different cortical depths, deconvolution or inferential statistical models are not used to measure activation. VASO is a quantitative measurement that is proportional to physical units (ml per 100 ml), which means the units can be directly interpreted without the need to convert to percentage signal change. The signal change is calculated as follows:

[0145] First, for ease of interpretation, we convert the VASO signal v t for each subject at each time point t from negative contrast to positive contrast as follows:

[0146] V t = V t * (-100)

[0147] Subsequently, calculate the average signal for 10 runs, with the resting state denoted as and the task state denoted as To ensure that the activation or resting state has reached a stable state, we discard the first two and the last two time points and calculate the average value of the resting state and the average value of the task state

[0148]

[0149]

[0150] Finally, calculate the signal transformation

[0151]

[0152] A hierarchical module for upsampling the preprocessed image and calculating signal features, and hierarchically segmenting the primary motor area of the brain based on the signal features.

[0153] Generate a smooth boundary through upsampling based on the processed data and calculate the signal features of each layer, and hierarchically segment the L ROI region;

[0154] 1) Upsampling: For a dataset with a resolution of 0.7 - 0.8 mm, upsample by 4 - 5 times.

[0155] 2) Manually delineate the boundaries of gray matter: In the finger tapping task, obvious activation regions can be seen in the motor cortex. We select the position L of the individualized motor area ROI , and manually draw the gray matter - white matter boundary and gray matter - cerebrospinal fluid boundary of L ROI .

[0156] 3) Calculate the cortical depth of the ROI: Use LAYNII to calculate 11 equi - depth layers.

[0157] Finally, extract the activation data based on the calculated cortical depth:

[0158] 1) Obtain the voxel value of each in the 11 ROIs to extract the activity values of different layers.

[0159] 2) Obtain the mean, standard deviation, and number of voxels of each layer and visualize them.

[0160] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A gray matter stratification method for primary motor areas of the brain based on vascular space occupancy, characterized in that: Methods include: Step S1, scanning a BOLD image when the finger performs a closing movement and locating the position of the movement area based on the BOLD image to obtain an individualized movement area; Step S2, adjusting the position of the BOLD image based on the individualized motor area, and acquiring a VASO image based on the adjusted image; Step S3, using NORDIC noise reduction technology to reduce noise on the VASO image to obtain a noise-reduced image; Step S4, preprocessing the denoised image to obtain a preprocessed image; Step S5, up-sampling the pre-processed image and calculating signal features, and stratifying the primary motor areas of the brain based on the signal features.

2. The gray matter stratification method of the primary motor area of ​​the brain based on vascular space occupancy according to claim 1, characterized in that: In step S1, the acquisition parameters of BOLO images are: single GRE echo planar imaging, TR / TE = 3000 / 26.6ms, flip angle = 90°, FOV = 100×50mm 2 , matrix size = 128 × 64, resolution = 0.8 × 0.8 × 0.8 mm 3 , slices = 30 layers.

3. The gray matter stratification method of the primary motor area of ​​the brain based on vascular space occupancy according to claim 1, characterized in that: In step S1, scanning the BOLD image when the finger performs a closing movement specifically includes: The left index finger and left thumb are used to make closing and opening movements, and the finger movements are collected according to the preset tapping frequency. While collecting the left hand finger movements, the primary motor cortex images on both sides of the central sulcus of the right hemisphere are also collected.

4. The gray matter stratification method of the primary motor area of ​​the brain based on vascular space occupancy according to claim 1, characterized in that: The VASO image is subjected to noise reduction using the NORDIC noise reduction technology, and the content of the noise reduction image obtained specifically includes: Using MP-PCA in NORDIC to estimate the thermal noise level and g-factor map of the VASO image, and obtaining a first processed image with uniform thermal noise based on the thermal noise level and g-factor map; Selecting a local voxel fragment from the first processed image, and constructing a Casorati matrix based on the local voxel fragment; Performing principal component analysis on the Casorati matrix using singular value decomposition to obtain analysis results; Based on the analysis results, local voxel fragments are screened, the local voxel fragments are averaged, and the VAS O image and the scaled g-factor mapping sub-image are reversely normalized to obtain a denoised image.

5. A gray matter stratification system of the primary motor area of ​​the brain based on vascular space occupancy, the system being used to implement the gray matter stratification method of the primary motor area of ​​the brain as claimed in any one of claims 1 to 4, characterized in that: The system includes: A data acquisition module, used for scanning the BOLD image when the finger performs a closing movement and locating the position of the movement area based on the BOLD image to obtain an individualized movement area; A position adjustment module, used for adjusting the position of the BOLD image based on the individualized motion area, and acquiring a VASO image based on the adjusted image; An image noise reduction module, used for reducing noise on the VASO image using NORDIC noise reduction technology to obtain a noise-reduced image; A preprocessing module, used for preprocessing the denoised image to obtain a preprocessed image; A stratification module is used to upsample the preprocessed image and calculate signal features, and stratify the primary motor area of ​​the brain based on the signal features.

6. The gray matter stratification system of the primary motor area of ​​the brain based on vascular space occupancy according to claim 5, characterized in that: In the data acquisition module, the acquisition parameters of BOLO images are: using a single GRE echo planar imaging image, TR / TE=3000 / 26.6ms, flip angle=90°, FOV=100×50mm 2 , matrix size = 128 × 64, resolution = 0.8 × 0.8 × 0.8 mm 3 , slices = 30 layers.

7. The gray matter stratification system of the primary motor area of ​​the brain based on vascular space occupancy according to claim 5, characterized in that: In the data acquisition module, the left index finger and the left thumb are used to make closing and opening movements, and the finger movements are collected according to a preset tapping frequency. While collecting the left hand finger movements, the primary motor cortex images on both sides of the central sulcus of the right hemisphere are collected.

8. The gray matter stratification system of the primary motor area of ​​the brain based on vascular space occupancy according to claim 6, characterized in that: The image noise reduction module specifically includes: Using MP-PCA in NORDIC to estimate the thermal noise level and g-factor map of the VASO image, and obtaining a first processed image with uniform thermal noise based on the thermal noise level and g-factor map; Selecting a local voxel fragment from the first processed image, and constructing a Casorati matrix based on the local voxel fragment; Performing principal component analysis on the Casorati matrix using singular value decomposition to obtain analysis results; Based on the analysis results, local voxel fragments are screened, the local voxel fragments are averaged, and the VAS O image and the scaled g-factor mapping sub-image are reversely normalized to obtain a denoised image.

Images