Quality control method and device for resting-state magnetic resonance cerebrovascular reactive imaging

Abstract

The invention discloses a quality control method for resting-state magnetic resonance cerebrovascular reactive imaging. The quality control method comprises the following steps: firstly, acquiring a resting-state magnetic resonance cerebrovascular reactivity image queue which needs to be subjected to quality control; then, quality control parameters (DI, Pn and Zscore) of the resting-state magnetic resonance cerebrovascular reactivity image are quantified; and finally, substituting the obtained quality control parameters into a quality evaluation model to obtain a comprehensive index QEI for evaluating image quality, selecting a proper QEI cutoff value according to the distribution condition of the resting-state magnetic resonance cerebrovascular reactivity images QEI in the queue and clinical analysis requirements, and rejecting the images of which the QEI is lower than the cutoff value to realize quality control. According to the method, the quality of the resting-state magnetic resonance cerebrovascular reactivity image can be automatically quantified, so that objective quality control is provided for large-scale clinical data analysis and clinical diagnosis based on the resting-state magnetic resonance cerebrovascular reactivity image, and clinical transformation of a resting-state magnetic resonance cerebrovascular reactivity imaging technology is facilitated.

Description

Technical Field

[0001] The invention relates to the field of magnetic resonance imaging and image processing, and in particular to a quality control method for resting magnetic resonance cerebrovascular reactivity imaging. Background Art

[0002] Resting MRI cerebrovascular reactivity (CVR) imaging is an emerging MRI technique that uses the spontaneous fluctuation of blood CO2 content during breathing as an intrinsic vasoactive stimulus and simultaneously collects BOLD signals to monitor the corresponding changes in cerebral hemodynamics, which can detect vascular health non-invasively and sensitively. Compared with traditional CVR imaging techniques based on CO2 inhalation, resting CVR imaging has low compliance requirements for subjects, high comfort, wide applicability, and is easy to obtain in a standard clinical environment. It has advantages as an imaging marker of cerebrovascular reserve in large-scale clinical data analysis based on public databases such as ADNI, UKB, and HCP, and can be combined with other clinical biochemical indicators and neuropsychological assessments to conduct research on CVR-related diseases. Resting CVR imaging technology has the potential to provide important information for the diagnosis and treatment evaluation of cerebrovascular diseases such as acute ischemic stroke, neurodegenerative diseases such as Alzheimer's disease, systemic diseases, and even psychiatric diseases. However, due to the limited degree of spontaneous fluctuation of blood CO2 content during breathing, resting CVR images are easily affected by background physiological noise, with poor signal-to-noise ratio and stability. Poor image quality will have an adverse effect on clinical analysis and diagnosis, and it is necessary to exclude poor quality CVR images from the analysis. Resting CVR image quality control based on visual inspection is subjective, especially in large-scale clinical data analysis, which requires high manpower and time costs. Therefore, a set of automatic quality control algorithms for resting CVR imaging is needed to objectively evaluate the quality of resting CVR images and better play the value of resting CVR imaging technology in standard clinical environments and large-scale clinical data analysis. Summary of the invention

[0003] Currently, there is a lack of automatic quality control algorithms for resting-state CVR imaging. To solve the problems described in the background technology, the present invention has developed a quality control method for resting-state CVR imaging. Combining the characteristics of CVR images with poor quality, three quality control parameters are extracted from the resting-state CVR images themselves, and the three quality control parameters are introduced into the quality assessment model to obtain the comprehensive index QEI for assessing image quality. The method can automatically quantify the reliability and authenticity of resting-state CVR images in reflecting cerebrovascular reserve, thereby achieving objective quality control and contributing to the clinical transformation of resting-state CVR imaging technology.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] In a first aspect, the present invention provides a quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging, which comprises the following steps:

[0006] S1. Obtain a cohort of resting-state MRI cerebrovascular reactivity images that require quality control, wherein the resting-state MRI cerebrovascular reactivity images are obtained by preprocessing the resting-state BOLD image time series and then performing voxel-by-voxel general linear regression analysis;

[0007] S2. For each resting MRI cerebrovascular reactivity image in the cohort, three quality control parameters are quantitatively calculated. The first quality control parameter is the comprehensive variance of the resting MRI cerebrovascular reactivity values ​​in the gray matter, white matter and cerebrospinal fluid regions. The second quality control parameter is the proportion of voxels with negative resting MRI cerebrovascular reactivity values ​​in the gray matter region. The third quality control parameter is the whole brain average Z score converted from the T value image generated by the general linear regression analysis.

[0008] S3, for each resting-state magnetic resonance cerebrovascular reactivity image in the cohort, inputting the corresponding three quality control parameters into the quality assessment model to obtain a comprehensive assessment index for each image;

[0009] S4. Using the comprehensive evaluation index as a positively correlated representation index of image quality, quality control is performed on all resting magnetic resonance cerebrovascular reactivity images in the queue.

[0010] As a preferred embodiment of the first aspect, in S1, the method for obtaining a resting magnetic resonance cerebrovascular reactivity image based on a resting BOLD image time series is as follows:

[0011] The resting-state BOLD image time series is subjected to head motion correction, spatial smoothing, linear detrending and time-domain filtering in sequence to obtain the BOLD signal time series of each voxel, and at the same time, the BOLD signal time series of all voxels in the whole-brain mask after removing the skull are spatially averaged to obtain a reference signal time series; the voxel-by-voxel BOLD signal time series is used as the dependent variable, the reference signal time series is used as the regression independent variable, and the six motion vectors and a linear trend generated by the head motion correction are used as covariates, and general linear regression analysis is performed voxel by voxel, and the regression coefficient corresponding to the regression independent variable of each voxel is used as the resting-state magnetic resonance cerebrovascular reactivity in relative units, thereby obtaining a resting-state magnetic resonance cerebrovascular reactivity map in relative units;

[0012] Preferably, the head motion correction, spatial smoothing, linear detrending, time domain filtering and general linear regression analysis are all performed by a statistical parametric mapping tool.

[0013] As a preferred embodiment of the first aspect, the spatial smoothing uses a Gaussian filter with a half-width of 8 mm, and the passband range of the time domain filtering is 0 to 0.1164 Hz.

[0014] As a preferred embodiment of the first aspect, the quantification method of the first quality control parameter is as follows:

[0015] The gray matter mask, white matter mask, and cerebrospinal fluid mask in T1 space were registered to BOLD space, and then the resting MRI cerebrovascular reactivity values ​​of the voxels in the mask were extracted using the three registered masks. Then, the comprehensive variance DI was calculated according to the following formula:

[0016]

[0017] Where: V k represents the variance of resting-state MRI cerebrovascular reactivity values ​​within mask k, m k represents the mean value of resting-state MRI cerebrovascular reactivity within mask k, N k represents the number of voxels in mask k. When k = 1, 2, and 3, mask k represents the gray matter mask, white matter mask, and cerebrospinal fluid mask, respectively.

[0018] As a preferred embodiment of the first aspect, the quantification method of the second quality control parameter is as follows:

[0019] The gray matter mask in T1 space was aligned to the BOLD space, and then the resting-state MRI cerebrovascular reactivity values ​​of the voxels in the mask were extracted using the aligned gray matter mask. The proportion Pn of the number of voxels with negative resting-state MRI cerebrovascular reactivity values ​​to the total number of voxels in the gray matter mask was obtained.

[0020] As a preferred embodiment of the first aspect, the quantification method of the third quality control parameter is as follows:

[0021] First, the SPM_T map generated by the current resting-state magnetic resonance cerebrovascular reactivity image in the general linear regression analysis is converted into a Z-score map, and then the mean Z-score of all voxels within the whole-brain mask range in the Z-score map is calculated, and the Z-score mean is standardized within the cohort to obtain the standardized whole-brain average Z-score Zscore.

[0022] As a preferred embodiment of the first aspect, the quality assessment model is in the form of:

[0023] QEI=k1+k2*DI+k3*Pn+k4*Zscore+k5*Pn*Zscore

[0024] Wherein: k1, k2, k3, k4, k5 are 5 fitting parameters, DI, Pn, Zscore are the first quality control parameter, the second quality control parameter, and the third quality control parameter, respectively.

[0025] Preferably, the quality assessment model is in the form of:

[0026] QEI=0.6890-0.0083*DI-2.4727*Pn+0.0265*Zscore-0.6329*Pn*Zscore.

[0027] As a preferred embodiment of the first aspect, when the comprehensive evaluation index QEI is used for quality control, the resting magnetic resonance cerebrovascular reactivity images in the queue with a comprehensive evaluation index QEI lower than the cutoff value preset by the comprehensive evaluation index QEI are eliminated as low-quality images.

[0028] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging as described in any one of the first aspects above is implemented.

[0029] In a third aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0030] The memory is used to store computer programs;

[0031] The processor is used to implement the quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging as described in any one of the first aspects above when executing the computer program.

[0032] Compared with the prior art, the present invention has the following technical effects:

[0033] The present invention proposes a quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging, which can automatically quantify the reliability and authenticity of resting-state CVR images in reflecting cerebrovascular reserve; compared with the existing resting-state CVR image quality control based on visual inspection, it is more objective and does not require professional imaging experts, especially in large-scale clinical data analysis, with low manpower and time costs; it provides objective quality control for clinical diagnosis and data analysis based on resting-state CVR, an imaging marker, reduces the adverse effects of poor-quality resting-state CVR, and contributes to the clinical transformation of resting-state CVR imaging technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a preprocessing flowchart for obtaining CVR images from resting-state BOLD images.

[0035] Figure 2 It is a quantitative flow chart of the quality control parameters DI, Pn, Zscore and the comprehensive quality evaluation index QEI of resting-state CVR images.

[0036] Figure 3 4 are typical resting-state CVR images of different qualities and QEI result diagrams obtained using the quality control method developed by the present invention.

[0037] Figure 4 It is a correlation result diagram and a specific example diagram of the test-retest stability verification of the quality control method developed by the present invention using the resting-state BOLD data of the same person scanned twice in the ADNI3 and HCP public databases. DETAILED DESCRIPTION

[0038] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0039] The present invention extracts specific quality control parameters from resting magnetic resonance cerebrovascular reactivity (CVR) imaging and calculates a comprehensive evaluation index QEI, thereby realizing automatic quality control. In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described below in conjunction with the accompanying drawings, and its specific technical effects will be demonstrated by examples.

[0040] In an embodiment of the present invention, a quality control method for resting state magnetic resonance cerebrovascular reactivity imaging is provided, which comprises the following steps:

[0041] Step 1: Obtain a resting-state magnetic resonance cerebrovascular reactivity image queue that needs to be quality controlled, wherein each resting-state magnetic resonance cerebrovascular reactivity image in the queue is obtained by performing voxel-by-voxel general linear regression analysis after resting-state BOLD image time series preprocessing, and the magnetic resonance imaging parameters of the resting-state BOLD image time series corresponding to all resting-state magnetic resonance cerebrovascular reactivity images in the queue are the same.

[0042] In addition, the resting state BOLD image time series is a sequence composed of a series of resting state BOLD images at multiple consecutive sampling moments. The acquisition of the resting state BOLD image time series belongs to the prior art, and the process of converting the resting state BOLD image time series into the resting state magnetic resonance cerebrovascular reactivity image can also be implemented by calling the tools in the prior art. In order to facilitate a better understanding by those skilled in the art, the principle of this conversion process in the present invention is described in detail below.

[0043] See also Figure 1As shown in the figure, the resting BOLD image time series is processed by motion correction, spatial smoothing, linear detrending and temporal filtering in sequence to obtain the BOLD signal time series of each voxel in the image domain. The spatial smoothing uses a Gaussian filter with a half-width of 8 mm, and the passband range of the temporal filtering is 0-0.1164 Hz to resist the background physiological noise and extract the component with the greatest correlation with the end-tidal CO2 concentration in the BOLD signal. The motion correction aligns the BOLD image of each sampling point to the BOLD image of the first sampling point to eliminate the displacement caused by the head movement. This process can derive the head motion vector (3 displacement vectors and 3 rotation vectors) and linear trend of each sampling point. In addition, the resting BOLD image time series after motion correction is processed by time averaging and skull removal to generate a whole-brain mask, which can be used to extract the voxel signal of brain tissue. After obtaining the whole-brain mask, the BOLD signal time series of all voxels in the whole-brain mask are spatially averaged to obtain the reference signal time series. Finally, for each voxel in the brain tissue, a general linear regression analysis was performed with its BOLD signal time series as the dependent variable, the reference signal time series as the regression independent variable, and the head motion vector sequence and a linear trend generated by the head motion correction as the covariate. The CVR value of the voxel was obtained. The whole-brain CVR value was obtained by performing general linear regression analysis on each voxel, and then divided by the mean CVR value of the whole brain to obtain the resting-state CVR map in relative units.

[0044] In an embodiment of the present invention, the above-mentioned head motion correction, spatial smoothing, linear detrending, time domain filtering and general linear regression analysis can all be implemented by a statistical parametric mapping tool SPM12 in MATLAB.

[0045] The dependent variable, regression independent variable and covariate in the above general linear regression analysis are all time domain signals in vector form, so this regression is a time domain general linear regression analysis. The form of the general linear regression model can be expressed as follows:

[0046] BOLD(t)=β1referencesignal+β2motion+β3t+β0

[0047] In the formula, BOLD(t) is the preprocessed voxel BOLD signal time series, reference signal is the reference signal time series obtained by spatially averaging the BOLD signal time series of all voxels in the whole brain mask, motion is the 6 head motion vectors generated by head motion correction, and t is the linear trend. β0, β1, β2, and β3 are all regression coefficients, which are given by Get the CVR value of the voxel.

[0048] It should be noted that in the above-mentioned time-domain general linear regression analysis, since all the signals involved in the regression are time-domain signals, it is actually equivalent to taking each parameter at each moment as a regression sample, and the regression samples at all moments as a sample set to fit the regression coefficients in the above-mentioned general linear regression model. Of course, this is only to facilitate understanding of the essence of the above-mentioned time-domain general linear regression analysis. In software such as MATLAB, the above-mentioned time-domain general linear regression analysis can be implemented by directly inputting in the form of vectors or matrices, and it is not necessary to pre-split the samples at each moment.

[0049] Step 2. For each resting-state MRI cerebrovascular reactivity image in the above-mentioned cohort, three quality control parameters are quantitatively calculated. The first quality control parameter is the comprehensive variance DI of the resting-state MRI cerebrovascular reactivity values ​​in the gray matter, white matter and cerebrospinal fluid regions. The second quality control parameter is the proportion Pn of voxels with negative resting-state MRI cerebrovascular reactivity values ​​in the gray matter region. The third quality control parameter is the whole-brain average Z score Zscore obtained by converting the T value image generated by the general linear regression analysis.

[0050] See also Figure 2 As shown, the process of extracting three quality control parameters DI, Pn, and Zscore from CVR images in an embodiment of the present invention is shown. The specific calculation methods of the three parameters are as follows:

[0051] 1) For the calculation of the comprehensive variance DI of the CVR of the three regions of gray matter (GM), white matter (WM), and cerebro-Spinal Fluid (CSF), CVR varies according to the tissue type. The maximum or minimum CVR value caused by noise increases the spatial variability of the same tissue, so DI can be used to reflect the noise level. In the embodiment of the present invention, the calculation method of the comprehensive variance DI of the CVR of the three regions is as follows: first use ANTs software to align the GM, WM, and CSF masks of the T1 space to the BOLD space, and then use the three registered masks to extract the resting state magnetic resonance cerebrovascular reactivity CVR values ​​of the voxels in the mask, and then calculate the comprehensive variance DI according to the following formula:

[0052]

[0053] Where: V k represents the variance of resting-state MRI cerebrovascular reactivity values ​​within mask k, m k represents the mean value of resting-state MRI cerebrovascular reactivity within mask k, N k represents the number of voxels in mask k. When subscript k = 1, 2, or 3, mask k represents the gray matter mask, white matter mask, and cerebrospinal fluid mask, respectively.

[0054] 2) Calculation of the proportion Pn of voxels with negative CVR values ​​in the gray matter region. CVR is the ability of blood vessels to expand when stimulated, and is usually positive in the gray matter region. The presence of a large number of voxels with negative CVR values ​​in the gray matter may indicate that the noise is large. In an embodiment of the present invention, the method for calculating the proportion Pn of voxels with negative CVR values ​​in the gray matter region is as follows: first use ANTs software to align the GM mask in the T1 space to the BOLD space, and then use the aligned gray matter mask to extract the resting magnetic resonance cerebrovascular reactivity values ​​of the voxels in the mask, and statistically obtain the proportion Pn of the number of voxels with negative resting magnetic resonance cerebrovascular reactivity values ​​to the total number of voxels in the gray matter mask.

[0055] 3) Since the resting-state CVR images are generated by general linear regression analysis based on the BOLD image time series, the whole-brain average Z score generated by this process can be used to evaluate its sensitivity. In the embodiment of the present invention, the calculation method of the whole-brain average Z score Zscore generated by the time-domain general linear regression analysis is as follows: first, the SPM_T map generated by the current resting-state magnetic resonance cerebrovascular reactivity image in the general linear regression analysis is converted into a Z score map. In the embodiment of the present invention, the "spm_t2z()" function of the SPM tool can be used to convert the SPM_T map generated by the general linear regression analysis into a Z score map; then calculate the Z score mean of all voxels within the whole-brain mask range in the Z score map, and perform intra-group standardization on the Z score mean in the resting-state magnetic resonance cerebrovascular reactivity image queue to obtain the standardized whole-brain average Z score Zscore.

[0056] It should be noted that the above-mentioned within-group standardization when calculating the average Z score of the whole brain is to convert the mean Z score Zscore corresponding to each CVR image in the queue into i As the observed value, the Z score was re-standardized and calculated across the entire cohort. The average of the Z scores corresponding to all CVR images in the entire cohort was Z mean , the standard deviation of the mean Z score corresponding to all CVR images is Z st , then the mean Z score Zscore corresponding to each CVR image i The average Z score of the whole brain after re-group standardization is Zscore = (Zscore i -Z mean ) / Z st .

[0057] Step 3: for each resting-state MRI cerebrovascular reactivity image in the queue, the corresponding three quality control parameters are input into the quality assessment model to obtain a comprehensive assessment index QEI for each image.

[0058] Since the three quality control parameters (DI, Pn, and Zscore) are obtained in step 2 above, which reflect the image quality in different dimensions, this step needs to integrate the three into a single metric QEI representing the resting-state CVR image quality, and substitute the quality control parameters DI, Pn, and Zscore into the quality assessment model to obtain the comprehensive indicator QEI for evaluating image quality.

[0059] In the present invention, the form of the above-mentioned quality assessment model is not limited, and the QEI that can best reflect the image quality shall be obtained. Since the respiratory controlled CVR imaging has an accuracy comparable to that of the traditional CVR imaging based on CO2 inhalation, the average structural similarity SSIM between the respiratory controlled CVR graph and the resting CVR graph is used as the gold standard (Gold Standard) for assessing the quality of the resting CVR image. Furthermore, in an embodiment of the present invention, a multivariate linear regression model is performed on the relationship between the three quality control parameters (DI, Pn, Zscore) and the quality assessment gold standard, and a total of 4 fitting models are tried: (1) "GoldStandard~1+DI+Pn+Zscore" containing only linear terms; (2) "Gold Standard~1+DI+Pn+Zscore+Pn*Zscore" containing linear terms and cross terms after stepwise regression; (3) "Gold Standard~1+DI+Pn+Pn" containing linear terms and square terms after stepwise regression. 2 "; (4) Perform univariate linear regression on the three quality assessment indicators and the quality assessment gold standard respectively, and then calculate the geometric mean of the three univariate linear regressions. Among the four fitting models, the model with the highest Pearson correlation coefficient between the fitted value QEI and the gold standard is "Gold Standard ~ 1 + DI + Pn + Zscore + Pn * Zscore" which contains linear terms and cross terms after stepwise regression.

[0060] In the present invention, the best model form among the above four fitting models is:

[0061] QEI=k1+k2*DI+k3*Pn+k4*Zscore+k5*Pn*Zscore

[0062] Wherein: k1, k2, k3, k4, k5 are five fitting parameters, DI, Pn, Zscore are three quality control parameters.

[0063] Each fitting parameter in the above quality assessment model can be obtained by fitting using an actual data set. In an embodiment of the present invention, the resting BOLD images of 17 subjects with three scanning parameters and the BOLD images with respiratory control were collected, and general linear regression analysis was performed to obtain 51 groups of resting CVR images and corresponding respiratory control CVR images. After fitting, a set of optimal parameters was obtained, as follows:

[0064] QEI=0.6890-0.0083*DI-2.4727*Pn+0.0265*Zscore-0.6329*Pn*Zscore

[0065] Therefore, the quality control parameters DI, Pn, and Zscore are substituted into the above model formula to obtain the comprehensive evaluation index QEI for evaluating image quality.

[0066] Step 4: Using the comprehensive evaluation index QEI as a positively correlated characterization index of CVR image quality, quality control is performed on all resting magnetic resonance cerebrovascular reactivity images in the queue.

[0067] It should be noted that the above-mentioned comprehensive evaluation index QEI reflects the quality of the CVR image. The higher the comprehensive evaluation index QEI, the better the quality of the CVR image. When the above-mentioned comprehensive evaluation index QEI is used for quality control, the specific quality control method can be set according to the actual application scenario requirements. One implementation method is to use a preset minimum cutoff value of the comprehensive evaluation index QEI, and remove the resting magnetic resonance cerebrovascular reactivity images in the queue whose comprehensive evaluation index QEI is lower than the cutoff value as low-quality images.

[0068] The minimum cutoff value of the above QEI can be set according to the density distribution of the QEI values ​​of all images in the cohort and the needs of clinical analysis, or it can be set in a manually assisted manner such as expert knowledge, for example, using the lower quartile or other quantile of the QEI of all CVR image samples as the cutoff value to filter the data.

[0069] In order to verify the effectiveness of the resting CVR image quality comprehensive evaluation index QEI proposed in the present invention for quality control, the quality control effect thereof is demonstrated below based on steps 1 to 4 of the above method in combination with verification experiments so that those skilled in the art can better understand the essence of the present invention.

[0070] Verification experiment 1:

[0071] In order to qualitatively verify the consistency between the resting-state CVR image quality comprehensive evaluation index QEI obtained by the quality control method developed by the present invention and the visual inspection effect, we designed the following human experiment: T1-weighted MPRAGE scans and resting-state BOLD scans were performed on 23 healthy subjects using a 64-channel head coil. The resting-state BOLD scan was acquired using gradient echo planar imaging. The scanning parameters were as follows: the field of view size was 220×220mm 2 , the voxel size is 2.50×2.50×2.50mm 3 , the scanning matrix size is 88×88×64, TR / TE=607 / 32ms, 64 slices with a gap of 2.50mm in the axial direction, the phase encoding direction is P to A, the number of measurement time points is 976, and the MB acceleration factor is 8. Figure 3 As shown, the QEI calculated by the quality control method developed by the present invention for four typical resting-state CVR images of different qualities has good consistency with the visual inspection results.

[0072] Verification experiment 2:

[0073] In order to quantitatively verify the applicability of the quality control method for resting-state CVR imaging developed by the present invention in the resting-state BOLD data of the public database, we used two resting-state BOLD data of the same person scanned in ADNI3 about one year apart and two resting-state BOLD data of the same person scanned in HCP about half a year apart to verify the test-retest stability of the quality control method developed by the present invention. There will be no significant variability in the two CVR images scanned within one year of a healthy person, unless caused by noise. If the resting-state CVR images of the same person are both of high quality, the test-retest stability (the average structural similarity SSIM of the two test CVR images) will be higher, and if at least one CVR image is of poor quality, the test-retest stability will decrease. Therefore, if the quality control method developed by the present invention can effectively evaluate the quality of resting-state CVR images, the smaller value of the QEI of the CVR images obtained from the same person twice is significantly positively correlated with the SSIM of the two images.

[0074] In the test-retest stability verification 1, two resting-state BOLD data of the same person were selected from the data of the ADNI3 Axial MB rsfMRI (Eyes Open) scan sequence, which were scanned about one year apart, for a total of 24 groups. The resting-state BOLD scan parameters are as follows: the field of view size is 220×220mm 2 , the voxel size is 2.50×2.50×2.50mm 3The scanning matrix size is 88×88×64, TR / TE=607 / 32ms, 64 slices with a gap of 2.50mm in the axial direction, the phase encoding direction is P to A, the number of measurement time points is 976, and the MB acceleration factor is 8.

[0075] In the test-retest stability validation 2, 45 sets of resting-state BOLD data were selected from the HCP Young Adult cohort, which were scanned twice about half a year apart from the same person. The resting-state BOLD scan parameters were as follows: the field of view size was 208×208mm 2 , the voxel size is 2×2×2mm 3 The scanning matrix size is 104×104×72, TR / TE=720 / 33ms, 72 slices with a gap of 2mm in the axial direction, the phase encoding direction is P to A, the number of measurement time points is 1200, and the MB acceleration factor is 8.

[0076] See also Figure 4 (A) The Pearson correlation coefficient r of lower QEI in the CVR graphs of the same person tested twice was 0.8091, which was significantly correlated (p value was much less than 0.001). Figure 4 (B) The Pearson correlation coefficient r of lower QEI in the CVR graphs of the same person tested twice was 0.8901, which was significantly correlated (p value was much less than 0.001). Figure 4 (C) is a typical schematic diagram of three groups of test-retest stability verification. In the test-retest stability verification, the smaller QEI value of the CVR image obtained by two scans of the same person is significantly positively correlated with the SSIM of the two images, which together demonstrates that the quality control method for resting-state CVR imaging developed by the present invention is applicable to CVR images obtained from resting-state BOLD data of people of different ages and with multiple scanning parameters.

[0077] It should be pointed out that the above implementation modes are only preferred implementation modes of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A quality control method for resting state magnetic resonance cerebrovascular reactivity imaging, characterized in that: The following steps are involved: S1. Obtain a cohort of resting-state MRI cerebrovascular reactivity images that require quality control, wherein the resting-state MRI cerebrovascular reactivity images are obtained by preprocessing the resting-state BOLD image time series and then performing voxel-by-voxel general linear regression analysis; S2. For each resting MRI cerebrovascular reactivity image in the cohort, three quality control parameters are quantitatively calculated. The first quality control parameter is the comprehensive variance of the resting MRI cerebrovascular reactivity values ​​in the gray matter, white matter and cerebrospinal fluid regions. The second quality control parameter is the proportion of voxels with negative resting MRI cerebrovascular reactivity values ​​in the gray matter region. The third quality control parameter is the whole brain average Z score converted from the T value image generated by the general linear regression analysis. S3, for each resting-state magnetic resonance cerebrovascular reactivity image in the cohort, inputting the corresponding three quality control parameters into the quality assessment model to obtain a comprehensive assessment index for each image; S4. Using the comprehensive evaluation index as a positively correlated representation index of image quality, quality control is performed on all resting magnetic resonance cerebrovascular reactivity images in the queue.

2. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: In S1, the method for obtaining a resting-state magnetic resonance cerebrovascular reactivity image based on a resting-state BOLD image time series is as follows: The resting-state BOLD image time series is subjected to head motion correction, spatial smoothing, linear detrending and time-domain filtering in sequence to obtain the BOLD signal time series of each voxel, and at the same time, the BOLD signal time series of all voxels in the whole-brain mask after removing the skull are spatially averaged to obtain a reference signal time series; the voxel-by-voxel BOLD signal time series is used as the dependent variable, the reference signal time series is used as the regression independent variable, and the six motion vectors and a linear trend generated by the head motion correction are used as covariates, and general linear regression analysis is performed voxel by voxel, and the regression coefficient corresponding to the regression independent variable of each voxel is used as the resting-state magnetic resonance cerebrovascular reactivity in relative units, thereby obtaining a resting-state magnetic resonance cerebrovascular reactivity map in relative units; Preferably, the head motion correction, spatial smoothing, linear detrending, time domain filtering and general linear regression analysis are all performed by a statistical parametric mapping tool.

3. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: The spatial smoothing uses a Gaussian filter with a half-width of 8 mm, and the passband range of the time domain filtering is 0 to 0.1164 Hz.

4. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: The first quality control parameter is quantified as follows: The gray matter mask, white matter mask, and cerebrospinal fluid mask in T1 space were registered to BOLD space, and then the resting MRI cerebrovascular reactivity values ​​of the voxels in the mask were extracted using the three registered masks. Then, the comprehensive variance DI was calculated according to the following formula: Where: V k represents the variance of resting-state MRI cerebrovascular reactivity values ​​within mask k, m k represents the mean value of resting-state MRI cerebrovascular reactivity within mask k, N k represents the number of voxels in mask k. When k = 1, 2, and 3, mask k represents the gray matter mask, white matter mask, and cerebrospinal fluid mask, respectively.

5. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: The second quality control parameter is quantified as follows: The gray matter mask in T1 space was aligned to the BOLD space, and then the resting-state MRI cerebrovascular reactivity values ​​of the voxels in the mask were extracted using the aligned gray matter mask. The proportion Pn of the number of voxels with negative resting-state MRI cerebrovascular reactivity values ​​to the total number of voxels in the gray matter mask was obtained.

6. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: The third quality control parameter is quantified as follows: First, the SPM_T map generated by the current resting-state magnetic resonance cerebrovascular reactivity image in the general linear regression analysis is converted into a Z-score map, and then the mean Z-score of all voxels within the whole-brain mask range in the Z-score map is calculated, and the Z-score mean is standardized within the cohort to obtain the standardized whole-brain average Z-score Zscore.

7. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: The quality assessment model is in the form of: QEI=k1+k2*DI+k3*Pn+k4*Zscore+k5*Pn*Zscore Wherein: k1, k2, k3, k4, k5 are 5 fitting parameters, DI, Pn, Zscore are the first quality control parameter, the second quality control parameter, and the third quality control parameter, respectively; Preferably, the quality assessment model is in the form of: QEI=0.6890-0.0083*DI-2.4727*Pn+0.0265*Zscore-0.6329*Pn*Zscore.

8. The quality control method for resting state magnetic resonance cerebrovascular reactivity imaging according to claim 1, characterized in that: When quality control is performed using the comprehensive evaluation index QEI, the resting magnetic resonance cerebrovascular reactivity images in the queue whose comprehensive evaluation index QEI is lower than the cutoff value are eliminated as low-quality images according to the preset minimum cutoff value of the comprehensive evaluation index QEI.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging as claimed in any one of claims 1 to 8 is implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the quality control method for resting-state magnetic resonance cerebrovascular reactivity imaging as described in any one of claims 1 to 8 when executing the computer program.

Images