Multi-center magnetic resonance brain image correction method and system
By extracting physiological indicators from multicenter magnetic resonance imaging data and using the Image-Combat algorithm for standardization, the problem that ComBat cannot correct for nonlinear scanner effects was solved, achieving high-quality data correction and improved consistency.
Patent Information
- Application Number
- CN202511246109.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-02
- Publication Date
- 2025-12-05
AI Technical Summary
Existing ComBat correction methods cannot effectively correct complex, nonlinear scanner effects in multicenter magnetic resonance imaging data, resulting in poor correction results.
By extracting physiological indicators from multiple regions of interest, quantifying them based on mathematical parameters and scanner nonlinear effect parameters, and combining them with the Image-Combat algorithm for standardization and Bayesian parameter estimation, nonlinear batch effects caused by scanner differences and operator factors are eliminated.
It improves the comparability and consistency of multicenter imaging data, enhances the credibility of radiomics research, preserves biological information, and reduces cross-center differences.
Smart Images
Figure CN121069286A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of medical image processing, in particular to a multi-center magnetic resonance brain image correction method and system. BACKGROUND
[0002] In radiomics research, data based on large quantities can often reveal and show more potential research value. However, in the field of biomedicine, especially in the research of imaging, the image data collected from a single scanner for a research target is often limited. Especially in a limited time, according to the experimental conditions set by the research target, the image data collected from a single scanner will be even less. In view of the above situation, if the image data of multiple scanners in multiple centers can be collected, even in a limited time, sufficient data can be collected to support the research, and the reliability of the research results will be greatly improved. However, in reality, many radiological features and related biological indicators are sensitive to scanner effects (also known as batch effects), so it is very important to perform related adjustment and correction processing before collecting image data of scanners in different centers.
[0003] At present, many methods have been proposed to eliminate the scanner effect in multi-center image data, among which the ComBat correction method has the best effect and is used most widely. The ComBat method was first proposed in 2007 in the study of genomics to eliminate the influence of batch effects on experiments, and has been gradually optimized and improved to be applied to related research in medical imaging. In previous studies, ComBat correction method was often combined with resampling or RAVEL intensity normalization, but all of them have obtained the conclusion that the effect is not as good as using ComBat method only.
[0004] When the ComBat method corrects multi-center image data, it assumes that the batch effect is a linear offset. However, the effect caused by the real difference between scanners, protocol changes, software upgrades or operator differences may be nonlinear and complex. ComBat cannot correct these complex and nonlinear interactions, so when ComBat method is applied to multi-center image data, the corrected data still has batch effect, and the correction effect is not good. SUMMARY
[0005] The present application provides a multi-center magnetic resonance brain image correction method and system, which solves the problem that the existing ComBat cannot correct these complex and nonlinear interactions, and thus the corrected data still has batch effect.
[0006] In a first aspect, the present application provides a multi-center magnetic resonance brain image correction method, comprising the following steps: Multiple magnetic resonance brain images were collected from different subjects at multiple centers. Multiple physiologically relevant regions of interest (ROIs) are selected using brain imaging ROI templates. Each MRI brain image is then calibrated based on these ROIs to obtain multiple calibrated regions. From each calibrated region, multiple physiological indicators are extracted to characterize the correction effect. The physiological index values of subjects from different centers were quantified based on mathematical parameters of physiological indicators and nonlinear effect parameters of physiological indicators obtained by scanners. The mathematical parameters were estimated, and the physiological index data were standardized using the estimated mathematical parameters to obtain standardized physiological index values. The nonlinear effect parameters were estimated, and the physiological index values of different centers were corrected based on the standardized physiological index values, the estimated mathematical parameters, and the estimated nonlinear effect parameters.
[0007] Preferably, before calibrating each MRI brain image based on multiple regions of interest, each MRI brain image undergoes format conversion, time point correction, head motion correction, irrelevant variable regression, registration, and resampling preprocessing.
[0008] Preferably, the multiple regions of interest include the precentral gyrus, insula, anterior cingulate and paracingulate gyrus, hippocampus, amygdala, and thalamus, and multiple physiological indicators include ReHo, ALFF, fALFF, and VMHC.
[0009] Preferably, the mathematical parameters include the average value and the regression coefficient vector, and the nonlinear effect parameters include the additive scanning effect and the multiplicative protocol effect.
[0010] Preferably, the correction of multiple physiological index data based on standardized physiological index data and nonlinear effect parameter estimates is as follows: ; In the formula, For the calibrated scanner middle subjects On the indicators The value, As an indicator Throughout the subjects The estimated standard deviation in To estimate the obtained multiplication agreement effect parameters, Z ijg For scanners middle subjects On the indicators Standardized value To estimate the obtained additive scanner parameters, As an indicator Throughout the subjects the estimated mean value in the middle, designing a regression coefficient vector corresponding to a matrix of each different covariate.
[0011] Preferably, the corrected multiple physiological indicators are further tested by using Friedman non-parametric test and ICC inter-class correlation analysis.
[0012] In a second aspect, the present application provides a multi-center magnetic resonance brain image correction system, comprising: The acquisition module is configured to acquire multiple magnetic resonance brain images corresponding to different subjects from multiple centers respectively; The extraction module is configured to select multiple regions of interest related to physiology by using a brain image region of interest template, calibrate each magnetic resonance brain image based on the multiple regions of interest, and obtain multiple calibration regions; and extract multiple physiological indicators for representing correction effects from each calibration region. The correction module is configured to quantify physiological indicator values of different subjects in different centers based on mathematical parameters of the physiological indicators and non-linear effect parameters of the physiological indicators of the scanner; estimate the mathematical parameters, standardize the physiological indicator data by using the estimated values of the mathematical parameters, and obtain standardized physiological indicator values; estimate the non-linear effect parameters, and correct the physiological indicator values of different centers based on the standardized physiological indicator values, the estimated values of the mathematical parameters, and the estimated values of the non-linear effect parameters.
[0013] Compared with the prior art, the present application has the following beneficial effects: The present application firstly acquires multiple magnetic resonance brain images corresponding to different subjects from multiple centers respectively, and extracts corresponding physiological indicators for each magnetic resonance brain image. The physiological indicator values of different subjects in different centers are quantified based on mathematical parameters of the physiological indicators and non-linear effect parameters of the physiological indicators of the scanner. By introducing the non-linear effect parameters, the present application can more accurately model and correct complex non-linear batch effects caused by factors such as scanner differences, protocol changes, and operators in multi-center data. The physiological indicator data is standardized, the non-linear effect parameters are estimated, and the physiological indicator values of each center are corrected based on the standardized physiological indicator values, the estimated values of the mathematical parameters, and the estimated values of the non-linear effect parameters. By using the ComBat correction algorithm combined with standardization and Bayesian parameter estimation, the present application can effectively eliminate cross-center differences while retaining biological information, enhance the comparability between image data from different sources, and provide a high-quality and high-consistency data basis for subsequent imageomics research. BRIEF DESCRIPTION OF DRAWINGS
[0014] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description only aim to some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.
[0015] Figure 1 A flow chart of a multi-center magnetic resonance brain image data correction method of the present application; Figure 2 An interactive platform interface of the multi-center magnetic resonance brain image correction and verification of the present application; Figure 3 A pretreatment flow chart of the present application; Figure 4 A physiological index extraction flow chart of the present application; Figure 5 A distribution chart of the hippocampal ALFF index before Image-ComBat correction of the present application; Figure 6 A distribution chart of the hippocampal ALFF index after Image-ComBat correction of the present application; Figure 7 A box chart of the hippocampal ALFF index before Image-ComBat correction of the present application; Figure 8 A box chart of the hippocampal ALFF index after Image-ComBat correction of the present application; Figure 9 A distribution chart of the thalamic ReHo index before Image-ComBat correction of the present application; Figure 10 A distribution chart of the thalamic ReHo index after Image-ComBat correction of the present application; Figure 11 A box chart of the thalamic ReHo index before Image-ComBat correction of the present application; Figure 12 A box chart of the thalamic ReHo index after Image-ComBat correction of the present application; Figure 13 A distribution chart of the amygdala fALFF index before Image-ComBat correction of the present application; Figure 14 A distribution chart of the amygdala fALFF index after Image-ComBat correction of the present application; Figure 15Box plot of the fALFF index of the amygdala region of the present application before Image-ComBat correction; Figure 16 Box plot of the fALFF index of the amygdala region of the present application after Image-ComBat correction; Figure 17 Distribution plot of the VMHC index of the precuneus region of the present application before Image-ComBat correction; Figure 18 Distribution plot of the VMHC index of the precuneus region of the present application after Image-ComBat correction; Figure 19 Box plot of the VMHC index of the precuneus region of the present application before Image-ComBat correction; Figure 20 Box plot of the VMHC index of the precuneus region of the present application after Image-ComBat correction. DETAILED DESCRIPTION
[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.
[0017] Embodiment 1 Based on the above defects, with reference to Figure 1 The present application discloses a multi-center magnetic resonance brain image correction method, comprising the following steps: Step 1: Collect multiple magnetic resonance brain images corresponding to different subjects from multiple centers respectively.
[0018] In this embodiment, the magnetic resonance brain image data (original image data) of 40 healthy subjects (10 healthy subjects from each of four hospitals) from the four hospitals is collected.
[0019] The resting state blood oxygen level dependent functional magnetic resonance imaging (rs-BOLD-fMRI) of the subjects is selected for analysis in this embodiment. Due to its resting state characteristics, it is widely used in the clinical research of neurosurgical diseases and mental diseases in recent years. BOLD-fMRI reflects the change of deoxyhemoglobin concentration in blood due to the activation of neurons or physiological neural metabolism. Its imaging principle is that because oxyhemoglobin and deoxyhemoglobin have diamagnetic and paramagnetic properties, respectively, hydrogen protons are affected by different magnetic field strengths, causing magnetic field inhomogeneity, which causes the transverse relaxation time T2 determined by the hydrogen proton dephasing speed to be different. In the normal neural metabolism process of the human body, the change of the brain blood flow and the oxygenation / deoxyhemoglobin ratio will cause the magnetic susceptibility to change, which will be detected by magnetic resonance, thereby completing imaging.
[0020] Second step: for the original image data, the original image data of the subjects is preprocessed by using the DPABI tool in Matlab, and the specific operation process is as shown in Figure 3
[0021] Before the correction process, a unified preprocessing step needs to be performed on the nuclear magnetic image data, and the absence of the preprocessing step will affect the subsequent experimental research process.
[0022] The preprocessing process needs to combine the functional image and structural image data of the subjects. First, the original image DICOM format is converted to 3D NIFTI format to improve the stability and efficiency of data processing. Considering the possible instability of the scanner at the initial stage and the adaptation process of the subjects at the initial stage of scanning, the first 10 time points of each subject's data are removed, and then the remaining image data is corrected in terms of slice number. In order to avoid the influence of image offset and artifacts caused by individual factors of the subjects on the experimental research, head motion correction needs to be performed. In order to remove noise and irrelevant signal interference, irrelevant variable regression processing needs to be performed. Then, the structural image of the whole brain map of the subjects is segmented into white matter, gray matter, and cerebrospinal fluid by using a unified segmentation algorithm, and is registered, spatially standardized, and image filtered to eliminate differences between subjects and other interference. Finally, the images processed above are resampled to 3mm×3mm×3mm, and the data smoothing processing needs to be combined with the calculation requirements of different indicators.
[0023] Third step: for the preprocessed image data, the marking of the region of interest (ROI) and the extraction of radiological features or physiological indicators are continued in DPABI, and the specific operation process is as shown in Figure 4
[0024] The basic indexes for extraction and analysis include ReHo, ALFF, fALFF and VMHC. In the calculation process of a series of indexes such as ReHo, ALFF, fALFF and VMHC, the calculation region needs to be specified, and a template is needed to demarcate the calculation region of interest at this time. In brain image research, standard space templates can be used or formulated according to needs. In the present application, the ALL (Anatomical Automatic Labeling) template provided by the MNI (Montreal Neurological Institute) is selected. The ALL template has a total of 116 regions, including 90 regions belonging to the cerebral cortex, and the remaining 26 regions belonging to the cerebellum and subcortical gray matter structure. At the same time, the present application selects six important regions related to movement, sensation, emotion and memory from the 116 regions for index extraction and subsequent analysis, and the six important regions include:
[0025] Precentral gyrus: contains a large number of large ones, is the movement center.
[0026] Insula: the emotion control center in the brain, controls the generation and origin of many feelings and emotions.
[0027] Anterior cingulate and paracingulate gyrus: associated with the somatic motor area and somatosensory area.
[0028] Hippocampus: mainly responsible for short-term storage conversion and orientation functions.
[0029] Amygdala: part of the limbic system, brain tissue that produces emotions, recognizes emotions, regulates emotions, and controls learning and memory.
[0030] Thalamus: the highest center of sensation, the most important sensory relay station.
[0031] Step 4: Perform Image-Combat correction processing on the extracted multi-center radiology features or physiological index data to eliminate the difference caused by the scanner effect.
[0032] The basic principle of the Image-Combat algorithm is to adjust the data in the framework of the frequency statistical linear model to eliminate the technical variation caused by equipment, protocol, operator and experimental batch changes between different batches of data. It uses the parameter estimation method in frequency statistics to estimate the parameters in the model through least squares method, Bayesian experience method and other techniques.
[0033] The algorithm assumes that the scanner The subject The measurement expression value of the physiological index can be expressed in general form as: ; in, For scanners middle subjects On the indicators The value, As an indicator In all subjects The average value in Design matrix for the covariates of interest, Design a regression coefficient vector corresponding to the matrix for each different covariate. For scanners For indicators The additive scanning effect For the error term ( Scanners under the influence of For indicators The multiplication agreement effect.
[0034] The algorithm typically includes three steps: data standardization, scanner effect parameter estimation, and scanner effect elimination. Step two, the core of the Image-Combat algorithm, is an empirical Bayesian method for removing the scanner effect. Its idea is to obtain prior information from experimental conditions and, in conjunction with this prior information, use probability distributions to describe the uncertainty of the Bayesian model parameters. By continuously acquiring new data, the probability distributions of the model's corresponding parameters are updated to more closely approximate reality. The additive scanning effect parameters are estimated using a Bayesian posterior method. Multiplication Agreement Effect Parameter This represents the degree of influence of different scanners on the data, allowing for the correction and adjustment of the corresponding scanner effects. The specific implementation of each step is as follows:
[0035] (1) Data standardization. First, the parameters are estimated using the least squares method. , and ( , and They are respectively , and (estimated value), with limitations. To ensure the identifiability of the parameters. Secondly, estimation. The values are:
[0036] ; Finally, the data is standardized based on the estimated values of each parameter, resulting in: ; in for the scanner in the subject on the index , for the scanner in the subject on the index , for the index in all subjects , is the design matrix for the covariates of interest, is the regression coefficient vector corresponding to each different covariate design matrix, for the index in all subjects .
[0037] (2) Estimate the parameters of the scanner effect in the model using the Bayesian posterior distribution.
[0038] Assume that the standardized feature (index) values follow a specific normal distribution: .
[0039] Since the mean and standard deviation have been removed in the standardization process, the index values on the scanner in the subject are only affected by the batch effect, including the additive scanner effect and the multiplicative protocol effect .
[0040] Assume that the standardized index values follow a normal distribution with mean and variance , and subsequently estimate the batch effect parameters and using the Bayesian posterior method.
[0041] Since the standardized index values are assumed to follow a specific normal distribution as the prior condition, the conjugate posterior distributions of and are normal distribution and inverse gamma distribution (Inverse Gamma), respectively, i.e. where , , and are empirically estimated from standardized data using the method of moments, so that ; ; The above two parameters capture the differences between batches.
[0042] (3) Subtract and divide by the estimated additive scanner parameter and multiplicative protocol effect parameter from the raw data, respectively.
[0043] ; Through the above steps, Image-Combat can effectively eliminate batch effects and enhance the comparability of data between different batches. At the same time, during the batch effect adjustment process, the differences in actual biological effects will not be affected by the algorithm, thereby preserving the biological information between samples.
[0044] Among the 24 processes of 6 brain regions and 4 indicators, the Image-Combat correction results of the VMHC indicator in the precentral gyrus, the ALFF indicator in the hippocampus, the fALFF indicator in the amygdala, and the ReHo indicator in the thalamus were selected, and were displayed in the form of distribution map and box plot, Figures 5 to 20 The correction results are shown.
[0045] Example 2 The verification method of the correction effect of multi-center image data in existing research is not perfect, which may lead to the inability to accurately evaluate the actual influence of the correction method on the homogeneity of multi-center data.
[0046] Step 5: Perform Friedman non-parametric test analysis and ICC inter-group correlation analysis on the multi-center image data after correction to verify the correction effect of the Image-Combat algorithm. Most previous studies have focused on applying the ComBat algorithm to correct multi-center data, but have not considered which statistical method to use to verify the correction effect of ComBat algorithm. The specific principle of the verification method is as follows.
[0047] Friedman non-parametric test.
[0048] Non-parametric Friedman test is based on permutation and randomization, and does not depend on the specific distribution of the population. The basic idea is to sort each observation value within each group, and calculate the rank of each observation value according to the sorted result. Then, by comparing the average value of the rank of each sample, it is determined whether there is a difference between the samples.
[0049] The basic steps of Friedman test are as follows: (1) Sort the observations of each sample by size and assign a rank (starting from 1) to each observation.
[0050] (2) For each sample, calculate the average of the ranks of all observations.
[0051] (3) Calculate the Friedman statistic Q, the formula is as follows: ; Where is the number of treatments, is the number of blocks, represents the rank of the treatment in blocks, .
[0052] (7) Query the distribution of the Friedman test statistic Q, if the p-value of the Friedman test is greater than 0.05, it means that at the significance level of 0.05, the null hypothesis cannot be rejected. The null hypothesis usually means that the medians or means of all groups are equal, that is, there is no significant difference between the samples.
[0053] ICC correlation analysis ICC (Intraclass Correlation Coefficient) can be used to evaluate and measure the consistency or similarity of data within the same group. In medical imaging, ICC is widely used in the study of the reliability and consistency of medical imaging data in different situations, which can help to evaluate and improve the reliability of medical decision and scheme. In the present application, ICC can be used to evaluate the consistency of image data between different sites, and the ICC index of image data between different sites before and after correction is calculated, and the effect of Image-Combat correction method is measured according to the change of ICC value before and after correction.
[0054] ICC is based on the principle of variance decomposition, which decomposes the total variation observed into different sources of variation. For multi-group or multi-site data, the total variation is always composed of inter-group variation and intra-group variation. ICC measures the proportion of variation between data in the same group relative to the total variation.
[0055] The formula for calculating ICC can be selected according to the specific data category and analysis direction, as follows: The value of absolute consistency in single measurement in one-way random pattern, the formula is as follows: ; The value of absolute consistency in average measurement in one-way random pattern, the formula is as follows: ; The value of absolute consistency in single measurement in two-way random / mixed mode, the calculation formula is: ; The value of absolute consistency in average measurement in two-way random / mixed mode, the calculation formula is: ; The value of consistency in single measurement in two-way random / mixed mode, the calculation formula is: ; The value of consistency in average measurement in two-way random / mixed mode, the calculation formula is: ; Wherein, 1 and in the ICC brackets respectively represent single measurement and average measurement, and respectively represent absolute consistency and consistency, is the row variable mean square value, is the column variable mean square value, is the within-group variation mean square value, is the error mean square value.
[0056] In statistical analysis, the selection of the ICC model involves three aspects: single / double, mixed / random, and consistency / absolute consistency. In view of the characteristics of the data of the application, the two-way mixed consistency average measurement ICC model is selected. The value of ICC ranges from 0 to 1. Less than 0.5 indicates poor consistency, 0.5 to 0.75 indicates moderate consistency, 0.75 to 0.9 indicates good consistency, and greater than 0.9 indicates very good consistency.
[0057] Friedman non-parametric test and ICC correlation analysis are performed on the indicators before and after correction, and the results are as follows: Before Image-ComBat correction, only 6 of the total 24 indicators from 6 brain regions have Friedman test p values greater than 0.05, and the p values of the remaining 18 indicators are all less than 0.05, which means that the significance level of most indicators is above 0.05, so the null hypothesis (the null hypothesis usually refers to the equality of the medians or means of all groups) is rejected, that is, the indicator values of the four centers have significant differences before correction. The results after correction are shown in Table 1, the p values of all indicators are improved and are all greater than 0.05, which indicates that the scanner effect is eliminated and the significant differences of the indicator values of the four centers are significantly weakened after correction.
[0058] Table 1 Friedman test P value statistics of multiple indicators before and after ComBat correction After Image-ComBat correction, the ICC values of a total of 24 indicators from 6 brain regions are improved to different degrees after correction, and all are improved to a high correlation level of more than 0.9 after correction. The consistency of many indicators with poor correlation before correction is significantly enhanced after correction. The specific ICC values before and after correction are shown in Table 2.
[0059] Table 2. ICC value statistics of multiple indicators before and after ComBat correction Embodiment 3 Based on the same concept, the application also provides a multi-center magnetic resonance brain image correction system, which refers to Figure 2 , comprising an acquisition module, an extraction module and a correction module.
[0060] The acquisition module is used to acquire a plurality of magnetic resonance brain images corresponding to different subjects from different centers respectively.
[0061] The extraction module is used to select a plurality of regions of interest related to physiology by a brain image region of interest template, calibrate each magnetic resonance brain image based on the plurality of regions of interest to obtain a plurality of calibration regions; and extract a plurality of physiological indicators for representing the correction effect from each calibration region.
[0062] The correction module is used to quantify the physiological indicator values of different center subjects based on the mathematical parameters of the physiological indicators and the nonlinear effect parameters of the physiological indicators of the scanner; estimate the mathematical parameters, standardize the physiological indicator data by the mathematical parameter estimate value to obtain standardized physiological indicator values; estimate the nonlinear effect parameters, and correct the physiological indicator values of different centers based on the standardized physiological indicator values, the mathematical parameter estimate value and the nonlinear effect parameter estimate value.
[0063] The application can provide a reference research scheme for the homogenization correction of multi-center medical image data and a suitable statistical analysis method for testing the correction effect, so as to improve the reliability and repeatability of multi-center image related research, so that the follow-up research and expansion of imageomics are not affected by multiple variation sources. At the same time, it can avoid the waste of experimental resources caused by the failure of research scheme design to a certain extent, and thus improve the overall efficiency of scientific research.
[0064] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.
[0065] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.
Claims
1. A method for correcting multicenter magnetic resonance brain images, characterized in that, The method comprises the following steps: Collecting multiple magnetic resonance brain images corresponding to different subjects from multiple centers respectively; Selecting multiple regions of interest related to physiology through a region of interest template of the brain image, labeling each magnetic resonance brain image based on the multiple regions of interest to obtain multiple labeled regions, and extracting multiple physiological indicators for representing correction effects from each labeled region; Quantifying physiological indicator values of subjects in different centers based on mathematical parameters of the physiological indicators and nonlinear effect parameters of the physiological indicators of a scanner; Estimating the mathematical parameters, standardizing the physiological indicator data through the estimated values of the mathematical parameters to obtain standardized physiological indicator values, estimating the nonlinear effect parameters, and correcting the physiological indicator values of different centers based on the standardized physiological indicator values, the estimated values of the mathematical parameters, and the estimated values of the nonlinear effect parameters.
2. A multi-center magnetic resonance brain image correction method as claimed in claim 1, characterized in that, Before labeling each magnetic resonance brain image based on the multiple regions of interest, performing pre-processing such as format conversion, time point correction, head motion correction, irrelevant variable regression, registration, and resampling on each magnetic resonance brain image.
3. The multi-center magnetic resonance brain image correction method of claim 1, wherein, The multiple regions of interest include the anterior cingulate gyrus, the insula, the anterior cingulate gyrus and the paracingulate gyrus, the hippocampus, the amygdala, and the thalamic region, and the multiple physiological indicators include ReHo, ALFF, fALFF, and VMHC.
4. The multi-center magnetic resonance brain image correction method of claim 1, wherein, The mathematical parameters include mean values and regression coefficient vectors, and the nonlinear effect parameters include additive scanning effects and multiplicative protocol effects.
5. The multi-center magnetic resonance brain image correction method of claim 1, wherein, The correction of the multiple physiological indicator data based on the standardized physiological indicator data and the estimated values of the nonlinear effect parameters is as follows: ； where is the value of the index on the post-scaners is the index is the estimated multiplicative scanner effect parameter Z ijg is the value of the index on the post-scaners is the estimated additive scanner parameter is the index is the regression coefficient vector corresponding to each different covariate design matrix. 6. The multi-center magnetic resonance brain image correction method of claim 1, wherein, The method further comprises testing the corrected multiple physiological indicators by using Friedman nonparametric test and ICC interclass correlation analysis.
7. A multi-center magnetic resonance brain image correction system, comprising: The method comprises the following steps: A collecting module is configured to collect multiple magnetic resonance brain images corresponding to different subjects from multiple centers respectively; An extracting module is configured to select multiple regions of interest related to physiology through a region of interest template of the brain image, label each magnetic resonance brain image based on the multiple regions of interest to obtain multiple labeled regions, and extract multiple physiological indicators for representing correction effects from each labeled region; A correcting module is configured to quantify physiological indicator values of subjects in different centers based on mathematical parameters of the physiological indicators and nonlinear effect parameters of the physiological indicators of a scanner; Estimate the mathematical parameters, standardize the physiological indicator data through the estimated values of the mathematical parameters to obtain standardized physiological indicator values, estimate the nonlinear effect parameters, and correct the physiological indicator values of different centers based on the standardized physiological indicator values, the estimated values of the mathematical parameters, and the estimated values of the nonlinear effect parameters.