Method for determining range and peak point of neural activity synchronization region of interest based on pixel time sequence and application

Abstract

The invention discloses a method for determining the range and peak points of a neural activity synchronization region of interest (ROI) based on a pixel time sequence and application. According to the method, the consistency of the pixel time sequence is calculated according to the bidirectional rank of the pixel time sequence so as to determine the ROI range, and the pixel which can best reflect the overall time sequence characteristics of the ROI is selected as the ROI peak point according to the principal component weighted projection value of the pixel time sequence, so that the interference of noise on the determination of the ROI range is reduced; the accuracy and the stability of the neural activity synchronism calculation taking the ROI as the unit can be improved; the method can be used independently or in combination with an anatomical template, the problem that a traditional method depends on the template matching degree and precision is solved, and the problem that the ROI range and peak points of a tested individual are difficult to obtain through the traditional method is solved; according to the method, the calculation of traversing the pixels is independent among the pixels, so that the implementation of parallel acceleration calculation and control error propagation is facilitated; the method can be used for neural activity synchronism calculation and analysis based on medical images or neural images.

Description

Technical Field

[0001] The present invention belongs to the field of medical imaging and neuroimaging technology, and specifically relates to a method and application for determining the range and peak point of a region of interest based on pixel timing in the calculation and analysis of neural activity synchronization. Background Art

[0002] Neural activity synchronization can reflect the brain's ability to coordinate processing of external stimuli, as well as its ability to spontaneously activate in the absence of specific stimuli. It is an important basis for constructing brain networks and a key indicator for studying cognitive function, the mechanisms of neurological and psychiatric disorders, and clinical assessment. Neural activity data can be collected using either electrophysiological or neuroimaging techniques. Neuroimaging data typically has a higher spatial resolution, allowing temporal correlations to be calculated at the pixel level or at the region of interest (ROI) level, which can then be used to reflect neural activity synchronization.

[0003] Calculating temporal correlations on a pixel-by-pixel basis is susceptible to noise, and due to the large number of pixels, multiple comparisons are problematic, increasing the probability of Type I or Type II errors in statistical tests. Calculating temporal correlations on a ROI-by-ROI basis can overcome these issues, but requires predetermining the ROI's range and peak point—that is, which pixels the ROI contains and which pixel it is centered on. Existing techniques for determining the ROI's range primarily rely on anatomical templates or by setting the ROI's range to a fixed value (or a fixed number of pixels) between 10 and 100 cubic millimeters. Existing techniques typically determine the ROI's peak point by selecting the pixel with the strongest temporal variation or the pixel corresponding to the ROI's geometric center.

[0004] The following problems exist when using existing technologies to determine the range of ROI. Determining the range of ROI based on anatomical templates is highly dependent on the matching degree and accuracy of the template. Since anatomy and function do not correspond precisely, there is a problem of misalignment between functional subregions and anatomical subregions when using this method to determine the range of ROI, especially at the junction of various brain regions, the misalignment is more serious. In addition, if the template is different from the subject group involved in the current study, there will be a large systematic error in determining the range of ROI based on the template. Using a fixed ROI range may be too rough for brain regions with fine structure and rich functions, and may be too fine for brain regions with simple structure and concentrated functions. Therefore, setting the range of ROI to a fixed value is highly dependent on the personal experience of the researcher, and improper setting will seriously affect the statistical power and interpretability of the results of neural synchronization calculations.

[0005] The peak point of the ROI determined by the prior art has the following problems. Since the time series of the pixel with the most dramatic change may not be similar to the time series of the surrounding pixels, the time series of the pixel with the most dramatic change cannot fully reflect the characteristics of the overall time series of the ROI. Since the anatomical sub-region and the functional sub-region do not correspond accurately, the time series of the pixel at the geometric center of the ROI cannot fully reflect the characteristics of the overall time series of the ROI. SUMMARY

[0006] The present application aims at the deficiencies of the prior art and provides a method for determining the range and peak point of a region of interest of neural activity synchronicity based on pixel time series.

[0007] Another object of the present application is to provide an application of the method for determining the range and peak point of a region of interest of neural activity synchronicity based on pixel time series.

[0008] The specific technical solution for achieving the object of the present application is as follows.

[0009] The method for determining the range and peak point of a region of interest of neural activity synchronicity based on pixel time series has the following characteristics: the range of the ROI is determined according to the bidirectional rank of the pixel time series, and the pixel that can best reflect the overall time series characteristics of the ROI is selected as the peak point of the ROI according to the principal component weighted projection value of the pixel time series.

[0010] The bidirectional rank refers to the rank of the observation data of the same pixel at each time point, and the rank of the same pixel at each time point.

[0011] The principal component weighted projection value refers to the projection value of the pixel time series on the principal component, and the projection value is weighted by the eigenvalue of the principal component analysis to obtain the principal component weighted projection value of the pixel time series.

[0012] Further, the method specifically includes the following steps.

[0013] Step 1: In the image foreground range or the range selected by the researcher, select a pixel as the initial peak point of the ROI, and the initial range of the ROI only contains the pixel.

[0014] Step 2: Expand the range of the ROI by one pixel in each direction, then calculate the bidirectional rank of the time series of all pixels in the range of the ROI, and then calculate the Kendall concordance coefficient W of the bidirectional rank, which represents the consistency of the time series of the pixel.

[0015] Step 3:

[0016] 3.1: If W is greater than or equal to the threshold value W_th, repeat step 2 until W is less than the threshold value W_th, and then perform step 3.2.

[0017] 3.2: If W is less than a threshold value W_th, then the range of the ROI is reduced by one pixel in the expansion direction, and the reduced range is taken as the range of the ROI;

[0018] Step 4: Principal component analysis is performed on the time series of all pixels in the ROI range, and a principal component weighted projection value of each pixel time series is calculated. The pixel with the largest weighted projection value is taken as the peak point of the ROI;

[0019] Step 5: In the image foreground range or the range selected by the researcher, steps 1-4 are performed for each pixel until all pixels are traversed;

[0020] Step 6: For multiple ROIs with the same peak point, they are merged into one ROI; the peak points of the merged ROIs are the same, which is the peak point of the merged ROI; the range of the merged ROI is equal to the union of the ranges of the merged ROIs;

[0021] Step 7: For ROIs with different peak points, do not merge them, and keep their respective ranges and peak points;

[0022] Alternatively

[0023] For ROIs with different peak points, the following steps are used to merge them:

[0024] 7.1: According to the distance between the peak points of the ROIs, perform agglomerative hierarchical clustering on all ROIs;

[0025] 7.2: According to the clustering results, calculate the intra-class maximum distance D_max for each class, then prune the clustering results, and only keep the classes with D_max less than or equal to a threshold value D_th; according to the pruning results, calculate the intra-class Kendall concordance coefficient Winc for each class, then prune the pruning results again, and only keep the classes with Winc greater than or equal to a threshold value Winc_th;

[0026] Alternatively

[0027] According to the clustering results, calculate the intra-class Kendall concordance coefficient Winc for each class, then prune the clustering results, and only keep the classes with Winc greater than or equal to a threshold value Winc_th; according to the pruning results, calculate the intra-class maximum distance D_max for each class, then prune the pruning results again, and only keep the classes with D_max less than or equal to a threshold value D_th;

[0028] Alternatively

[0029] According to the clustering results, calculate the intra-class maximum distance D_max for each class, then prune the clustering results, and only keep the classes with D_max less than or equal to a threshold value D_th;

[0030] Alternatively

[0031] According to the clustering result, the Kendall's Winc is calculated in each class, and then the clustering result is pruned to only keep the class whose Winc is greater than or equal to a threshold Winc_th;

[0032] 7.3: According to the pruning result, the ROIs are merged, and the range of the merged ROI is equal to the union of the ranges of the merged ROIs;

[0033] 7.4: For each merged ROI, the principal component analysis is performed on all the pixel time series in the range of the ROI, the principal component weighted projection value of each pixel time series is calculated, and the pixel with the maximum weighted projection value is taken as the peak point of the ROI.

[0034] Further, the steps 1-4 are performed for each pixel in a pixel-by-pixel sequential manner or in a multi-pixel parallel manner.

[0035] The method is applied to the calculation and analysis of neural activity synchrony based on medical images or neuroimages.

[0036] The present application takes a single pixel as the initial peak point and the initial range of the ROI in the image foreground range or the range selected by the researcher, calculates the pixel time series consistency according to the bidirectional rank of the pixel time series, expands the range of the ROI around the pixel if the pixel time series consistency is greater than or equal to a threshold, until the pixel time series consistency is less than the threshold, calculates the principal component weighted projection value of the pixel time series in the range of the ROI, selects the pixel with the maximum weighted projection value, which is the pixel whose time series can best reflect the characteristics of the overall time series of the ROI, and takes the pixel as the peak point of the ROI, and then traverses all the pixels and merges the repeated or similar ROIs.

[0037] The present application has the following advantages: the pixel time series consistency is calculated according to the bidirectional rank of the pixel time series to determine the ROI range, the pixel whose time series can best reflect the characteristics of the overall time series of the ROI is selected as the peak point of the ROI according to the principal component weighted projection value of the pixel time series, the interference of noise on the determination of the ROI range is reduced, and the accuracy and stability of the calculation of the neural activity synchrony in the unit of ROI are improved; the present application can be used alone or in combination with the anatomical template, and the problem of the dependence of the traditional method on the matching degree and accuracy of the template is overcome, and the problem that the traditional method is difficult to obtain the ROI range and peak point of the individual subject is solved; the calculation process of traversing the pixels in the present application is independent between the pixels, which is not only conducive to the implementation of parallel acceleration calculation, but also conducive to the control of error propagation. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 The flowchart of the method of the present application;

[0039] Figure 2This is a schematic diagram of the agglomerative hierarchical clustering results described in the present invention. DETAILED DESCRIPTION

[0040] The present invention adopts the following technical solution: within the foreground range of the image or the range selected by the researcher, a single pixel is used as the initial peak point and initial range of the ROI; the pixel timing consistency is calculated according to the bidirectional rank of the pixel timing, and if the pixel timing consistency is greater than or equal to the threshold, the range of the ROI is expanded to the surrounding area until the pixel timing consistency is less than the threshold; within the ROI range, the weighted projection value of the principal component of the pixel timing is calculated, and the pixel with the largest weighted projection value is selected. The pixel timing that best reflects the characteristics of the overall ROI timing is used as the ROI peak point; all pixels are traversed, and then repeated or similar ROIs are merged.

[0041] Pixel timing consistency is assessed using the Kendall harmonic coefficient (W). The Kendall harmonic coefficient (W) ranges from 0 to 1, with W closer to 1 indicating a higher degree of consistency in the timing of each pixel. When the timing fluctuations of pixels within the ROI vary significantly, the Kendall harmonic coefficient (W) obtained using existing methods cannot accurately reflect pixel timing consistency.

[0042] To solve this problem, the present invention proposes to use bidirectional rank calculation W, as follows:

[0043] Assume that the ROI contains N pixels, and each pixel time series contains K observations at each moment. Use an N-row, K-column matrix Mat to represent the observation value of each pixel in the ROI at each moment.

[0044] First, the observation data of the same pixel at each moment are sorted and assigned a rank to obtain the rank matrix Mat_r.

[0045] Mat_r = Rank(Mat, row) (Formula 1)

[0046] Then, the ranks of each pixel at the same moment are sorted and assigned ranks to obtain a bidirectional rank matrix Mat_rc.

[0047] Mat_rc = Rank(Mat_r, col) (Equation 2)

[0048] Among them, Rank() is a function whose function is to sort the first parameter Mat by row or column according to the second parameter of the function, which is row or col, and then replace the sorted values ​​with ordinal numbers.

[0049] Rij is used to represent the bidirectional rank of the i-th pixel at the j-th moment, and Rij is equal to the element Mat_rc(i, j) in the i-th row and j-th column of the bidirectional rank matrix Mat_rc.

[0050] Rij = Mat_rc(i, j) (Formula 3)

[0051] Then the rank sum S of each pixel is calculated i ,

[0052] (Formula 4)

[0053] The mean deviation square sum S of the rank sum is calculated

[0054] (Formula 5)

[0055] Wherein represents the average value of the rank sum of N pixels,

[0056] (Formula 6)

[0057] The theoretical maximum mean deviation square sum S is calculated max ,

[0058] (Formula 7)

[0059] The Kendall concordance coefficient W is calculated.

[0060] (Formula 8)

[0061] If the data of different pixels are equal at the same time, the denominator needs to be corrected when calculating the Kendall concordance coefficient W. It is assumed that the jth moment needs to be corrected, Tj is the correction term, and t is the number of data in the same group of rank at the jth moment.

[0062] (Formula 9)

[0063] (Formula 10)

[0064] The principal component analysis of the pixel time sequence mainly includes the following steps: first, the observation data of N pixels at K moments are standardized, then the covariance matrix is calculated, and then the covariance matrix is decomposed to obtain the eigenvalue and the eigenvector; the eigenvalues are sorted from large to small; the principal components whose eigenvalues are greater than 1 are retained, or the principal component with the largest eigenvalue is retained, or multiple principal components in the front are retained. It is assumed that a total of L principal components are retained.

[0065] In order to efficiently and accurately select the pixel that can best reflect the overall time sequence characteristics of the ROI, the present application proposes to use the eigenvalue to weight the projection value of the pixel time sequence on the principal component, and then take the pixel with the maximum weighted projection value as the peak point of the ROI, and the specific method is as follows:

[0066] The projection value of the pixel time series on the jth principal component wherein, represents the ith element of the eigenvector of the jth principal component, represents the ith element of the pixel time series.

[0067] Equation 11

[0068] Then the weighted square sum of the projection values of the pixel time series on each principal component is calculated and the square root of the result is taken as the principal component weighted projection value WPC of the pixel time series. Wherein, represents the weighting coefficient of the jth principal component, represents the eigenvalue of the pth principal component.

[0069] Equation 12

[0070] Equation 13

[0071] The WPC of each pixel time series is calculated using Equations 11-13, and then the pixel with the largest WPC in the ROI range is selected as the peak point of the ROI, and the pixel time series most reflects the characteristics of the overall time series of the ROI.

[0072] The "merging of repeated or similar ROIs" can be achieved by using agglomerative hierarchical clustering and pruning. Agglomerative hierarchical clustering is a bottom-up clustering method that constructs a hierarchical tree by gradually merging similar objects or classes. The core steps are as follows: in the initialization stage, each ROI is regarded as a class; the distance matrix between all pairs of classes is calculated, the two classes with the closest distance are found and merged to generate a new class, and then the distance matrix is updated and the distance between the new class and other classes is recalculated; repeat the above process of calculating the distance between class pairs and merging to generate a new class until merging into one class.

[0073] As shown in Figure 2 , the result of agglomerative hierarchical clustering can be represented as a tree diagram, and the bottom layer of classes are the ROIs to be merged. From bottom to top, each layer is merged until one class is obtained. According to the maximum distance between pixels in the class or the consistency of the pixel time series, the tree diagram is pruned and the classes whose maximum distance between pixels or pixel time series consistency meet the conditions are retained.

[0074] According to the result after pruning, the ROIs are merged. As shown in Figure 2 , the classes that meet the conditions and are retained include ROI-2-1, ROI-1-1, ROI-1-2, ROI-1, ROI-2, ROI-3, ROI-4 and ROI-5. Search the pruned tree diagram from top to bottom, and select the uppermost class in the branch that meets the conditions as the ROI after merging of the branch. As shown inFigure 2 As shown, ROI-1, ROI-2 and ROI-3 are merged into ROI-2-1, and ROI-4 and ROI-5 are merged into ROI-1-2. That is, finally, two types of ROI-2-1 and ROI-1-2 are merged. The range of ROI-2-1 is the union of ROI-1, ROI-2 and ROI-3, and the range of ROI-1-2 is the union of ROI-4 and ROI-5.

[0075] Embodiment 1

[0076] The embodiment includes the following steps:

[0077] (1) In the image foreground range or the range selected by the researcher, a pixel is selected as the initial peak point of the ROI, and the initial range of the ROI only contains the pixel;

[0078] (2) The range of the ROI is expanded by one pixel in each direction, then the bidirectional rank of the time sequence of all pixels in the ROI range is calculated, and the Kendall concordance coefficient W of the bidirectional rank is calculated, which represents the consistency of the pixel time sequence;

[0079] (3.1) If W is greater than or equal to the threshold value W_th, repeat step (2) until W is less than the threshold value W_th, and then perform step (3.2);

[0080] (3.2) If W is less than the threshold value W_th, the range of the ROI is reduced by one pixel along the expansion direction, and the reduced range is taken as the range of the ROI;

[0081] (4) Principal component analysis is performed on the time sequence of all pixels in the ROI range, and the principal component weighted projection value of each pixel time sequence is calculated, and the pixel with the maximum weighted projection value is taken as the peak point of the ROI;

[0082] (5) In the image foreground range or the range selected by the researcher, steps (1) to (4) are performed for each pixel until all pixels are traversed;

[0083] (6) For multiple ROIs with the same peak point, they are merged into one ROI; the peak points of the merged ROIs are the same, and the peak point of the merged ROI is the same as the peak point of the merged ROI; the range of the merged ROI is equal to the union of the ranges of the merged ROIs;

[0084] (7) ROIs with different peak points are not merged, and their respective ranges and peak points are retained;

[0085] The "performing steps (1) to (4) for each pixel" in the above step (5) can be implemented in a pixel-by-pixel sequential execution manner, or in a multi-pixel parallel execution manner.

[0086] The "image foreground range" refers to a range retained after the image removes a background part.

[0087] The "researcher-selected range" can be a range drawn by the researcher or a range determined by the researcher according to a template.

[0088] The "time sequence" refers to a sequence of signal intensity changes over time.

[0089] Preferably, the threshold value W_th is 0.8.

[0090] Embodiment 2

[0091] This embodiment includes the following steps:

[0092] (1) In the image foreground range or the researcher-selected range, select a pixel as an initial peak point of the ROI, and the initial range of the ROI only contains the pixel;

[0093] (2) Expand the range of the ROI by one pixel in each direction, then calculate the bidirectional rank of the time sequence of all pixels in the ROI range, and then calculate the Kendall's W coefficient of concordance of the bidirectional rank, and use W to represent the consistency of the pixel time sequence;

[0094] (3.1) If W is greater than or equal to the threshold value W_th, repeat step (2) until W is less than the threshold value W_th, and then perform step (3.2);

[0095] (3.2) If W is less than the threshold value W_th, then reduce the range of the ROI by one pixel along the expansion direction, and the reduced range is the range of the ROI;

[0096] (4) Perform principal component analysis on the time sequence of all pixels in the ROI range, and calculate the principal component weighted projection value of each pixel time sequence, and take the pixel with the maximum weighted projection value as the peak point of the ROI;

[0097] (5) In the image foreground range or the researcher-selected range, perform steps (1) to (4) on each pixel until all pixels are traversed;

[0098] (6) For multiple ROIs with the same peak point, merge them into one ROI; the peak points of the merged ROIs are the same, and the peak point is the peak point of the merged ROI; the range of the merged ROI is equal to the union of the ranges of the merged ROIs;

[0099] (7) For ROIs with different peak points, the following steps are used to merge:

[0100] (7.1) According to the distance between the peak points of the ROIs, perform agglomerative hierarchical clustering on all ROIs;

[0101] (7.2) According to the clustering result, the maximum distance D_max in each class is calculated, and then the clustering result is pruned to retain only the classes with D_max less than or equal to a threshold D_th; according to the pruning result, the Kendall's Winc in each class is calculated, and then the pruning result is pruned again to retain only the classes with Winc greater than or equal to a threshold Winc_th;

[0102] (7.3) According to the pruning result, the ROIs are merged, and the range of the merged ROI is equal to the union of the ranges of the merged ROIs corresponding to the merged ROIs;

[0103] (7.4) For each merged ROI, the principal component analysis is performed again on all pixel time series in the range of the ROI, and the principal component weighted projection value of each pixel time series is calculated, and the pixel with the maximum weighted projection value is taken as the peak point of the ROI.

[0104] The steps (1) to (4) performed on each pixel can be implemented in a pixel-by-pixel sequential manner or in a multi-pixel parallel manner.

[0105] The "image foreground range" refers to the range retained after removing the background part of the image.

[0106] The "researcher-selected range" can be a range drawn by the researcher or a range determined by the researcher according to a template.

[0107] The "time series" refers to a sequence of signal intensity changes over time.

[0108] The "merging the ROIs according to the pruning result" refers to searching the pruned tree diagram from top to bottom, and selecting the uppermost class in the branch that meets the conditions as the ROI after merging the branch.

[0109] Preferably, the threshold W_th is 0.8.

[0110] Preferably, the threshold Winc_th is 0.5.

[0111] Preferably, the threshold D_th is 10 pixels.

[0112] Example 3

[0113] This example is the same as Example 2, except that step (7.2) is different, as follows:

[0114] (7.2) According to the clustering result, calculate the Kendall's Winc coefficient of each class, then prune the clustering result, only keep the class whose Winc is greater than or equal to the threshold Winc_th; according to the pruning result, calculate the maximum distance D_max of each class, then prune the pruning result again, only keep the class whose D_max is less than or equal to the threshold D_th.

[0115] Embodiment 4

[0116] This embodiment is the same as embodiment 2, only step (7.2) is different, which is as follows:

[0117] (7.2) According to the clustering result, calculate the maximum distance D_max of each class, then prune the clustering result, only keep the class whose D_max is less than or equal to the threshold D_th.

[0118] Embodiment 5

[0119] This embodiment is the same as embodiment 2, only step (7.2) is different, which is as follows:

[0120] (7.2) According to the clustering result, calculate the Kendall's Winc coefficient of each class, then prune the clustering result, only keep the class whose Winc is greater than or equal to the threshold Winc_th.

[0121] Embodiment 6

[0122] A method for determining the range and peak point of a region of interest (ROI) based on pixel timing of neural activity synchronization, and application of the method in calculation and analysis of neural activity synchronization based on medical images or neuroimaging.

[0123] Taking the calculation of neural activity synchronization between brain regions of a human brain based on functional magnetic resonance imaging (fMRI) as an example. Assume that the fMRI data contains 1000 frames, i.e. contains brain maps of the same subject at 1000 time points, each frame contains 32 layers, each layer contains 64 rows and 64 columns, i.e. each frame contains 32x64x64=131072 pixels. The steps of any one of embodiments 1-5 are used to determine the range and peak point of the ROI based on pixel timing, and it is assumed that 100 ROIs are finally obtained. These ROIs can be used alone, or the ROI range can be intersected with an anatomical template to select the part that is consistent in structure and function as the range of the ROI. Then calculate the time series correlation coefficient between each pair of ROIs in units of ROIs to obtain a 100x100 correlation coefficient matrix for describing the neural activity synchronization between brain regions.

Claims

1. A method for determining the range and peak of neural activity synchronization interest region based on pixel time series, characterized in that: The pixel timing consistency is calculated based on the bidirectional rank of the pixel timing to determine the ROI range. The pixel that best reflects the overall timing characteristics of the ROI is selected as the ROI peak point based on the weighted projection value of the principal component of the pixel timing. The bidirectional rank is obtained by sorting and ranking the observation data of the same pixel at each moment, and then sorting and ranking the ranks of each pixel at the same moment. The principal component weighted projection value refers to calculating the projection value of the pixel time series on the principal component, and then weighting the projection value using the eigenvalue of the principal component analysis to obtain the principal component weighted projection value of the pixel time series.

2. The method according to claim 1, characterized in that The specific steps include: Step 1: Select a pixel in the image foreground or the range selected by the researcher as the initial peak of the ROI. The initial range of the ROI only includes this pixel. Step 2: Expand the ROI by one pixel in each direction, then calculate the bidirectional rank of all pixel timings within the ROI, and then calculate the Kendall harmony coefficient W of the bidirectional rank, using W to represent the consistency of the pixel timing. Step 3: 3.1: If W is greater than or equal to the threshold W_th, repeat step 2 until W is less than the threshold W_th, then execute step 3.2; 3.2: If W is less than the threshold W_th, the range of the ROI is reduced by one pixel along the expansion direction, and the reduced range is used as the range of the ROI; Step 4: Perform principal component analysis on all pixel time series within the ROI range, calculate the weighted projection value of the principal component of each pixel time series, and take the pixel with the largest weighted projection value as the peak point of the ROI; Step 5: Perform steps 1 to 4 for each pixel in the image foreground range or the range selected by the researcher until all pixels are traversed; Step 6: For multiple ROIs with the same peak point, merge them into one ROI; if the peak point of each merged ROI is the same, the peak point is the peak point of the merged ROI; the range of the merged ROI is equal to the union of the ranges of each merged ROI; Step 7: ROIs with different peak points are not merged, and their respective ranges and peak points are retained; or For ROIs with different peak points, the following steps are used to merge them: 7.1: Perform agglomerative hierarchical clustering on all ROIs based on the distance between ROI peaks; 7.2: Based on the clustering results, calculate the maximum intra-class distance D_max for each class, then prune the clustering results, retaining only the classes whose D_max is less than or equal to the threshold D_th; Based on the pruning results, calculate the intra-class Kendall concordance coefficient Winc for each class, then prune the pruning results again, retaining only the classes whose Winc is greater than or equal to the threshold Winc_th; Alternatively, based on the clustering results, the Kendall harmony coefficient Winc is calculated for each class, and then the clustering results are pruned to retain only the classes whose Winc is greater than or equal to the threshold Winc_th; based on the pruning results, the maximum distance D_max within the class is calculated for each class, and then the pruning results are pruned again to retain only the classes whose D_max is less than or equal to the threshold D_th; Alternatively, based on the clustering results, the maximum intra-class distance D_max is calculated for each class, and then the clustering results are pruned to retain only the classes whose D_max is less than or equal to the threshold D_th; Alternatively, based on the clustering results, the Kendall harmony coefficient Winc within each class is calculated, and then the clustering results are pruned to retain only the classes whose Winc is greater than or equal to the threshold Winc_th; 7.3: Merge the ROIs according to the pruning results. The range of the merged ROI is equal to the union of the ranges of its corresponding merged ROIs. 7.4: For each merged ROI, re-perform principal component analysis on all pixel time series within its range, calculate the weighted projection value of the principal component of each pixel time series, and take the pixel with the largest weighted projection value as the peak point of the ROI.

3. The method according to claim 2, characterized in that The steps 1 to 4 are executed for each pixel in a manner of sequentially executing the steps pixel by pixel, or in a manner of executing the steps in parallel on multiple pixels.

4. An application of the method according to claim 1 or 2 in the calculation and analysis of neural activity synchronization based on medical images or neuroimaging.

Images