A method for calculating polarity jump of fMRI spatial source phase
Patent Information
- Application Number
- CN202611059255.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-16
- Publication Date
- 2026-10-09
AI Technical Summary
但是，现有关于扫描相位的极性跳变研究仅限于简单的定性分析，缺乏定量的计算方法，难以推广到fMRI空间源相位以挖掘全新的脑功能信息
[0018]再次，当采用疾病标志物脑区细粒度体素块的三方向源相位极性跳变平均次数作为SVM分类特征时，分类准确率较之采用源幅值平均值特征可提高6%～10%和5%～9%（分别对应扫描相位和希尔伯特变换生成相位）。当将两种特征联合使用时，分类准确率的提升可达到9%～11%和7%～10%（见图4）。
Smart Images
Figure CN122885633A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of signal processing technology and relates to a method for calculating the polarity jump of the spatial source phase in functional magnetic resonance imaging (fMRI). Background Technology
[0002] fMRI is an advanced brain functional imaging technique with outstanding advantages such as safety, non-invasiveness, and high spatial resolution. It has been widely used in brain function research and the development of disease biomarkers and pathogenesis studies for neuropsychiatric brain diseases (such as schizophrenia and depression).
[0003] Raw fMRI data obtained from MRI scanners are in complex form, including amplitude and phase data. Existing research indicates that phase data contains unique brain functional information not found in amplitude data, revealing brain functional patterns from different perspectives. Combining fMRI phase and amplitude data to form complex data can significantly improve the detection of spatial brain activation components and brain functional changes associated with neuropsychiatric brain disorders. For example, spatial brain activation components become more complete, making it more sensitive to brain functional abnormalities caused by neuropsychiatric brain disorders.
[0004] Currently, the default image reconstruction algorithms of MRI scanners focus primarily on the quality of fMRI amplitude images, neglecting the quality of phase images. This results in persistent phase distortion in fMRI phase data, rendering it unusable in subsequent data analysis. Consequently, many publicly available fMRI datasets store only amplitude data, discarding phase data, which significantly limits the performance and potential of fMRI analysis. A promising development is a method that uses Hilbert transform to generate phase data from fMRI amplitude data. This method provides phase data that approximates the phase data acquired by the MRI scanner (referred to as scan phase), thereby significantly improving the performance of fMRI analysis (A method for improving fMRI analysis performance using phase generated by Hilbert transform, CN121661196A, 2026.3.13). By applying the Hilbert transform method to generate phase data, existing fMRI data analysis can be upgraded from amplitude data analysis to complex data analysis, thereby comprehensively improving brain function research and the development of biomarkers and pathogenesis studies for neuropsychiatric diseases.
[0005] fMRI spatial source phase mapping uses data-driven blind source separation methods, such as independent component analysis (ICA), to extract phase maps of complex brain activation components from fMRI complex data. Compared to MRI scan phase mapping or Hilbert transform-generated phase mapping, fMRI spatial source phase mapping undergoes effective noise reduction through blind source separation methods, more closely resembling real neural activity, thus providing more sensitive voxel-level fine features for brain disease research.
[0006] In fact, the spatial source phase of fMRI contains two aspects of information: phase magnitude and polarity. However, existing studies mainly use the magnitude of the spatial source phase to denoise brain activation components, and research on its polarity remains lacking. Studies based on MRI scan phase have detected polarity jumps in the phase near activation peaks, suggesting that polarity jumps contain brain functional information. However, existing studies on polarity jumps in scan phases are limited to simple qualitative analysis, lacking quantitative calculation methods, making it difficult to generalize to fMRI spatial source phase to uncover novel brain functional information. Summary of the Invention
[0007] This invention provides a method for calculating the polarity jump of fMRI spatial source phase, which can quantitatively calculate the phase jump along the entire region or a specified sub-region of activated components in three-dimensional brain space. , or The number and average number of phase polarity jumps between adjacent voxels in a single direction, and along... , , The total or average number of phase polarity jumps between adjacent voxels in three directions provides a basic quantitative method for mining polarity jump information.
[0008] The technical solution adopted in this invention is as follows: A method for calculating the polarity jump of the phase of an fMRI spatial source includes the following steps (see...). Figure 1 ): Step 1: Constructing fMRI complex data make This represents fMRI amplitude data. This represents phase data (generated from an MRI scanner or Hilbert transform). It is the total number of primes in the brain; This refers to the number of time points, i.e., the number of scans of the whole brain. Amplitude data is used. With phase data Constructing fMRI complex data as follows: Step 2: Extracting brain-activating components of interest in complex spatial domains Using data-driven blind source separation methods, such as the entropy bound minimization (EBM) algorithm for complex ICA, from complex fMRI data... Extract Individual brain activation components ( Spatial reference components (such as the template from the paper Smith SM, Fox PT, Miller KL, Glahn DC, Fox PM, Mackay CE, Filippini N, Watkins KE, Toro R, Laird AR, Beckmann CF. Correspondence of the brain's functional architecture during activation and rest, Proceedings of the National Academy of Sciences, vol.106, no.31, pp.13040-13045, 2009, referred to as the Smith template) are selected for the components of interest (such as the default mode network, DMN). Pearson correlation calculations are then performed to... The component of interest with the highest correlation coefficient with the spatial reference component was selected from the components. Phase correction and denoising were performed using the method described in the paper "Yu MC, Lin QH, Kuang LD, Gong XF, Cong F, Calhoun VD. ICA of full complex-valued fMRI data using phase information of spatial maps, Journal of Neuroscience Methods, vol. 249, pp. 75-91, 2015" to obtain the denoised complex spatial brain activation components.
[0009] Step 3: Calculate the spatial source amplitude map and spatial source phase map. Based on the complex spatial brain activation components of interest obtained in the second step, their magnitude and phase angle are calculated to obtain the spatial source amplitude map and spatial source phase map of the component, respectively.
[0010] Step 4: Calculate the number of source phase polarity jumps within the region using three methods. Select the entire region or sub-region of the component of interest, denoted as For example, the anterior cingulate cortex (ACC) of the DMN component is represented by its source phase diagram as follows: Among them, the region .
[0011] The first method is to calculate the number of source phase polarity jumps in a single direction within the region. Taking direction as an example, the number of phase polarity jumps between adjacent voxels is calculated as follows: In the formula, Indicates along Single phase sequence in direction In the middle, the number of polarity jumps between adjacent voxels: In the formula, For indicator functions, when If the phase signs of adjacent voxels in the directional phase sequence are opposite, the condition in parentheses is true and the value is 1; otherwise, it is 0.
[0012] Similarly, it is possible to calculate along or Number of phase polarity jumps between adjacent voxels in one direction , .
[0013] When only targeting a region When analyzing a single voxel point (such as an activation peak point), only one phase sequence exists along a single direction. , , Each has only one value, which can be used to further calculate the total number of polarity jumps in the three directions. ).
[0014] The second method is to calculate the average number of source phase polarity jumps in a single direction within the region. Because polarity jumps occur relatively sparsely, to avoid the influence of numerous zero-polarity changes on the results, the region... In each direction, only the phase sequence results with non-zero polarity transitions are averaged; Taking direction as an example, calculate non-zero values. The number is as follows: Based on this, define The average number of source phase polarity jumps in the direction is as follows: In the formula, max It is a maximum value operation.
[0015] Similarly, it is possible to calculate along or Average number of phase polarity jumps between adjacent voxels in one direction , .
[0016] The third method is to calculate the total or average number of source phase polarity jumps in the three directions. This is done within the separately calculated regions. Inside , , The average number of phase polarity jumps between adjacent voxels in three directions is obtained , and Then, the total number of source phase polarity jumps in the three directions can be calculated as follows: Alternatively, the average number of source phase polarity jumps in the three directions can be calculated as follows: This invention provides a method for calculating the polarity jump of spatial source phase in fMRI, which can be used to reveal the brain functional information contained in the polarity jump, providing a panoramic view and spatial localization information of source amplitude-source phase for brain functional abnormalities, providing new evidence of abnormal blood oxygenation levels for disease mechanism research, and providing more sensitive new features for disease classification. Using this method, firstly, a stable difference in the number of polarity jumps in the brain regions of disease markers can be found between patients and healthy controls. For example, for the activation peak in the ACC brain region and its six surrounding submaximal values, the total number of source phase polarity jumps in three directions can be calculated respectively. The levels of healthy controls (HC) were consistently higher than those of schizophrenia patients (SZ). (See...) Figure 2 This indicates that polarity shifts contain specific information about brain function.
[0017] Secondly, by combining the spatial source-phase polarity jumps in the brain regions containing disease biomarkers with the spatial source amplitude, a panoramic image of source amplitude-source phase can be constructed, especially a detailed image of the source amplitude-source phase polarity jumps near activation extrema (see...). Figure 3 Under the condition that the magnitude of activation corresponds to the level of blood oxygenation, the number of source phase polarity jumps in the panoramic image can provide abnormal change patterns of blood oxygenation levels for disease mechanism research, such as insufficient increase of blood oxygenation level in the left auditory network (L-AUD) in SZ; moreover, the polarity jump position provides precise voxel-level localization.
[0018] Furthermore, when the average number of three-directional source phase polarity jumps in fine-grained voxel blocks of disease marker brain regions is used as a classification feature for SVM, the classification accuracy can be improved by 6%–10% and 5%–9% compared to using the source amplitude average feature (corresponding to the scan phase and the Hilbert transform-generated phase, respectively). When the two features are used in combination, the improvement in classification accuracy can reach 9%–11% and 7%–10% (see...). Figure 4 ). Attached Figure Description
[0019] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 The source phase polarity jump map of the activation peak and its neighborhood in the ACC region for an HC subject and an SZ subject; Figure 2 In the diagram, (a) is the source phase polarity diagram, where red indicates a positive polarity source phase and blue indicates a negative polarity source phase. (b) is the amplitude curve and phase polarity jump, where (1) is the activation peak value and (2)-(7) are the six activation sub-maximum values. The red arrow indicates that the polarity changes from negative to positive in the direction toward the peak value, and the blue arrow indicates that the polarity changes from positive to negative in the direction toward the peak value. Figure 3 For an HC subject, the activation extreme point along the left auditory region is... Detailed plot of source amplitude-source phase polarity jump in direction; Figure 4 The SVM classification results for 82 subjects (40 HC and 42 SZ) based on fine-grained voxel block features of three brain regions were presented. The classification features were the average number of source phase polarity jumps and the average source amplitude in three directions. The three brain regions were ACC, L-AUD, and right executive control network (R-ECN). Detailed Implementation
[0020] The following describes two embodiments of the present invention in detail with reference to the technical solutions.
[0021] Example 1: fMRI amplitude data and MRI scan phase data of a HC subject and an SZ subject are known, and have been preprocessed with head motion correction, spatial normalization and spatial smoothing; phase jump analysis is performed near the peak in sub-regions that are highly sensitive to schizophrenia.
[0022] Step 1: Construct fMRI complex data using formula (1) Step 2: Extracting brain-activating components of interest in complex spatial domains Using the complex EBM algorithm, 110 spatial brain activation components were extracted from the complex fMRI data of each subject. Next, using the Smith template of DMN and AUD components as spatial reference components, the two components of interest with the highest Pearson correlation coefficient with the spatial reference components were selected from 110 components. Phase correction and denoising were then performed to obtain two denoised complex spatial brain activation components.
[0023] Step 3: Calculate the spatial source amplitude map and spatial source phase map. Based on the complex spatial brain activation components of interest obtained in the second step, their magnitude and phase angle are calculated to obtain the spatial source amplitude map and spatial source phase map of the two components, respectively.
[0024] Step 4: Calculate the number of source phase polarity jumps within the region. From the component of interest (DMN), the ACC subregion, which is highly sensitive to schizophrenia, was selected. Using formulas (2) and (3), at the activation peak of the region and its six neighboring second-largest activation values, the activation values of an HC subject and an SZ subject were calculated along the lines of the region. , and Number of phase polarity jumps between adjacent voxels in three unidirectional directions , , Furthermore, the total number of polarity jumps in the three directions is calculated. ), the results are shown Figure 2 ; L-AUD subregions highly sensitive to schizophrenia were selected from the AUD components, and the activation peak of an HC subject along the region was calculated using formulas (2) and (3). Number of source phase polarity jumps in the direction ( ), and show a detailed plot of the source amplitude-source phase polarity jump (see Figure 3 This can provide evidence of abnormal blood oxygen levels.
[0025] Example 2: fMRI amplitude data, MRI scan phase data, and phase data generated from fMRI amplitude data by Hilbert transform were known for 40 HC subjects and 42 SZ subjects. The fMRI amplitude data and MRI scan phase data had been preprocessed with head motion correction, spatial normalization, and spatial smoothing. Fine-grained source phase polarity jump analysis was performed on sub-regions that are highly sensitive to schizophrenia, and then classified.
[0026] Step 1: Construct fMRI complex data using formula (1) Step 2: Extracting brain-activating components of interest in complex spatial domains Using the complex EBM algorithm, 110 spatial brain activation components were extracted from the complex fMRI data of each subject. Next, using Smith templates of DMN, AUD, and ECN as spatial reference components, the three components of interest with the highest Pearson correlation coefficient with the spatial reference components were selected from 110 components. Phase correction and noise reduction were then performed to obtain three complex spatial brain activation components for each subject.
[0027] Step 3: Calculate the spatial source amplitude map and spatial source phase map. Based on the complex spatial brain activation components of interest obtained in the second step, their magnitude and phase angle are calculated to obtain the spatial source amplitude map and spatial source phase map of the three components, respectively.
[0028] Step 4: Calculate the number of source phase polarity jumps within the region. Based on the sliding window technique, fine-grained voxel blocks of size 5×5×5 voxels are obtained from the region of interest. 3 The sliding step size is 2 voxels, that is ; For each voxel block, calculate the average source amplitude of 125 voxels. And the average number of source phase polarity jumps in three unidirectional directions.
[0029] The regions of interest were selected from three sub-regions—ACC, L-AUD, and R-ECN—that are highly sensitive to schizophrenia. Formulas (2) to (5) were used to calculate the region of interest for each subject along the three sub-regions. , and Average number of phase polarity jumps between adjacent voxels in three unidirectional directions , , Next, the average number of source phase polarity jumps in the three directions is calculated using formula (7). .
[0030] Step 5: SVM Classification For all fine-grained voxel block features of all HC and SZ subjects, including those calculated in step four... and the calculation in step five Single-feature and joint-feature classification were performed, and the results are shown below. Figure 4 ; Compared to The classification accuracy was improved by 6%–10% and 5%–9% (corresponding to the scanning phase and the Hilbert transform-generated phase, respectively). When the two were used together, the classification accuracy was improved by 9%–11% and 7%–10%, respectively.
Claims
1. A method for calculating the polarity jump of the phase of an fMRI spatial source, characterized in that... The following steps are required: Step 1: Constructing fMRI complex data make This represents fMRI amplitude data. This represents phase data, generated from an MRI scanner or Hilbert transform. It is the total number of primes in the brain; This refers to the number of time points, i.e., the number of scans of the whole brain; amplitude data is used. With phase data Constructing fMRI complex data as follows: Step 2: Extracting brain-activating components of interest in complex spatial domains Using a data-driven blind source separation method, from complex fMRI data Extract Individual brain activation components, ; Select the spatial reference component of the component of interest, and use Pearson correlation calculations to... The component of interest with the highest correlation coefficient with the spatial reference component was selected from the components. Phase correction and denoising were performed using the method described in the paper "Yu MC, Lin QH, Kuang LD, Gong XF, Cong F, Calhoun VD. ICA of full complex-valued fMRI data using phase information of spatial maps, Journal of Neuroscience Methods, vol. 249, pp.75-91, 2015" to obtain the denoised complex spatial brain activation components. Step 3: Calculate the spatial source amplitude map and spatial source phase map. Based on the complex spatial brain activation components of interest obtained in the second step, their magnitude and phase angle are calculated to obtain the spatial source amplitude map and spatial source phase map of the component, respectively. Step 4: Calculate the number of source phase polarity jumps within the region using three methods. Select the entire region or sub-region of the component of interest, denoted as Its source phase diagram is represented as Among them, the region ； The first method is to calculate the number of source phase polarity jumps in a single direction within the region; Taking direction as an example, the number of phase polarity jumps between adjacent voxels is calculated as follows: In the formula, Indicates along Single phase sequence in direction In the middle, the number of polarity jumps between adjacent voxels: In the formula, For indicator functions, when If the phase signs of adjacent voxels in the directional phase sequence are opposite, the condition in parentheses is true and the value is 1; otherwise, it is 0. Similarly, it is possible to calculate along or Number of phase polarity jumps between adjacent voxels in one direction , ; When only targeting a region When analyzing a single voxel, only one phase sequence exists along a single direction. , , Each has only one value, which can be used to further calculate the total number of polarity jumps in the three directions. ); The second method is to calculate the average number of source phase polarity jumps in a single direction within the region; Because polarity jumps occur relatively sparsely, to avoid the influence of numerous zero-polarity changes on the results, in the region... In each direction, only the phase sequence results with non-zero polarity transitions are averaged; Taking direction as an example, calculate non-zero values. The number is as follows: Based on this, define The average number of source phase polarity jumps in the direction is as follows: In the formula, max This is a maximum value operation; Similarly, it is possible to calculate along or Average number of phase polarity jumps between adjacent voxels in one direction , ; The third method is to calculate the total or average number of source phase polarity jumps in the three directions; in the calculation regions respectively Inside , , The average number of phase polarity jumps between adjacent voxels in three directions is obtained , and Then, the total number of source phase polarity jumps in the three directions can be calculated as follows: Alternatively, the average number of source phase polarity jumps in the three directions can be calculated as follows: 。