Brain function image signal enhancement method based on spatial features

Abstract

The invention discloses a brain function image signal enhancement method based on spatial features, and belongs to the technical field of medical image processing. The method comprises the following steps: dividing data into a plurality of time slices through sliding window segmentation, and extracting spatial features and time features from each slice by adopting singular value decomposition; signal enhancement is carried out on voxel, cluster and feature dimensions, and enhanced space and time features are combined by adopting a weighted average method, so that enhanced fMRI data are generated. According to the method, the spatial features of each time slice are finely processed, so that signals related to neural activity are more accurately improved; signal features of all dimensions are enhanced, a cluster-like enhancement method is introduced, and the stability and reliability of signals are improved while space details are kept. The accuracy and reproducibility of a brain network are improved in resting state data processing; the statistical confidence coefficient of the activation graph is enhanced in task state data processing, the scanning time is shortened on the premise that the graph quality is guaranteed, and the method has wide application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method for enhancing brain function image signals based on spatial features. Background Art

[0002] Functional Magnetic Resonance Imaging (fMRI) provides valuable data resources for brain science research with its high temporal and spatial resolution, as well as its advantages of non-invasiveness and radiation-free. It has now become the main technical means to study brain cognitive functions. The neurophysiological basis of fMRI technology is that when the activity of a certain area of ​​the brain increases, the proportion of deoxyhemoglobin in the area decreases, resulting in an enhancement of the blood oxygen level dependent (BOLD) signal, that is, an increase in the magnetic resonance signal. Therefore, fMRI is an indirect imaging by detecting physiological reactions such as blood flow changes caused by neural activity, rather than directly measuring neural activity. However, the signals related to neuronal activity in fMRI data are usually very weak, accounting for only 1%-5% of the total signal. In addition, fMRI data are easily affected by a variety of noises, including machine thermal noise, subject head movement, and physiological noise. The nature of these noises changes over time, and the fluctuation of the noise is similar to the amplitude of the change of brain function signals, making it more difficult to effectively separate the signal from the noise. In summary, the low sensitivity and noise interference of brain function signals detected by fMRI technology limit the signal-to-noise ratio of fMRI data and increase the complexity of data processing and analysis. These challenges emphasize the importance of signal enhancement in fMRI data analysis, which helps to improve the sensitivity and accuracy of signals and reduce noise interference, thereby more clearly identifying and analyzing brain activity.

[0003] High-field fMRI provides higher BOLD signal specificity and sensitivity, but because more slices and phase encoding steps are required to achieve high spatial resolution, this limits the temporal resolution; in addition, high field strength may bring more serious noise interference, affecting the improvement of signal-to-noise ratio. The Brain Networks Enhancement Model (BNEM) proposed by Nizhuan Wang et al. suppresses noise through a 3D Wavelet Noise Filter (3DWNF) and uses a Spatial Reproducibility Enhancement Algorithm (SREA) to improve the reproducibility of statistical parameter maps, thereby enhancing brain functional signals; although this method improves spatial repeatability, it may exclude the detection of certain functional networks on downsampled data sets with insufficient time points; in addition, the choice of wavelet basis may limit the applicability of this method to different types of fMRI data. The information-assisted dictionary learning method (IADL) proposed by Manuel Morante et al. significantly outperforms traditional methods in constructing design matrices, can generate activation clusters for more participants, and is more anatomically accurate and reliable; however, the sensitivity of this method in detecting task-related sources is still limited, and may reduce the sensitivity to some smaller or weaker task-related signals. The locally low-rank method (LLR) proposed by Nolan K Meyer et al. denoises the raw image data before fMRI post-processing, improving the statistical confidence of task-related activation maps; however, this method only produces the best results when performing the denoising step on complex-valued fMRI images, and may not work well for real-valued images. Although these technologies have made some progress in improving the signal-to-noise ratio of fMRI data and enhancing brain function signals, there are still some technical and application challenges that require further research and improvement. Summary of the invention

[0004] In view of the problems existing in the existing fMRI data signal enhancement technology, this paper proposes a brain function imaging signal enhancement method based on spatial features, which aims to improve the quality of individual fMRI data by specifically enhancing brain function signals. This method can significantly improve the accuracy and reproducibility of brain networks in resting state data; in task state data, it can enhance the statistical confidence of task-based activation maps, while shortening the scanning time while maintaining the quality of the activation map.

[0005] To solve the above problems, the present invention adopts the following technical solutions:

[0006] A brain function imaging signal enhancement method based on spatial features, comprising the following steps:

[0007] Beneficial effects of the present invention: The present invention proposes a method for enhancing brain function imaging signals based on spatial features. Compared with the traditional global signal enhancement method, the present invention can more accurately enhance the signals related to neural activity by performing fine processing on the spatial features of each fragment. Compared with the processing method that only relies on a single dimension of information, the present invention comprehensively uses multiple processing steps of voxels, clusters and feature dimensions to enhance signal features of different dimensions, and introduces a cluster-like enhancement method. While ensuring spatial details, the present invention improves the stability and reliability of the signal, effectively promotes the accurate identification and analysis of functional networks, and has strong practicality and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a flowchart of a brain function imaging signal enhancement method and system based on spatial features according to an embodiment of the present invention.

[0009] Figure 2 : This is a diagram of signal enhancement effects under different smoothing kernel sizes in an embodiment of the present invention, where a is a group analysis activation map obtained by different processing flows, b is a FWHM result map, c is a tSNR result map, and d is an activation consistency analysis map.

[0010] Figure 3 1 is a diagram of the signal enhancement effect at different scanning times in an embodiment of the present invention, wherein a is a group analysis activation spectrum obtained by different processing flows, b is a FWHM result diagram, c is a tSNR result diagram, and d is an activation consistency analysis diagram. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] like Figure 1 As shown, the embodiment of the present invention provides a method for enhancing brain functional imaging signals based on spatial features, which aims to improve the signal-to-noise ratio of fMRI data through multi-dimensional processing and enhance signals related to neural activity. This embodiment is based on 3T language task state data in the HCP public database. The specific steps are as follows:

[0013] Step 1, first, adopt sliding window method, based on the stimulation mode of language task, implement sliding window segmentation process to original fMRI data, the purpose is to divide the whole time series data into multiple smaller time segments, so as to facilitate the extraction and analysis of local signal features. In sliding window process, select a fixed size window and slide on the whole time series with a certain step length. The data in each window is considered to be an independent time period, and the size of the window can be adjusted according to task requirements and signal characteristics. In this way, the whole time series is divided into several small time segments, which helps to capture the brain function activities that change in time. The sliding window size of the present embodiment is set to 70 time points, and the step length is 20 time points.

[0014] Step 2: Convert each time segment into a two-dimensional space-time matrix, and use singular value decomposition (SVD) to decompose the data of each time segment. SVD is a matrix decomposition method that decomposes an original matrix into the product of three matrices, namely, a spatial feature matrix, a temporal feature matrix, and a diagonal matrix. Specifically, for an fMRI data matrix Y containing spatial and temporal information, SVD decomposes it into:

[0015] Y=USV T

[0016] Among them, U is the spatial feature matrix, which represents the characteristics of the data in the spatial domain; S is a diagonal matrix containing singular values, which reflect the strength of each feature in the data; V is the temporal feature matrix, which represents the characteristics of the data in the temporal domain. In this way, SVD can effectively extract the main spatial and temporal features from the data.

[0017] Step 3: After SVD decomposition, the next step is to enhance the spatial feature vector of each segment, especially the components related to neural activity. In this process, the goal is to identify and enhance those signal components related to brain activity from the spatial features, while suppressing noise or components unrelated to brain activity. Signal enhancement based on spatial features mainly includes three aspects: voxel dimension, cluster dimension and feature dimension:

[0018] Step 301: voxel dimension processing.

[0019] Spatial reconstruction and standardization: First, the spatial feature vector of each time segment is reconstructed into a three-dimensional spatial distribution map; then, Z-score standardization is used to transform the values ​​of all voxels into a standard normal distribution (Gaussian distribution); this operation helps to eliminate outliers (such as noise and outliers) in the original data, so that the data conforms to the Gaussian distribution assumption.

[0020] Threshold setting and binarization: Based on the spatial map after Z-score standardization, significant voxels are screened out by setting a threshold; usually, when the Z-score value of a voxel exceeds a certain set threshold, the voxel is considered valid, and these valid voxels are converted into a binary map; the threshold setting can be adjusted based on statistical characteristics (such as P value, the threshold value p<0.05 is set in this embodiment) to ensure that the selected voxels are statistically significant.

[0021] Voxel screening: Finally, based on the number of valid voxels, clusters with sufficient voxels in the spatial distribution map are screened out; through the label connected component algorithm, under the condition of a connection standard of 18, clusters with the number of voxels that do not meet the standard (less than 50 valid voxels) are removed to ensure the stability and reliability of the results.

[0022] Step 302: cluster dimension processing.

[0023] Mild smoothing: Apply a small-sized Gaussian smoothing to the purified signal clusters, use the smoothing function in SPM and set the smoothing kernel parameter (Full Width at Half Maximum, FWHM) to a voxel size to balance spatial resolution and signal enhancement, and avoid over-smoothing leading to false activation signals. Due to the influence of thermal noise, the signal-to-noise ratio of the original data is low, so a small-sized Gaussian smoothing kernel is applied to smooth the purified signal clusters; the goal of this step is to improve the signal-to-noise ratio without sacrificing spatial details. Through smoothing, the effective signal area will be more prominent and the noise part will be weakened, thereby improving the signal strength of the cluster.

[0024] Cluster-like enhancement: After light smoothing, the signal is further enhanced using the Threshold-free Cluster Enhancement (TFCE) method. The key to the cluster-like enhancement method is to weight the clusters according to their strength and size to reduce noise interference and enhance the true brain function signal; unlike traditional smoothing operations, cluster-like enhancement technology enhances the minimum and maximum values ​​in the local area. These extreme values ​​are usually located in the same position, avoiding the displacement of the extreme value position during the smoothing process; therefore, this method can effectively remove thermal noise and scattered noise, while enhancing brain function signals, avoiding false activation signals that may be introduced by over-smoothing; the enhanced eigenvalue is given by the following formula:

[0025]

[0026] Among them, TFCE represents the enhanced eigenvalue, e(h) represents the number of voxels in the cluster, h is the strength of the cluster, E is the index of cluster expansion (increasing E can make spatially larger clusters obtain higher weights), H is the index of cluster strength (increasing H can make clusters with higher strength obtain greater weights), and dh represents the threshold step size; by adjusting the values ​​of E and H, the clusters can be adjusted according to the needs of different experiments to enhance the signals related to neural activity. In this embodiment, E is set to 0.5 and H is set to 2. By identifying high-intensity voxels, classifying clusters according to connectivity, and then reconstructing according to cluster size and strength, scattered noise is reduced and the strength of brain function signals is significantly improved.

[0027] Step 303: feature dimension processing.

[0028] Correlation analysis between time segments: Considering that in fMRI data, the spatial distribution of different time segments usually has a high degree of commonality; therefore, by calculating the correlation of spatial feature vectors between different time segments, a graph connection matrix is ​​constructed to quantify the spatial feature similarity between each time segment.

[0029] Screening of unstable spatial distributions: Use a classification method (this embodiment uses the K-nearest neighbor classification method) to automatically estimate the appropriate threshold and screen out those spatial distributions with low correlation in the connection graph; the purpose of this step is to remove those components that are unstable between different time segments, which may be caused by noise or inconsistent neural activities; by screening and retaining features that are consistently present in multiple time segments, the stability and reliability of the data can be effectively enhanced, thereby ensuring that signals related to neural activity are more prominent, and ultimately improving the accuracy of functional network identification.

[0030] Step 4: Signal reconstruction and weighted averaging. The enhanced spatial feature vectors and the corresponding temporal feature vectors will be combined through reconstruction and weighted averaging to generate enhanced fMRI data; due to the overlap of sliding windows, some time points will be used multiple times, and their repetition frequency will be used as a weight for weighted averaging to generate enhanced fMRI data for subsequent analysis.

[0031] Figure 2 The ad in the figure evaluated the differences between the signal enhancement method and the standard process by comparing the effects of mild smoothing processing with different smoothing kernel sizes (0.5 times, 1 times, 1.5 times, and 2 times the voxel size), involving activation maps, smoothness (FWHM), temporal signal-to-noise ratio (tSNR), and activation consistency. Figure 2 a in the figure compares the group analysis activation maps obtained by different processing procedures; Figure 2The FWHM results in b show that the smoothing effect of the signal enhancement method in the feature space is slightly lower than that of the standard processing in the voxel space; Figure 2 The tSNR results in c show that the tSNR of the standard treatment gradually increases from a low level, while the signal enhancement method maintains a high level. Although the tSNR of the standard treatment may exceed that of the signal enhancement method at a high smoothing level, its activation effect is not good, which may be due to signal distortion caused by excessive smoothing; Figure 2 The activation consistency analysis of d in shows that the correlation of both methods increases with the increase of smoothing kernel size, but the correlation of standard processing is always lower than that of signal enhancement method, which indicates that signal enhancement method improves the intensity of individual-specific activation and enhances the consistency of individual activation map with group activation results. Overall, signal enhancement method performs best under mild smoothing of 1 times voxel size, which not only significantly improves the intensity and range of activation signal, but also maintains high spatial resolution, avoiding signal blurring and false activation caused by over-smoothing.

[0032] Figure 3 The ad in the figure mainly shows the activation maps, FWHM, tSNR and activation consistency of the unprocessed, standard processed and signal enhanced method processed at different scanning times (half scanning time and full scanning time). Figure 3 Figure a shows the group analysis activation maps obtained by different processing procedures, indicating that the signal enhancement method shows superior effects in the comparison of different scanning times, especially when the scanning time is shortened to half, the presentation quality of conditional activation is better than the full scanning time results obtained by the standard processing procedure. Figure 3 The FWHM results in (b) show that the signal enhancement method is slightly less smooth than the standard processing, while Figure 3 The tSNR results in c show that the tSNR of the signal enhancement method is slightly higher than that of the standard processing; Figure 3 The activation consistency analysis of d in further showed that the signal enhancement method can significantly improve the consistency of individual-specific activation regardless of the scanning time. Overall, the signal enhancement method can not only significantly improve the quality of the activation map, but also effectively shorten the necessary scanning time.

[0033] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A brain function imaging signal enhancement method based on spatial features, characterized in that: The following steps are involved: Step 1: Segment the original fMRI data by means of a sliding window to divide the data into multiple time segments; Step 2: For each time segment of data, singular value decomposition technology is used to extract the spatial feature matrix and the temporal feature matrix; Step 3: Targetedly enhance the spatial feature vector of each segment, including: Step 3.1, voxel dimension processing: First, the spatial feature vector of each time segment is reconstructed into a three-dimensional spatial distribution map, and Z-score standardization is performed to convert the voxel value into a standard normal distribution. Then, a threshold is set based on the standardized spatial map to screen out significant voxels, and a binary map is generated for the significant voxels. Finally, the label connected component algorithm is used to identify valid clusters, and clusters whose voxel numbers do not meet the standards are removed. Step 3.2: Cluster dimension processing: Firstly, a slight smoothing is applied to the spatial feature vector after voxel dimension processing to improve the signal-to-noise ratio; then the signal is further enhanced by cluster-like enhancement method; Step 3.3, feature dimension processing: After processing the cluster dimension, the spatial feature vectors are analyzed to analyze the correlation of spatial features in different time segments, and to screen out low-correlation and unstable spatial distributions. Step 4: Signal reconstruction and weighted averaging: The enhanced spatial features and the temporal features extracted in step 2 are combined by weighted averaging to generate enhanced fMRI data.

2. The method according to claim 1, characterized in that The window size and step size of the sliding window are set according to the data characteristics.

3. The method according to claim 1, characterized in that The threshold was set at P < 0.05 to screen significant voxels; when the connectivity criterion was 18, clusters with less than 50 valid voxels were removed.

4. The method according to claim 1, characterized in that: In the light smoothing step, the smoothing function in SPM is used and the smoothing kernel parameter is set to one voxel size.

5. The method according to claim 1, characterized in that In the cluster enhancement step, a threshold-free cluster enhancement method is used. The enhanced eigenvalue is given by the following formula: Among them, TFCE represents the enhanced eigenvalue, e(h) represents the number of voxels in the cluster, h is the strength of the cluster, E is the index of cluster expansion, H is the index of cluster strength, and dh represents the threshold step size; by adjusting the values ​​of E and H, the clusters can be adjusted according to the needs of different experiments to enhance the signals related to neural activity.

6. The method according to claim 1, characterized in that The feature dimension processing is specifically as follows: by calculating the correlation of spatial feature vectors between different time segments, a graph theory connection matrix is ​​constructed to quantify the spatial feature similarity between each time segment, and using classification methods to automatically estimate the appropriate threshold, screen out those spatial distributions with low correlation in the connection graph, and remove components that are unstable between different time segments.

Images