Processing method for resting-state magnetic resonance functional imaging based on multi-scale information fusion
By using Fuse-Former layers and multi-scale information fusion technology, the problems of low disease recognition rate and insufficient high-order information interaction in resting-state magnetic resonance imaging have been solved, achieving higher disease prediction accuracy, especially in the accurate prediction of Alzheimer's disease and autism spectrum disorder.
Patent Information
- Application Number
- CN202410983426.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-26
AI Technical Summary
Current resting-state magnetic resonance functional imaging technology has low disease recognition rate, neglects high-order information interaction, has high model complexity, and lacks short-term window feature mining, resulting in low disease prediction accuracy.
A multi-scale information fusion-based processing method is adopted. The BOLD signal is operated on through the Fuse-Former layer. Combined with global and local feature extraction modules, global and local classification labels are extracted using multi-head self-attention and label fusion technology for disease prediction.
It improved the accuracy of disease prediction and enhanced disease identification performance, particularly in predicting Alzheimer's disease and autism spectrum disorders.
Smart Images

Figure CN119027370A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of neuroimaging technology, and in particular to a processing method for resting-state functional magnetic resonance imaging based on multi-scale information fusion. Background Art
[0002] Functional Magnetic Resonance Imaging (fMRI) is the most commonly used technology in neuroimaging research. It measures brain Blood Oxygenation Level Dependent based on the inconsistency between local oxygen consumption and blood flow increase after neuronal excitation, that is, indirectly reflects the functional synergistic activities of spatially isolated brain regions through BOLD signals. Among them, Resting-State Functional Magnetic Resonance Imaging (rs-fMRI), that is, resting-state functional magnetic resonance imaging, refers to brain imaging scans performed when the subjects are in a relaxed and resting state. It can reflect the most basic and essential state of the brain functional network, and brain lesions can cause abnormal low-frequency fluctuations in the BOLD signal, which in turn leads to abnormal brain functional networks. Therefore, using rs-fMRI to establish an accurate brain disease recognition model has become a feasible solution.
[0003] In the existing technology, deep learning technology has greatly promoted the progress of existing rs-fMRI analysis research, but it still faces the following problems. First, the recognition performance directly affects whether the research can be applied in actual scenarios. Compared with neuroimaging technologies such as structural magnetic resonance imaging (sfMRI) and positron emission tomography (PET), the disease recognition model based on rs-fMRI usually has a lower recognition rate. Second, the functional connectivity (FC) features generated by the BOLD signal can usually only reflect low-order information interactions, ignoring the high-order effects reflected by BOLD. Third, constructing graph data will increase the complexity of model development, and may lose some detailed information, affecting the prediction performance of the model. Fourth, studies have shown that functional connectivity may show dynamic changes in a short period of time. Current research lacks the discovery of key short-term window features. In short, the disease prediction accuracy of existing technologies is not high. Summary of the invention
[0004] The primary purpose of the present invention is to overcome the problems existing in the prior art and provide a processing method for resting-state functional magnetic resonance imaging based on multi-scale information fusion. The present invention can improve the accuracy of disease prediction.
[0005] As another object of the present invention, a non-volatile storage medium suitable for storing a computer program implemented according to the method described is provided.
[0006] In order to achieve the above object, the present invention provides a processing method for resting state functional magnetic resonance imaging based on multi-scale information fusion, the method comprising the following steps: Acquire resting-state functional magnetic resonance imaging and perform preprocessing; An input set is obtained based on preprocessed resting-state functional magnetic resonance imaging; The input set is input into an image processing model to obtain a processing result, wherein the image processing model includes a plurality of sequentially connected Fuse-Former layers, aggregation layers, linear layers and softmax activation functions, wherein the plurality of sequentially connected Fuse-Former layers are used to obtain global classification labels and local classification labels according to the input set; the aggregation layer is used to perform aggregation processing on the global classification labels and the local classification labels to obtain aggregated labels; the linear layer is used to perform feature conversion processing on the aggregated labels to obtain an output vector; and the softmax activation function obtains an image processing result according to the output vector.
[0007] Furthermore, the method of obtaining an input set based on the preprocessed resting-state functional magnetic resonance imaging includes: dividing the resting-state functional magnetic resonance imaging into multiple regions of interest using an external atlas; extracting the blood oxygen level dependent response of each region of interest; and processing the blood oxygen level dependent response to obtain an input set.
[0008] Furthermore, the processing of the blood oxygen level dependent response to obtain an input set specifically includes: Obtaining a response average of each voxel in the region of interest, performing z-score normalization processing of zero mean and unit variance on the response average to obtain a time series, wherein the time series includes multiple time indexes; The responses measured at each specific time index were transformed into blood oxygen level-dependent markers using linear projection; Merge a classification tag at each time index to obtain a merged tag; Add position code to the merge tag to obtain an input set, which is recorded as , which is determined by the following formula:
[0009] in, is a blood oxygen level dependent marker, For classification marks, Encode the position.
[0010] Further, a linear projection is used to transform the response measured at each specific time index into a blood oxygen level dependent marker, which is denoted as , which is determined by the following formula:
[0011] in, Determined by the following formula:
[0012] Among them, is the time index, is the linear projection function, is the encoding dimension.
[0013] Furthermore, each Fuse-Former layer includes a global feature extraction module and a local feature extraction module, wherein the global feature extraction module is used to obtain a global blood oxygen level dependency marker and a global classification marker, and the local feature extraction module is used to obtain a local blood oxygen level dependency marker and a local classification marker.
[0014] Furthermore, the global feature extraction module includes a first normalization layer, a fusion window multi-head self-attention layer, a label fusion layer, a second normalization layer and a multi-layer perceptron layer connected in sequence. The first normalization layer is used to perform normalization processing on the input set to obtain a normalized input set; The fusion window multi-head self-attention layer is used to divide the normalized input set into multiple continuous time windows; The label fusion layer is used to fuse the contextual blood oxygen level dependency labels of the continuous time window to obtain a fused label feature; The second normalization layer is used to normalize the fused marker features to obtain normalized fused marker features; The multi-layer perceptron layer is used to extract features from the normalized fusion mark features to obtain global blood oxygen level dependency marks and global classification marks, and the global blood oxygen level dependency marks and the global classification marks are used as inputs of the next Fuse-Former layer.
[0015] Furthermore, the local feature extraction module is used to obtain the local blood oxygen level dependency mark and the local classification mark, specifically including: Set a window with the same step size as the input set to traverse the input set; Starting from the second window, calculate the distribution difference between each window and the previous window; use The algorithm selects multiple windows with the largest difference in front and back distribution, and splices each selected window with the previous window of the selected window to obtain a fused window; Performing multi-head self-attention calculation on the fusion window to obtain a feature-enhanced fusion window, and fusing the feature-enhanced fusion windows to obtain a feature matrix; A local blood oxygen level dependency marker and a local classification marker are extracted from the feature matrix, and the local blood oxygen level dependency marker and the local classification marker are used as inputs of the next Fuse-Former layer.
[0016] Furthermore, the distribution difference is calculated as follows:
[0017] in, represents the i-th window.
[0018] A processing method for resting-state functional magnetic resonance imaging based on multi-scale information fusion is applied to predict brain diseases, and the obtained image processing results can predict brain diseases.
[0019] A computer-readable storage medium stores a computer program, and when the computer-stored program is executed by a processor, a processing method for resting-state magnetic resonance functional imaging based on multi-scale information fusion is implemented.
[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention directly operates the BOLD signal through the Fuse-Former layer, does not adopt a complex graph feature construction method, gets rid of the constraints of the graph neural network, and simply regresses to the time series feature recognition of multiple variables. At the same time, the Fuse-Former layer has the "global-local" feature perception ability of multi-scale information. The global feature extraction module interacts and fuses all window features by fusing window multi-head self-attention and tag fusion to obtain a global classification tag. The local feature extraction module selects the most different key time series features by comparing the feature distribution of each short-term window, obtains subtle change information in the resting state, and thus obtains local classification tags. Finally, the global classification tags and local classification tags are combined for disease prediction, which improves the accuracy of disease prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 A flowchart of a method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to an embodiment of the present invention; Figure 2 It is a flow chart of a processing method of resting state magnetic resonance functional imaging based on multi-scale information fusion according to an embodiment of the present invention; Figure 3 4 is a structural diagram of multiple cascaded Fuse-Former layers according to an embodiment of the present invention. DETAILED DESCRIPTION
[0022] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0023] In the description of the present invention, it should be noted that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.
[0024] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0025] Furthermore, in the description of the present invention, unless otherwise specified, “plurality” means two or more.
[0026] Example 1 like Figure 1-3 As shown, a processing method for resting state functional magnetic resonance imaging based on multi-scale information fusion in a preferred embodiment of the present invention includes: S1: Acquire resting-state functional magnetic resonance imaging and perform preprocessing; In one embodiment, during the image acquisition process, factors such as the subject's head movement, breathing, and heartbeat will generate noise, resulting in poor image quality. Therefore, when analyzing the data, preprocessing must be performed first to reduce the impact of irrelevant noise and improve the signal-to-noise ratio. This embodiment uses DPARSF software to implement the preprocessing process.
[0027] S2: obtain input set based on preprocessed resting-state functional magnetic resonance imaging; In one embodiment, S2 specifically includes: Since fMRI contains time series, for a given region of interest (ROI), the time series is considered as the average response of each voxel within the ROI and is z-score normalized to zero mean and unit variance; Subsequently, a learnable linear projection is used to transform the blood oxygen level dependent (BOLD) response measured at each specific time index into the input of the Fuse-Former, called a blood oxygen level dependent token (BOLDtoken). The BOLDtoken set can be expressed as:
[0028] We further aggregate a classification tag (CLS) token at each time index for subsequent classification tasks, and then add a learnable positional encoding to Fuse-Former. , to dynamically adjust and learn the position information representation.
[0029] It should be noted that for a given ROI, the time series is regarded as the average response of each voxel in the ROI and is normalized by z-score with zero mean and unit variance. At this time, the BOLD response can be expressed as , where T is the length of the time series in the fMRI scan and n represents the number of ROIs extracted. Subsequently, the BOLD response measured at each specific time index is converted into the input of the Fuse-Former layer, called BOLD token, using a learnable linear projection, that is:
[0030] in is the time index, is the linear projection function, is the encoding dimension. The BOLD token set can be expressed as:
[0031] Further concatenate a CLS token at each time index for subsequent classification tasks. Since the Transformer structure model lacks explicit position awareness when processing sequence data, a learnable position encoding is added to Fuse-Former. , to dynamically adjust and learn the position information representation. At this point, the input of Fuse-Former can be expressed as:
[0032] S3: Input the input set into the image processing model to obtain a processing result, wherein the image processing model includes a plurality of sequentially connected Fuse-Former layers, aggregation layers, linear layers and softmax activation functions, wherein the plurality of sequentially connected Fuse-Former layers are used to obtain global classification labels and local classification labels according to the input set; the aggregation layer is used to perform aggregation processing on the global classification labels and the local classification labels to obtain aggregated labels; the linear layer is used to perform feature conversion processing on the aggregated labels to obtain an output vector; the softmax activation function obtains an image processing result according to the output vector, and obtains a disease prediction result according to the image processing result.
[0033] In one embodiment, the image processing model includes a plurality of sequentially connected Fuse-Former layers, aggregation layers, linear layers, and softmax activation functions, wherein the cascaded Fuse-Former layers are as follows: Figure 3 As shown, each Fuse-Former layer includes a global feature extraction module and a local feature extraction module. The global feature extraction module is used to obtain a global blood oxygen level dependency marker and a global classification marker, and the local feature extraction module is used to obtain a local blood oxygen level dependency marker and a local classification marker.
[0034] It should be noted that the global feature extraction module includes a first normalization layer, a fusion window multi-head self-attention layer, a label fusion layer, a second normalization layer and a multi-layer perceptron layer connected in sequence, wherein the first normalization layer is used to normalize the input set to obtain a normalized input set; the fusion window multi-head self-attention layer is used to divide the normalized input set into multiple continuous time windows; the label fusion layer is used to fuse the contextual blood oxygen level dependency labels of the continuous time windows to obtain fused label features; the second normalization layer is used to normalize the fused label features to obtain normalized fused label features; the multi-layer perceptron layer is used to extract features from the normalized fused label features to obtain global blood oxygen level dependency labels and global classification labels, which are used as the input of the next Fuse-Former layer. Specifically, the input set X will first enter the first normalization layer (layer normalization) to accelerate the convergence of the Fuse-Former layer and improve the generalization ability. FW-MSA divides the standardized X input into The size is , the step length is In order to achieve information interaction between windows and eliminate the influence of boundary effects to a certain extent, the receptive field of a given window contains feature tokens and set them on both sides edge tokens. When performing multi-head self-attention on a window, assume that Q (query), K (key), and V (value) are defined as follows:
[0035]
[0036]
[0037] The self-attention of a single head can be expressed as follows:
[0038] Subsequently, this embodiment fuses the contextual blood oxygen level dependent tokens (BOLD tokens) of the continuous time window, a process called token fusion. The fused (aggregated) tokens can be expressed as:
[0039] in Indicates the index of MSA for each window. Token fusion can effectively promote information interaction between windows and realize comprehensive extraction of global features. Fusion features After passing through a layernormalization layer and an MLP layer, the global BOLD token and the global CLS token are output, which are called and .
[0040] It should be noted that after the input set X enters the local feature extraction module, the steps of extracting the local blood oxygen level dependency token (local BOLD token) and the local classification token (local CLS token) by KL distributed attention are as follows: Set a window with the same step size as the input set to traverse the input set X of Fuse-Former; Starting from the second window, the distribution difference between each window and the previous window is calculated. The distribution difference is calculated as follows:
[0041] in, represents the i-th window; Use the top-k algorithm to select the one with the largest difference in distribution before and after windows. The selected windows Will be the same as the previous window Stitched together into a fusion window , to obtain more comprehensive information on BOLD level changes; Fusion Window The MSA calculation is performed separately. After completion, multiple fusion windows are merged, and the local BOLD token and local CLS token are separated from the features. In this embodiment, they are represented as and .
[0042] Example 2 This example uses Alzheimer's Disease (AD) as an example. The data in this example comes from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the Autism Brain Imaging Data Exchange I (ABIDE I). (1) ADNI is a global, open, and authoritative longitudinal multi-center database that aims to track AD and study clinical, imaging, genetic, and biochemical biomarkers. This study used rs-fMRI sample data from 118 AD subjects and 181 normal subjects (HC) at different time points in 76 subject samples. (2) ABIDE I aggregates and publishes ASD data samples collected from 17 centers around the world. The data from each center are scanned in a clinical environment, which conforms to the distribution characteristics of actual clinical data. This example uses 871 scans of 403 ASD subjects and 468 healthy controls. The above data are preprocessed through the resting-state fMRI data processing assistant (DPARSF) pipeline, including temporal layer correction, head motion correction, noise regression, spatial normalization, and spatial smoothing. The definition of ROIs was derived from the public brain atlas Schaefer atlas, which labeled 400 regions in seven intrinsic connectivity networks.
[0043] 1. Parameter settings Fuse-Former is trained for 20 epochs. The initial learning rate is 0.0002 and gradually decreases to 0.00001. The overall dropout rate is 0.1 and the hidden layer factor is 400. For the global feature extraction module, the window length The step length is 20. for ,in is the stride coefficient For the local feature extraction module, the window length The step length is 3. Fixed to 3, the number of fusion windows for different data sets and the number of stacks of MSA Not the same, ADNI ( , ) and ABIDE I ( , ). All experimental results adopt 10-fold cross validation, and the ratio of training set, validation set and test set is 8:1:1.
[0044] 2. Experimental results Fuse-Former achieved the best accuracy and AUC in both ADNI and ABIDE I, with ADNI achieving an accuracy of 0.738 and an AUC of 0.798, and ABIDE I achieving an accuracy of 0.743 and an AUC of 0.808. Compared with the existing advanced comparison models, Fuse-Former has a significant improvement in performance, thanks to the following two points: (1) Fuse-Former has a "global-local" structure with more powerful feature extraction capabilities, which fully analyzes and extracts the temporal features of rs-fMRI. (2) The local feature extraction module of Fuse-Former can obtain short-term key temporal window features and discover feature segments with more significant BOLD level changes in a shorter period of time. At the same time, the study found that the model that directly uses an external atlas to extract brain ROI generally outperforms other data processing methods. The comparison results of the Fuse-Former of this embodiment with various classification models are shown in the following table:
[0045] Among them, ACC stands for accuracy, Pre stands for precision, and Rec stands for recall.
[0046] In the two ablation experiments set up as follows, the number of fusion windows for different datasets and MSA stack quantity Impact on experimental results, number of fusion windows for different data sets Impact on Fuse-Former Performance (When the number of MSA stacks of ABIDE I =2, the number of MSA stacks of ADNI 3) is as follows:
[0047] Number of MSA stacks for different datasets Impact on Fuse-Former Performance (When the number of fusion windows of ABIDE I 2, the number of fusion windows of ADNI For 2) as follows:
[0048] For the number of fusion windows , when the number of MSA stacks is fixed Afterwards, Settings are 1, 2, and 3. Fuse-Former achieves the best performance in ADNI when When Fuse-Former obtains the best performance in ABIDE I; for the number of MSA stacks ,when Fuse-Former achieves the best performance in ADNI when Fuse-Former achieves best performance in ABIDE I. and The setting has an important impact on the local feature extraction module and needs to be selected according to the actual task.
[0049] Example 3 An embodiment of the present invention also provides a computer-readable storage medium, on which is stored a computer program for a method for processing resting-state magnetic resonance functional imaging based on multi-scale information fusion. When the computer program for the method for processing resting-state magnetic resonance functional imaging based on multi-scale information fusion is processed, the steps of the method for processing resting-state magnetic resonance functional imaging based on multi-scale information fusion are implemented.
[0050] In summary, the embodiment of the present invention provides a processing method for resting state magnetic resonance functional imaging based on multi-scale information fusion, which directly operates the BOLD signal through the Fuse-Former layer, does not adopt a complex graph feature construction method, gets rid of the constraints of the graph neural network, and simply regresses to the time series feature recognition of multiple variables. At the same time, the Fuse-Former layer has the "global-local" feature perception ability of multi-scale information. The global feature extraction module interacts and fuses all window features by fusing window multi-head self-attention and label fusion to obtain a global classification label. The local feature extraction module selects the most different key time series features by comparing the feature distribution of each short-time window, obtains subtle change information in the resting state, and thus obtains local classification labels. Finally, the global classification label and the local classification label are combined to predict the disease, thereby improving the accuracy of disease prediction.
[0051] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention. These improvements and substitutions should also be regarded as the scope of protection of the present invention.
Claims
1. A method for processing resting state magnetic resonance functional imaging based on multi-scale information fusion, characterized in that: The method comprises the following steps: Acquire resting-state functional magnetic resonance imaging and perform preprocessing; An input set is obtained based on preprocessed resting-state functional magnetic resonance imaging; The input set is input into an image processing model to obtain a processing result, wherein the image processing model includes a plurality of sequentially connected Fuse-Former layers, aggregation layers, linear layers and softmax activation functions, wherein the plurality of sequentially connected Fuse-Former layers are used to obtain global classification labels and local classification labels according to the input set; the aggregation layer is used to perform aggregation processing on the global classification labels and the local classification labels to obtain aggregated labels; the linear layer is used to perform feature conversion processing on the aggregated labels to obtain an output vector; and the softmax activation function obtains an image processing result according to the output vector.
2. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 1, characterized in that: The method of obtaining an input set based on the preprocessed resting-state functional magnetic resonance imaging includes: dividing the resting-state functional magnetic resonance imaging into multiple regions of interest using an external atlas; extracting the blood oxygen level dependent response of each region of interest; and processing the blood oxygen level dependent response to obtain an input set.
3. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 2, characterized in that: The processing of the blood oxygen level dependent response to obtain an input set specifically includes: Obtaining a response average of each voxel in the region of interest, performing z-score normalization processing of zero mean and unit variance on the response average to obtain a time series, wherein the time series includes multiple time indexes; The responses measured at each specific time index were transformed into blood oxygen level-dependent markers using linear projection; Merge a classification tag at each time index to obtain a merged tag; Add position code to the merge tag to obtain an input set, which is recorded as , which is determined by the following formula: in, is a blood oxygen level dependent marker, For classification marks, Encode the position.
4. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 3, characterized in that: Linear projection is used to transform the response measured at each specific time index into a blood oxygen level dependent signature, which is denoted as , which is determined by the following formula: in, Determined by the following formula: Among them, is the time index, is the linear projection function, is the encoding dimension.
5. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 1, characterized in that: Each Fuse-Former layer includes a global feature extraction module and a local feature extraction module, wherein the global feature extraction module is used to obtain a global blood oxygen level dependency marker and a global classification marker, and the local feature extraction module is used to obtain a local blood oxygen level dependency marker and a local classification marker.
6. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 5, characterized in that: The global feature extraction module includes a first normalization layer, a fusion window multi-head self-attention layer, a label fusion layer, a second normalization layer and a multi-layer perceptron layer connected in sequence. The first normalization layer is used to perform normalization processing on the input set to obtain a normalized input set; The fusion window multi-head self-attention layer is used to divide the normalized input set into multiple continuous time windows; The label fusion layer is used to fuse the contextual blood oxygen level dependency labels of the continuous time window to obtain a fused label feature; The second normalization layer is used to normalize the fused marker features to obtain normalized fused marker features; The multi-layer perceptron layer is used to extract features from the normalized fusion mark features to obtain global blood oxygen level dependency marks and global classification marks, and the global blood oxygen level dependency marks and the global classification marks are used as inputs of the next Fuse-Former layer.
7. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 6, characterized in that: The local feature extraction module is used to obtain the local blood oxygen level dependency mark and the local classification mark, and specifically includes: Set a window with the same step size as the input set to traverse the input set; Starting from the second window, calculate the distribution difference between each window and the previous window; use The algorithm selects multiple windows with the largest difference in front and back distribution, and splices each selected window with the previous window of the selected window to obtain a fused window; Performing multi-head self-attention calculation on the fusion window to obtain a feature-enhanced fusion window, and fusing the feature-enhanced fusion windows to obtain a feature matrix; A local blood oxygen level dependency marker and a local classification marker are extracted from the feature matrix, and the local blood oxygen level dependency marker and the local classification marker are used as inputs of the next Fuse-Former layer.
8. The method for processing resting state functional magnetic resonance imaging based on multi-scale information fusion according to claim 7, characterized in that: The distribution difference is calculated as follows: in, represents the i-th window.
9. An application of a processing method for resting state magnetic resonance functional imaging based on multi-scale information fusion, characterized in that , used for the prediction of brain diseases, the image processing results obtained can predict brain diseases.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer stored program is executed by the processor, the processing method of resting state magnetic resonance functional imaging based on multi-scale information fusion as described in any one of claims 1 to 8 is implemented.
Citation Information
Cited By
Alzheimer's disease prediction system
CN120748772A
