A method and apparatus for predicting early alzheimer's disease based on sfc features

Abstract

The embodiment of the present application relates to a kind of method and device based on SFC feature prediction early Alzheimer's disease, the method comprises: setting first brain region set and its corresponding first brain atlas, design first prediction model;The brain multi-modal MRI image of early Alzheimer's disease population and healthy population is carried out big data acquisition to obtain original sample set;According to first brain atlas and original sample set, generate first data set;First prediction model is trained based on first data set;After training, the brain multi-modal MRI image of any subject is received, and first SFC feature vector is calculated according to first brain atlas and current multi-modal image, and first SFC feature vector is input into first prediction model and is predicted to obtain current prediction result.The present application can mine SFC feature, and can be based on SFC feature early Alzheimer's disease and be classified and predicted.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and apparatus for predicting early Alzheimer's disease based on SFC features. Background Technology

[0002] Early detection of Alzheimer's Disease (AD) is crucial for slowing disease progression. Early detection methods for Alzheimer's disease based on various types of magnetic resonance imaging (MRI) have become a research hotspot in this field. Among them, functional magnetic resonance imaging (fMRI) can provide analytical information from the perspective of brain functional network disorder; diffusion tensor imaging (DTI) can provide analytical information from the perspective of white matter fiber damage.

[0003] Current techniques typically build binary classification models based on features from single-modal magnetic resonance imaging (fMRI, DTI). Clearly, this conventional classification method cannot deeply explore the structure-function coupling (SFC) relationship between the brain's structural connectivity (SC) and functional connectivity (FC) networks. Numerous neuroscience studies have shown that early Alzheimer's disease (AD) is precisely accompanied by disruption of the SFC relationship, meaning that structural connections cannot effectively support corresponding functional activities—a more biologically significant early marker. Therefore, building predictive models for early AD based on SFC features could further improve prediction accuracy. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on SFC features. This invention first defines a set of brain regions of interest (i.e., the first brain region set) based on multiple brain regions of interest associated with early Alzheimer's disease, and customizes a corresponding brain atlas (i.e., the first brain atlas) for the first brain region set. The matrix and vector structures of the structural matrix (SC matrix), functional matrix (FC matrix), and SFC feature vectors are also customized. Then, based on the vector structure of the SFC feature vectors, a neural network model (i.e., the first prediction model) for binary classification prediction of early Alzheimer's disease is customized. A raw sample set is constructed by collecting large amounts of multimodal MRI images (T1-weighted images, fMRI images, and DTI images) from two groups (early AD patients and healthy individuals), and a first dataset is built based on this raw sample set. The first prediction model is then trained on the first dataset. After model training, multimodal MRI images of any subject are received, and the images of each type are registered according to the first brain atlas. The corresponding SFC feature vectors are calculated based on the registered images, and the first prediction model is used to perform binary classification prediction based on the current SFC feature vectors. This invention can deeply explore the SFC relationship between SC and FC networks, and can perform binary classification prediction for early AD based on SFC features. It can not only effectively make up for the shortcomings of traditional schemes, but also improve the prediction accuracy.

[0005] To achieve the above objectives, a first aspect of the present invention provides a method for predicting early Alzheimer's disease based on SFC features, the method comprising:

[0006] Set up a first brain region set; generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and set up the matrix and vector feature structure of the structure matrix, function matrix, and structure-function coupling feature vector to obtain the corresponding SC matrix, FC matrix and SFC feature vector.

[0007] Based on the vector structure of the SFC feature vector, a neural network model for binary classification prediction of early Alzheimer's disease is designed, denoted as the first prediction model. The first prediction model is used to perform binary classification prediction based on the SFC feature vector input to the model and output the corresponding first prediction vector. The first prediction vector is composed of a first and a second prediction probability. The classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively.

[0008] The original sample set was obtained by collecting large data on multimodal MRI images of the brains of people with early Alzheimer's disease and healthy people; the multimodal MRI images of the brain include brain T1-weighted images, brain fMRI images, and brain DTI images, and the imaging time interval between any two types of images is less than a preset interval duration threshold.

[0009] Based on the first brain map, the images of each sample in the original sample set are registered, and the structure matrix, function matrix and structure-function coupling feature vector of the sample are calculated based on the registered images. Based on the calculation results and the sample type, corresponding training data is generated to form the corresponding first dataset.

[0010] The first prediction model is trained based on the first dataset;

[0011] After model training is completed, the multimodal MRI images of the brain of any subject are received as the current multimodal images; and the images of various types of the current multimodal images are registered according to the first brain atlas, and the corresponding structural matrix, functional matrix and structural-functional coupling feature vector are calculated based on the registered images to obtain the corresponding first SFC feature vector; the first SFC feature vector is input into the first prediction model to predict to obtain the corresponding first prediction vector; and the classification type corresponding to the highest prediction probability in the first prediction vector is taken as the current prediction result.

[0012] Preferably, the first brain region set consists of multiple brain regions including the Default Mode Network (DMN), Sensorimotor Network (SMN), Visual Network (VN), Executive Control Network (ECN), and Dorsal Attention Network (DAN), with the total number of brain regions N being a positive integer; the standard brain atlas includes at least the Yeo-7 atlas, Yeo-17 atlas, Power Atlas atlas, BNA atlas, MMP atlas, and AAL atlas; the first brain atlas consists of N brain region sub-atlases; the voxel space of both the standard brain atlas and the first brain atlas is the MNI standard space;

[0013] The SC matrix consists of N×N structural features s i,j Composition, 1 ≤ brain region index i, j ≤ N; the structural feature s i,j The characteristic components include total fiber count, fiber density, fiber length, cross-sectional area index, and FA (fiber averaging) characteristic; the total fiber count characteristic is the total number of white matter fibers from brain region i to brain region j; the fiber density characteristic is the white matter fiber density from brain region i to brain region j; the fiber length characteristic is the average length of white matter fibers from brain region i to brain region j; the cross-sectional area index is the cross-sectional area index of white matter fibers from brain region i to brain region j; and the FA characteristic is the average FA value of white matter fibers from brain region i to brain region j.

[0014] The FC matrix consists of N×N functional features f i,j Composition; the functional feature f i,j The correlation index of BOLD signals between brain region i and brain region j;

[0015] The SFC feature vector is composed of globally coupled feature c. global It consists of a brain region coupling feature sequence C; the brain region coupling feature sequence C is composed of N brain region coupling features c i Composition; the global coupling feature c global The global SDI index of the SC matrix and the FC matrix; the brain region coupling feature c i The SDI index of the i-th brain region;

[0016] The early Alzheimer's disease population includes multiple early Alzheimer's disease patients;

[0017] The healthy population consists of multiple healthy individuals who do not have early-stage Alzheimer's disease;

[0018] The original sample set includes multiple first sample records; each first sample record corresponds one-to-one with an early Alzheimer's disease patient or a healthy individual in the early Alzheimer's disease population and the healthy population; each first sample record includes the sample type, the brain T1-weighted image, the brain fMRI image, and the brain DTI image; the sample type includes patient samples and healthy samples;

[0019] The first dataset includes multiple first data records; each first data record corresponds one-to-one with a first sample record; each first data record includes a first training SFC feature vector and a first label vector; the first label vector is composed of first and second label probabilities; the classification types of the first and second label probabilities correspond to disease type and healthy type, respectively; if the first sample type corresponding to the current data record is a patient sample, then the first label probability is 1 and the second label probability is 0; if the first sample type corresponding to the current data record is a healthy sample, then the first label probability is 0 and the second label probability is 1.

[0020] Preferably, the first and second model input terminals of the first prediction model are respectively used to receive the global coupling feature c corresponding to the SFC feature vector. global and the brain region coupling feature sequence C{c i The model output is used to output the corresponding first prediction vector;

[0021] The first prediction model includes a first fully connected layer, a first activation layer, a second fully connected layer, a first convolutional layer, a second activation layer, a first pooling layer, a third fully connected layer, a first fusion layer, a fourth fully connected layer, and a first classification layer;

[0022] The input of the first fully connected layer is connected to the input of the first model, and its output is connected to the input of the first activation layer; the output of the first activation layer is connected to the input of the second fully connected layer; the output of the second fully connected layer is connected to the first input of the first fusion layer; the input of the first convolutional layer is connected to the input of the second model, and its output is connected to the input of the second activation layer; the output of the second activation layer is connected to the input of the first pooling layer; the output of the first pooling layer is connected to the input of the third fully connected layer; the output of the third fully connected layer is connected to the second input of the first fusion layer; the output of the first fusion layer is connected to the input of the fourth fully connected layer; the output of the fourth fully connected layer is connected to the input of the first classification layer; and the output of the first classification layer is connected to the model output.

[0023] The first fully connected layer is used for the global coupling feature c global The first feature vector H1 obtained by feature encoding is sent to the first activation layer;

[0024] The encoding method of the first feature vector H1 is as follows:

[0025] ;

[0026] W1 and B1 are the first weight matrix and the first bias vector of the model; the first weight matrix W1 and the first bias vector B1 both have the shape D1×1; D1 is the preset first feature dimension; the first feature vector H1 has the shape D1×1;

[0027] The first activation layer is used to activate the first feature vector H1 using the ReLU activation function to obtain the corresponding second feature vector H2, which is then sent to the second fully connected layer.

[0028] The activation method of the second feature vector H2 is as follows:

[0029] ;

[0030] The shape of the second feature vector H2 is D1×1;

[0031] The second fully connected layer is used to perform feature mapping on the second feature vector H2 to obtain the corresponding third feature vector H3, which is then sent to the first fusion layer.

[0032] The mapping method for the third feature vector H3 is as follows:

[0033] ;

[0034] W2 and B2 are the second weight matrix and the second bias vector of the model; the second weight matrix W2 has a shape of D2×D1, and the second bias vector B2 has a shape of D2×1; D2 is a preset second feature dimension; the third feature vector H3 has a shape of D2×1;

[0035] The first convolutional layer is used to apply D3 1×1 convolutional kernels to each of the brain region coupling features c in the brain region coupling feature sequence C. i Performing a one-dimensional convolution operation yields a feature vector of length D3. ; and the resulting N feature vectors The fourth feature vector H4, which has a shape of N×D3, is sent to the second activation layer;

[0036] Where D3 is the preset third feature dimension;

[0037] The second activation layer is used to activate the fourth feature vector H4 using the ReLU activation function to obtain the corresponding fifth feature vector H5, which is then sent to the first pooling layer.

[0038] The activation method of the fifth feature vector H5 is as follows:

[0039] ;

[0040] The shape of the fifth feature vector H5 is N×D3;

[0041] The first pooling layer is used to perform global pooling on each feature channel of the fifth feature vector H5 to obtain the corresponding sixth feature vector H6, which is then sent to the third fully connected layer.

[0042] The shape of the sixth feature vector H6 is 1×D3;

[0043] The third fully connected layer is used to perform feature mapping processing on the sixth feature vector H6 to obtain the corresponding seventh feature vector H7 and send it to the first fusion layer.

[0044] The mapping method for the seventh feature vector H7 is as follows:

[0045] ;

[0046] W3 and B3 are the third weight matrix and the third bias vector of the model; the third weight matrix W3 has a shape of D4×D3, and the third bias vector B3 has a shape of D4×1; D4 is a preset fourth feature dimension; the seventh feature vector H7 has a shape of D4×1.

[0047] The first fusion layer is used to perform vector concatenation processing on the third feature vector H3 and the seventh feature vector H7 to obtain the corresponding eighth feature vector H8, which is then sent to the fourth fully connected layer.

[0048] The shape of the eighth feature vector H8 is (D2+D4)×1;

[0049] The fourth fully connected layer is used to perform binary classification feature mapping on the eighth feature vector H8 to obtain the corresponding ninth feature vector H9 and send it to the first classification layer.

[0050] The mapping method for the ninth feature vector H9 is as follows:

[0051] ;

[0052] W4 and B4 are the fourth weight matrix and fourth bias vector of the model; the fourth weight matrix W4 has a shape of 2×(D2+D4), and the fourth bias vector B4 has a shape of 2×1; the ninth feature vector H9 has a shape of 2×1; the two vector data of the ninth feature vector H9 are denoted as corresponding... , ;

[0053] The first classification layer is used to use the Softmax function to perform binary classification probability calculation based on the ninth feature vector H9 to obtain the corresponding first and second predicted probabilities, and outputs the corresponding first predicted vector composed of the two predicted probabilities.

[0054] The first and second prediction probabilities are calculated as follows:

[0055] ,

[0056] .

[0057] Preferably, the step of registering the images of each sample in the original sample set according to the first brain atlas, calculating the structure matrix, function matrix, and structure-function coupling feature vector of the sample based on the registered images, and generating corresponding training data to form the corresponding first dataset based on the calculation results and sample type specifically includes:

[0058] Step 41: Take each of the first sample records in the original sample set as the current sample record;

[0059] Step 42: The sample type, brain T1-weighted image, brain fMRI image, and brain DTI image recorded in the current sample are used as the corresponding current sample type, current T1-weighted image, current fMRI image, and current DTI image.

[0060] Each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence, wherein the BOLD signal sequence consists of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values;

[0061] Step 43: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector.

[0062] Step 44: Identify the current sample type; if the current sample type is a patient sample, set the corresponding first label probability bit to 1 and the second label probability to 0; if the current sample type is a healthy sample, set the corresponding first label probability bit to 0 and the second label probability to 1; and form the corresponding first label vector from the current first label probability and the second label probability.

[0063] Step 45: The first training SFC feature vector and the first label vector corresponding to the current sample record are combined to form a corresponding first data record; and all the obtained first data records are combined to form the corresponding first dataset.

[0064] Preferably, training the first prediction model based on the first dataset specifically includes:

[0065] Step 51: Based on a preset first segmentation ratio, the first dataset is divided into two sub-datasets, denoted as the first training set and the first evaluation set.

[0066] Both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is N. tr The total number of records N in the first evaluation set ev The ratio of the total number N tr :Nev The first segmentation ratio is satisfied;

[0067] Step 52: Input the first training SFC feature vector of each first data record in the first training set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And record the first label vector of each of the first data records as the corresponding vector. ; and by each of the aforementioned vectors The corresponding vector Form a corresponding first prediction-label pair ( , );

[0068] Where 1 ≤ index u ≤ N tr ;

[0069] Step 53, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L M The corresponding first loss value is obtained through calculation;

[0070] Wherein, the model loss function L M Based on cross-entropy loss function L CE Implementation, specifically:

[0071] ;

[0072] Step 54: Identify whether the first loss value meets the preset first loss value range; if yes, proceed to step 55; if no, based on the preset first model optimizer, move towards making the model loss function L... M The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 52 after this round of modulation is completed;

[0073] The first model optimizer includes the Adam optimizer and the SGD optimizer.

[0074] Step 55: Input the first training SFC feature vector of each first data record in the first evaluation set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And record the first label vector of each of the first data records as the corresponding vector. ; and by each of the aforementioned vectors The corresponding vector Form a corresponding second prediction-label pair ( , ); and based on all the obtained second prediction-label pairs ( , The accuracy, precision, recall, and F1 score are calculated to obtain the corresponding first accuracy, first precision, first recall, and first F1 score.

[0075] Where 1 ≤ index q ≤ N ev ;

[0076] Step 56: Identify whether the first accuracy, first precision, first recall, and first F1 score all satisfy their respective first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 51 to continue training; if yes, stop training and confirm that the training of the first prediction model has ended.

[0077] Preferably, the step of registering various types of images of the current multimodal image according to the first brain atlas and calculating the corresponding first SFC feature vector based on the registered images, including the corresponding structure matrix, function matrix, and structure-function coupling feature vector, specifically includes:

[0078] Step 61: Use the brain T1-weighted image, brain fMRI image, and brain DTI image of the current multimodal image as the corresponding current T1-weighted image, current fMRI image, and current DTI image;

[0079] Each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence, wherein the BOLD signal sequence consists of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values;

[0080] Step 62: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector.

[0081] Preferably, the calculation of the SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image specifically includes:

[0082] Step 71: Use the FSL tool to perform voxel space registration from the current fMRI image to the current T1-weighted image to obtain the corresponding first registration relationship; use the FSL tool to perform voxel space registration from the current DTI image to the current T1-weighted image to obtain the corresponding second registration relationship; use the FSL tool to perform voxel space registration from the current T1-weighted image to the MNI standard space corresponding to the first brain atlas to obtain the corresponding third registration relationship; and perform MNI standard space registration on the three-dimensional voxel coordinates of the current fMRI image based on the first and third registration relationships; and perform MNI standard space registration on the three-dimensional voxel coordinates of the current DTI image based on the second and third registration relationships.

[0083] Step 72: Use MRtrix3 or FSL tools to identify white matter fiber streamlines in the whole brain based on the current DTI images to obtain the corresponding white matter fiber streamline set.

[0084] The white matter fiber streamline set includes multiple white matter fiber streamlines; each white matter fiber streamline is composed of multiple fiber voxel points; the voxel feature of each fiber voxel point includes three-dimensional voxel coordinates and FA value; the first and last fiber voxel points of each white matter fiber streamline are recorded as the corresponding start point and end point;

[0085] Step 73: Combine the i-th and j-th brain region sub-maps of the first brain map to form a corresponding brain region group P. i,j ; and group each of the brain regions P i,j The current brain region group is defined as follows: the two sub-image spaces in the current DTI image corresponding to the current brain region group are defined as the corresponding first and second sub-spaces; the white matter fiber streamlines are grouped together, and the white matter fiber streamlines whose starting points and ending points fall on the first and second sub-spaces are recorded as the corresponding current group fiber streamlines; and all the current group fiber streamlines are combined to form the corresponding fiber streamline subset F. i,j ;

[0086] Step 74, divide each of the fiber streamline subsets F i,jAs the corresponding current subset; the total number of white matter fiber streamlines in the current subset is counted and the statistical result is used as the corresponding fiber total number feature; the average volume of the first and second subspaces corresponding to the current subset is calculated, and the fiber density feature is calculated from the fiber total number feature and the average volume = fiber total number feature / average volume; the streamline length of each white matter fiber streamline in the current subset is calculated, and the average length of all streamlines is used as the corresponding fiber length feature; the minimum FA value of each white matter fiber streamline in the current subset is identified, and the average of all minimum FA values ​​is used as the corresponding cross-sectional area index; the average of all FA values ​​in the current subset is calculated and the calculation result is used as the corresponding FA feature; and the fiber total number feature, fiber density feature, fiber length feature, cross-sectional area index, and FA feature corresponding to the current subset are combined to form a corresponding structural feature s. i,j ; and the obtained N×N structural features s i,j Form the corresponding SC matrix;

[0087] Step 75: Take each brain region sub-map of the first brain atlas as the current atlas; take the sub-space image corresponding to the current atlas in the current fMRI image as the current image; calculate the mean of all BOLD signal sequences in the current image to obtain the corresponding mean signal sequence; delete the mean of the first M BOLD signals in the mean signal sequence, where M is a preset positive integer; and extract the low-frequency signal from the mean signal sequence according to a preset low-frequency signal range to obtain the corresponding low-frequency signal sequence.

[0088] Step 76: Calculate the Pearson correlation coefficient between the two low-frequency signal sequences corresponding to the i-th and j-th brain region sub-maps in the first brain map, and perform a Fisher z-transform on the calculation result to obtain the corresponding functional feature f. i,j ; and the obtained N×N functional features f i,j The corresponding FC matrix is ​​formed.

[0089] Preferably, the step of calculating the corresponding SFC feature vector based on the current SC matrix and FC matrix specifically includes:

[0090] Step 81, convert each of the structural features s of the SC matrix i,jAs the current structural feature; the total fiber count, fiber density, fiber length, cross-sectional area index, and FA feature of the current structural feature are normalized to obtain the corresponding normalized total count, normalized density, normalized length, normalized cross-sectional area index, and normalized FA value; the six normalized data are weighted according to six preset weighting coefficients to obtain the corresponding six weighted values; the six weighted values ​​are summed and the result is used as the corresponding connection strength sd. i,j ; and the obtained N×N connection strengths sd i,j Form the corresponding intensity matrix SD;

[0091] Step 82: Take the i-th row of the intensity matrix SD as the current row; calculate the corresponding cutoff position K=int(N×20%) based on the total number of brain regions N, where int() is the floor function; and sort the N-1 connection intensities sd of the current row in descending order. i,j≠i Sort the connections by their top K strengths (sd). i,j The index j forms the corresponding index vector ID. i ;

[0092] Step 83: Take the i-th row of the FC matrix as the current row; and use the N-1 functional features f of the current row. i,j≠i Form the corresponding observation function vector ; and based on the index vector ID corresponding to the current row i Calculate the corresponding predictive function vector ;

[0093] Wherein, the prediction function vector It consists of the corresponding N-1 predictive feature features composition;

[0094] The predictive functional features The calculation method is as follows:

[0095] ;

[0096] Step 84, for each of the observation function vectors The corresponding prediction function vector The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. i ; and based on each of the aforementioned coefficients ρ i Calculate the corresponding brain region coupling features c i ; and obtain N brain region coupling features c i This forms the corresponding brain region coupling feature sequence C;

[0097] Among them, the brain region coupling feature c i The calculation method is as follows:

[0098] ;

[0099] Step 85, N(N-1) / 2 of the connection strengths sd, excluding diagonal matrix elements, are formed from the upper triangular matrix of the strength matrix SD. i,j Form the corresponding vector V S ; and consisting of N(N-1) / 2 functional features f, excluding diagonal matrix elements, in the upper triangular matrix of the FC matrix. i,j Form the corresponding vector V F ; and for the vector V S and the vector V F The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. global ; and based on the coefficient ρ global Calculate the corresponding global coupling feature c global ;

[0100] Wherein, the global coupling feature c global The calculation method is as follows:

[0101] .

[0102] A second aspect of the present invention provides an apparatus for implementing the method for predicting early Alzheimer's disease based on SFC features as described in the first aspect above. The apparatus includes: a brain region customization module, a model customization module, a sample acquisition module, a dataset preparation module, a model training module, and a model application module.

[0103] The brain region customization module is used to set up a first brain region set; and generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and set up the matrix and vector feature structure of the structure matrix, function matrix, and structure-function coupling feature vector to obtain the corresponding SC matrix, FC matrix and SFC feature vector.

[0104] The model customization module designs a neural network model for binary classification prediction of early Alzheimer's disease based on the vector structure of the SFC feature vector, denoted as the first prediction model; the first prediction model is used to perform binary classification prediction based on the SFC feature vector input to the model and output the corresponding first prediction vector; the first prediction vector is composed of first and second prediction probabilities; the classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively;

[0105] The sample acquisition module is used to collect large amounts of brain multimodal MRI images of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample set; the brain multimodal MRI images include brain T1-weighted images, brain fMRI images, and brain DTI images, and the imaging time interval between any two types of images is lower than a preset interval duration threshold.

[0106] The dataset preparation module is used to register the images of each sample in the original sample set according to the first brain map, and to calculate the structure matrix, function matrix and structure-function coupling feature vector of the sample based on the registered images, and to generate corresponding training data based on the calculation results and sample type to form the corresponding first dataset.

[0107] The model training module trains the first prediction model based on the first dataset;

[0108] The model application module is used to receive the multimodal MRI images of the brain of any subject as the current multimodal images after the model training is completed; and to register various types of images of the current multimodal images according to the first brain atlas, and to calculate the corresponding first SFC feature vector based on the registered images by performing corresponding structural matrix, functional matrix and structural-functional coupling feature vector; and to input the first SFC feature vector into the first prediction model to predict and obtain the corresponding first prediction vector; and to take the classification type corresponding to the maximum prediction probability in the first prediction vector as the current prediction result.

[0109] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0110] The processor is used to couple with the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect above;

[0111] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0112] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0113] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on SFC features. As described above, in this embodiment of the invention, a set of brain regions of interest (i.e., the first brain region set) is first customized based on multiple brain regions of interest related to early Alzheimer's disease, and a corresponding brain atlas (i.e., the first brain atlas) is customized for the first brain region set. The matrix and vector structures of the structural matrix (SC matrix), functional matrix (FC matrix), and SFC feature vectors are also customized. Then, a neural network model (i.e., the first prediction model) for binary classification prediction of early Alzheimer's disease is customized based on the vector structure of the SFC feature vectors. An original sample set is constructed by collecting large amounts of multimodal MRI images (T1-weighted images, fMRI images, and DTI images) of the brains of two groups (early AD patients and healthy individuals), and a first dataset is constructed based on the original sample set. Then, the first prediction model is trained based on the first dataset. After the model training is completed, multimodal MRI images of the brains of any subject are received, and the images of various types of the current multimodal images are registered according to the first brain atlas. The corresponding SFC feature vectors are calculated based on the registered images, and the first prediction model is used to perform binary classification prediction based on the current SFC feature vectors. The embodiments of the present invention can deeply explore the SFC relationship between SC network and FC network, and can perform binary classification prediction of AD in the early stage based on SFC features, which not only effectively makes up for the shortcomings of traditional schemes, but also improves the prediction accuracy. Attached Figure Description

[0114] Figure 1 This is a schematic diagram of a method for predicting early Alzheimer's disease based on SFC features, provided in Embodiment 1 of the present invention.

[0115] Figure 2 A schematic diagram of the modules of the first prediction model provided in Embodiment 1 of the present invention;

[0116] Figure 3 This is a module structure diagram of a device for predicting early Alzheimer's disease based on SFC features, provided in Embodiment 2 of the present invention.

[0117] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0118] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0119] Embodiment 1 of this invention provides a method for predicting early Alzheimer's disease based on SFC features, such as... Figure 1 The schematic diagram of a method for predicting early Alzheimer's disease based on SFC features provided in Embodiment 1 of the present invention is shown. The method mainly includes the following steps:

[0120] Step 1: Set up the first brain region set; generate the first brain map corresponding to the first brain region set based on the preset standard brain map; and set up the structure matrix, function matrix, structure-function coupling feature vector matrix and vector feature structure to obtain the corresponding SC matrix, FC matrix and SFC feature vector.

[0121] Here, the first set of brain regions in this embodiment of the invention is set based on expert experience. The first set of brain regions consists of multiple types of brain regions, including the Default Mode Network (DMN), Sensorimotor Network (SMN), Visual Network (VN), Executive Control Network (ECN), and Dorsal Attention Network (DAN). The total number of brain regions N is a positive integer. All of the above-mentioned brain networks are brain networks related to the early diagnosis of Alzheimer's disease.

[0122] The standard brain atlas of this invention includes at least the Yeo-7 atlas, Yeo-17 atlas, Power Atlas atlas, BNA atlas, MMP atlas, and AAL atlas. The first brain atlas consists of N brain region sub-maps. The voxel space of both the standard brain atlas and the first brain atlas is the MNI standard space.

[0123] The data structure of the SC matrix in this embodiment of the invention is as follows: the SC matrix consists of N×N structural features s i,j Composition, 1 ≤ brain region index i, j ≤ N; structural features s i,j The characteristic components include total fiber count, fiber density, fiber length, cross-sectional area index, and FA (fiber averaging) characteristics. Total fiber count is the total number of white matter fibers from brain region i to brain region j. Fiber density is the white matter fiber density from brain region i to brain region j. Fiber length is the average length of white matter fibers from brain region i to brain region j. Cross-sectional area index is the cross-sectional area index of white matter fibers from brain region i to brain region j. FA is the average FA value of white matter fibers from brain region i to brain region j.

[0124] The data structure of the FC matrix in this embodiment of the invention is as follows: the FC matrix consists of N×N functional features f i,j Composition; Functional characteristics fi,j The correlation index of BOLD signals between brain region i and brain region j.

[0125] The data structure of the SFC feature vector in this embodiment of the invention is as follows: The SFC feature vector consists of globally coupled feature c global It consists of brain region coupling feature sequence C; brain region coupling feature sequence C consists of N brain region coupling features c i Composition; global coupling feature c global The SDI is the global spatial decoupling index of the SC and FC matrices; brain region coupling features c i The SDI index represents the brain region i. Compared to healthy individuals, early-stage AD patients exhibit a decreased structural-functional coupling relationship in their brains, resulting in a higher SDI index.

[0126] Step 2: Design a neural network model for binary classification prediction of early Alzheimer's disease based on the vector structure of SFC feature vectors, denoted as the first prediction model.

[0127] Here, the first prediction model in this embodiment of the invention is used to perform binary classification prediction based on the SFC feature vector input to the model and output the corresponding first prediction vector; wherein, the first prediction vector is composed of first and second prediction probabilities; the classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively.

[0128] The first and second model inputs of the first prediction model are used to receive the corresponding global coupling features c in the SFC feature vector, respectively. global and brain region coupling feature sequence C{c i The model output is used to output the corresponding first prediction vector.

[0129] like Figure 2 As shown in the schematic diagram of the first prediction model provided in Embodiment 1 of the present invention, the model components of the first prediction model include: a first fully connected layer, a first activation layer, a second fully connected layer, a first convolutional layer, a second activation layer, a first pooling layer, a third fully connected layer, a first fusion layer, a fourth fully connected layer, and a first classification layer.

[0130] like Figure 2As shown, the connection relationships of the model components in the first prediction model are as follows: the input of the first fully connected layer is connected to the input of the first model, and its output is connected to the input of the first activation layer; the output of the first activation layer is connected to the input of the second fully connected layer; the output of the second fully connected layer is connected to the first input of the first fusion layer; the input of the first convolutional layer is connected to the input of the second model, and its output is connected to the input of the second activation layer; the output of the second activation layer is connected to the input of the first pooling layer; the output of the first pooling layer is connected to the input of the third fully connected layer; the output of the third fully connected layer is connected to the second input of the first fusion layer; the output of the first fusion layer is connected to the input of the fourth fully connected layer; the output of the fourth fully connected layer is connected to the input of the first classification layer; and the output of the first classification layer is connected to the model output.

[0131] The model component functions of the first prediction model are shown below.

[0132] 1) First fully connected layer:

[0133] The first fully connected layer in this embodiment of the invention is used to handle the globally coupled feature c. global The first feature vector H1 obtained by feature encoding is sent to the first activation layer.

[0134] Here, the encoding method of the first feature vector H1 in this embodiment of the invention is as follows:

[0135] ;

[0136] Wherein, W1 and B1 are the first weight matrix and the first bias vector of the model; the first weight matrix W1 and the first bias vector B1 both have the shape D1×1; D1 is the preset first feature dimension; the first feature vector H1 has the shape D1×1.

[0137] 2) First activation layer:

[0138] In this embodiment of the invention, the first activation layer is used to activate the first feature vector H1 using the ReLU activation function to obtain the corresponding second feature vector H2, which is then sent to the second fully connected layer.

[0139] Here, the activation method of the second feature vector H2 in this embodiment of the invention is as follows:

[0140] ;

[0141] The shape of the second feature vector H2 is D1×1.

[0142] 3) Second fully connected layer:

[0143] In this embodiment of the invention, the second fully connected layer is used to perform feature mapping processing on the second feature vector H2 to obtain the corresponding third feature vector H3, which is then sent to the first fusion layer.

[0144] Here, the mapping method of the third feature vector H3 in this embodiment of the invention is as follows:

[0145] ;

[0146] Wherein, W2 and B2 are the second weight matrix and the second bias vector of the model; the shape of the second weight matrix W2 is D2×D1, and the shape of the second bias vector B2 is D2×1; D2 is the preset second feature dimension; the shape of the third feature vector H3 is D2×1.

[0147] 4) First convolutional layer:

[0148] The first convolutional layer in this embodiment of the invention is used to apply D3 1×1 convolutional kernels to the coupling features c of each brain region in the brain region coupling feature sequence C. i Performing a one-dimensional convolution operation yields a feature vector of length D3. ; and the resulting N eigenvectors The fourth feature vector H4, which has a shape of N×D3, is sent to the second activation layer.

[0149] Here, D3 in this embodiment of the invention is a preset third feature dimension.

[0150] 5) Second activation layer:

[0151] In this embodiment of the invention, the second activation layer is used to activate the fourth feature vector H4 using the ReLU activation function to obtain the corresponding fifth feature vector H5, which is then sent to the first pooling layer.

[0152] Here, the activation method of the fifth feature vector H5 in this embodiment of the invention is as follows:

[0153] ;

[0154] The fifth feature vector H5 has a shape of N×D3.

[0155] 6) First pooling layer:

[0156] In this embodiment of the invention, the first pooling layer is used to perform global pooling on each feature channel of the fifth feature vector H5 to obtain the corresponding sixth feature vector H6, which is then sent to the third fully connected layer.

[0157] Here, the shape of the sixth feature vector H6 in this embodiment of the invention is 1×D3.

[0158] 7) Third fully connected layer:

[0159] In this embodiment of the invention, the third fully connected layer is used to perform feature mapping processing on the sixth feature vector H6 to obtain the corresponding seventh feature vector H7, which is then sent to the first fusion layer.

[0160] Here, the mapping method of the seventh feature vector H7 in this embodiment of the invention is as follows:

[0161] ;

[0162] Wherein, W3 and B3 are the third weight matrix and the third bias vector of the model; the shape of the third weight matrix W3 is D4×D3, and the shape of the third bias vector B3 is D4×1; D4 is the preset fourth feature dimension; and the shape of the seventh feature vector H7 is D4×1.

[0163] 8) First Fusion Layer:

[0164] In this embodiment of the invention, the first fusion layer is used to perform vector concatenation processing on the third feature vector H3 and the seventh feature vector H7 to obtain the corresponding eighth feature vector H8, which is then sent to the fourth fully connected layer.

[0165] Here, the shape of the eighth feature vector H8 in this embodiment of the invention is (D2+D4)×1.

[0166] 9) Fourth fully connected layer:

[0167] In this embodiment of the invention, the fourth fully connected layer is used to perform binary classification feature mapping on the eighth feature vector H8 to obtain the corresponding ninth feature vector H9, which is then sent to the first classification layer.

[0168] Here, the mapping method of the ninth feature vector H9 in this embodiment of the invention is as follows:

[0169] ;

[0170] Where W4 and B4 are the fourth weight matrix and fourth bias vector of the model; the fourth weight matrix W4 has a shape of 2×(D2+D4), and the fourth bias vector B4 has a shape of 2×1; the ninth eigenvector H9 has a shape of 2×1; the two vector data of the ninth eigenvector H9 are denoted as corresponding... , .

[0171] 10) First classification layer:

[0172] In this embodiment of the invention, the first classification layer is used to calculate the corresponding first and second predicted probabilities by performing binary classification probability calculation based on the ninth feature vector H9 using the Softmax function, and outputs the corresponding first predicted vector composed of the two predicted probabilities.

[0173] Here, the calculation methods for the first and second prediction probabilities in this embodiment of the invention are as follows:

[0174] ,

[0175] .

[0176] Step 3: Collect big data on multimodal MRI images of the brains of people with early Alzheimer's disease and healthy people to obtain the corresponding raw sample set.

[0177] Here, the early Alzheimer's disease population in this embodiment of the invention includes multiple early Alzheimer's disease patients; the healthy population consists of multiple healthy individuals who do not have early Alzheimer's disease. The multimodal brain MRI images in this embodiment of the invention include brain T1-weighted images, brain fMRI images, and brain DTI images; it should be noted that the imaging time interval between any two of these three types of images is less than a preset interval duration threshold, thus ensuring the consistency of the multimodal images.

[0178] The original sample set of this invention includes multiple first sample records; each first sample record corresponds one-to-one with an early Alzheimer's disease patient or a healthy person in the early Alzheimer's disease population and the healthy population; each first sample record includes sample type, brain T1-weighted image, brain fMRI image, and brain DTI image; the sample type includes patient sample and healthy sample.

[0179] The big data collection channels in this invention include multiple publicly available neuroimaging datasets, such as the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, the Open Access Series of Imaging Studies (OASIS) dataset, and the Dominantly Inherited Alzheimer Network Observational Study (DIAN-OBS) dataset.

[0180] Step 4: Register the images of each sample in the original sample set according to the first brain map, and calculate the structure matrix, function matrix and structure-function coupling feature vector of the sample based on the registered images. Generate corresponding training data based on the calculation results and sample type to form the corresponding first dataset.

[0181] The first dataset includes multiple first data records; each first data record corresponds one-to-one with a first sample record; each first data record includes a first training SFC feature vector and a first label vector; the first label vector consists of first and second label probabilities; the classification types of the first and second label probabilities correspond to disease type and healthy type, respectively; if the first sample type corresponding to the current data record is a patient sample, then the first label probability is 1 and the second label probability is 0; if the first sample type corresponding to the current data record is a healthy sample, then the first label probability is 0 and the second label probability is 1.

[0182] Specifically, this includes: Step 41, taking each of the first sample records in the original sample set as the current sample record;

[0183] Step 42: Use the current sample record's sample type, brain T1-weighted image, brain fMRI image, and brain DTI image as the corresponding current sample type, current T1-weighted image, current fMRI image, and current DTI image.

[0184] In this context, each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a blood-oxygenation-level-dependent (BOLD) signal sequence, which consists of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values;

[0185] Step 43: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector.

[0186] Specifically, this includes: Step 431, calculating the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image;

[0187] Specifically, this includes: Step 4311, using the FSL tool to perform voxel space registration from the current fMRI image to the current T1-weighted image to obtain the corresponding first registration relationship; using the FSL tool to perform voxel space registration from the current DTI image to the current T1-weighted image to obtain the corresponding second registration relationship; using the FSL tool to perform voxel space registration from the current T1-weighted image to the MNI standard space corresponding to the first brain atlas to obtain the corresponding third registration relationship; and performing MNI standard space registration on the three-dimensional voxel coordinates of the current fMRI image based on the first and third registration relationships; and performing MNI standard space registration on the three-dimensional voxel coordinates of the current DTI image based on the second and third registration relationships.

[0188] Here, FSL tools refer to the analysis tools in the fMRIB (Functional Magnetic Resonance Imaging of the Brain) software library. This software library provides dedicated analysis tools for processing brain MRI, fMRI, DTI, and other imaging data. Common FSL tools used in this software library for image registration include FLIRT and FNIRT tools. Using FSL tools for brain region registration is a well-known technical solution in this field and will not be elaborated further here.

[0189] Step 4312: Use MRtrix3 or FSL tools to identify white matter fiber streamlines in the whole brain based on the current DTI images to obtain the corresponding white matter fiber streamline set.

[0190] The white matter fiber streamline set includes multiple white matter fiber streamlines; each white matter fiber streamline is composed of multiple fiber voxel points; the voxel features of each fiber voxel point include three-dimensional voxel coordinates and FA values; the first and last fiber voxel points of each white matter fiber streamline are recorded as the corresponding start point and end point;

[0191] Here, MRtrix3 is a processing tool specifically designed for DTI image analysis. MRtrix3 provides complete processing functions and workflow interfaces for whole-brain white matter fiber streamline analysis. In addition, the FMRIB software library also provides complete workflow interfaces and corresponding FSL tools for whole-brain white matter fiber streamline analysis. Using MRtrix3 or FSL tools for whole-brain white matter fiber streamline recognition is a well-known technical solution in this field and will not be elaborated further here.

[0192] Step 4313: Combine the i-th and j-th brain region sub-maps of the first brain map to form the corresponding brain region group P. i,j ; and group each brain region P i,jThe current brain region group is defined as follows: the two sub-image spaces in the current DTI image corresponding to the current brain region group are defined as the first and second sub-spaces respectively; white matter fiber streamlines are grouped together, and white matter fiber streamlines whose origin and end point fall in the first and second sub-spaces are recorded as the corresponding current group fiber streamlines; and the fiber streamlines obtained from all current group fiber streamlines are combined to form the corresponding fiber streamline subset F. i,j ;

[0193] Step 4314, divide each fiber streamline subset F i,j As the corresponding current subset; the total number of white matter fiber streamlines in the current subset is statistically analyzed, and the statistical result is used as the corresponding fiber total number feature; the average volume of the first and second subspaces corresponding to the current subset is calculated, and the corresponding fiber density feature is calculated from the fiber total number feature and the average volume = fiber total number feature / average volume; the streamline length of each white matter fiber streamline in the current subset is calculated, and the average length of all streamlines is used as the corresponding fiber length feature; the minimum FA value of each white matter fiber streamline in the current subset is identified, and the average of all minimum FA values ​​is used as the corresponding cross-sectional area index; the average of all FA values ​​in the current subset is calculated, and the calculation result is used as the corresponding FA feature; and a corresponding structural feature s is composed of the fiber total number feature, fiber density feature, fiber length feature, cross-sectional area index, and FA feature corresponding to the current subset. i,j ; and from the obtained N×N structural features s i,j Form the corresponding SC matrix;

[0194] Step 4315: Take each brain region sub-map of the first brain atlas as the current atlas; take the sub-space image corresponding to the current atlas in the current fMRI image as the current image; calculate the mean of all BOLD signal sequences in the current image to obtain the corresponding mean signal sequence; delete the mean of the first M BOLD signals in the mean signal sequence, where M is a preset positive integer; and extract the low-frequency signal from the mean signal sequence according to the preset low-frequency signal frequency range to obtain the corresponding low-frequency signal sequence.

[0195] Here, in this embodiment of the invention, the mean values ​​of the first M BOLD signals of the mean signal sequence are deleted in order to filter out unstable data in the early stage of signal acquisition;

[0196] The low-frequency signal frequency range of this embodiment of the invention is a pre-set low-frequency band, for example: 0.01-0.1Hz; the purpose of extracting low-frequency signals from the mean signal sequence based on this band is to retain low-frequency signal features related to neural activity and filter out physiological noise such as high-frequency noise, low-frequency drift, breathing and heartbeat.

[0197] Step 4316: Calculate the Pearson correlation coefficient between the two low-frequency signal sequences corresponding to the i-th and j-th brain region sub-maps in the first brain map, and perform a Fisher z-transform on the calculation result to obtain the corresponding functional feature f. i,j ; and from the obtained N×N functional features f i,j Form the corresponding FC matrix;

[0198] Step 432, and calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix;

[0199] Specifically, this includes: Step 4321, which involves dividing the structural features s of the SC matrix into... i,j As the current structural feature, the fiber total number, fiber density, fiber length, cross-sectional area index, and FA feature of the current structural feature are normalized to obtain the corresponding normalized total number, normalized density, normalized length, normalized cross-sectional area index, and normalized FA value. The six normalized data are then weighted according to six preset weighting coefficients to obtain six corresponding weighted values. The six weighted values ​​are then summed, and the result is used as the corresponding connection strength sd. i,j ; and the obtained N×N connection strengths sd i,j Form the corresponding intensity matrix SD;

[0200] Here, in the embodiments of the present invention, each structural feature s i,j The characteristics of total fiber count, fiber density, fiber length, cross-sectional area, and fiber affinity are denoted as the corresponding values. , , , , , The corresponding normalized total, normalized density, normalized length, normalized cross-sectional area index, and normalized FA value can be denoted as the corresponding values. , , , , , The six weighting coefficients can be denoted as the corresponding... , , , , , Therefore, the connection strength sd i,j The calculation method is as follows: ;

[0201] The strength matrix SD in this embodiment of the invention includes N×N connection strengths sd. i,j ;

[0202] Step 4322: Take the i-th row of the intensity matrix SD as the current row; calculate the corresponding cutoff position K=int(N×20%) based on the total number of brain regions N, where int() is the floor function; and sort the N-1 connection intensities sd of the current row in descending order. i,j≠i Sort the connections and select the top K connections by their strength sd. i,j The index j forms the corresponding index vector ID. i ;

[0203] Step 4323: Take the i-th row of the FC matrix as the current row; and use the N-1 functional features f of the current row. i,j≠i Form the corresponding observation function vector And based on the index vector ID corresponding to the current row. i Calculate the corresponding predictive function vector ;

[0204] Here, the present invention provides an embodiment for predicting the function vector. It consists of the corresponding N-1 predictive feature features composition;

[0205] Predictive Function Features of Embodiments of the Invention The calculation method is as follows:

[0206] ;

[0207] Step 4324, for each observation function vector Its corresponding predictive function vector The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. i ; and based on each coefficient ρ i Calculate the corresponding brain region coupling features c i ; and obtain N brain region coupling features c i This forms the corresponding brain region coupling feature sequence C;

[0208] Here, the brain region coupling feature c in the embodiment of the present invention i The calculation method is as follows:

[0209] ;

[0210] It should be noted that the result of (1-Spearman rank correlation coefficient ρ) is the SDI index;

[0211] Step 4325, N(N-1) / 2 connection strengths sd from the upper triangular matrix of the strength matrix SD, excluding diagonal matrix elements. i,j Form the corresponding vector VS ; and consists of N(N-1) / 2 functional features f, excluding diagonal matrix elements, in the upper triangular matrix of the FC matrix. i,j Form the corresponding vector V F ; and for vector V S sum vector V F The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. global ; and based on the coefficient ρ global Calculate the corresponding global coupling feature c global ;

[0212] Among them, the global coupling feature c global The calculation method is as follows:

[0213] ;

[0214] Step 433, and use the current SFC feature vector as the corresponding first training SFC feature vector;

[0215] Step 44: Identify the current sample type; if the current sample type is a patient sample, set the corresponding first label probability bit to 1 and the second label probability to 0; if the current sample type is a healthy sample, set the corresponding first label probability bit to 0 and the second label probability to 1; and form the corresponding first label vector from the current first label probability and the second label probability.

[0216] Step 45: A corresponding first data record is formed by the first training SFC feature vector and the first label vector corresponding to the current sample record;

[0217] Step 46, and the first dataset is composed of all the first data records obtained.

[0218] Step 5: Train the first prediction model based on the first dataset;

[0219] Specifically, it includes: Step 51, dividing the first dataset into two sub-datasets based on a preset first segmentation ratio, denoted as the corresponding first training set and first evaluation set;

[0220] Here, the first segmentation ratio in this embodiment of the invention is a preset ratio parameter, such as 8:2; both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is N. tr The total number of records N in the first evaluation set ev The ratio of the total number N tr :N ev Satisfies the first segmentation ratio;

[0221] Step 52: Input the first training SFC feature vector of each first data record in the first training set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And denote the first label vector of each first data record as the corresponding vector. ; and by each vector Its corresponding vector Form a corresponding first prediction-label pair ( , );

[0222] Where 1 ≤ index u ≤ N tr ;

[0223] Step 53, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L M The corresponding first loss value is obtained through calculation;

[0224] Here, the model loss function L in this embodiment of the invention M Based on cross-entropy loss function L CE Implementation, specifically:

[0225] ;

[0226] Step 54: Identify whether the first loss value meets the preset first loss value range; if yes, proceed to step 55; if no, based on the preset first model optimizer, move towards making the model loss function L... M The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 52 after this round of modulation is completed;

[0227] Here, the first loss value range in this embodiment of the invention is a pre-set numerical range; the first model optimizer includes the Adam optimizer and the SGD optimizer;

[0228] Step 55: Input the first training SFC feature vector of each first data record in the first evaluation set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And denote the first label vector of each first data record as the corresponding vector. ; and by each vector Its corresponding vector Form a corresponding second prediction-label pair ( , ); and based on all the obtained second prediction-label pairs ( , The accuracy, precision, recall, and F1 score are calculated to obtain the corresponding first accuracy, first precision, first recall, and first F1 score.

[0229] Where 1 ≤ index q ≤ N ev ;

[0230] Step 56: Identify whether the first accuracy, first precision, first recall, and first F1 score all satisfy their respective first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 51 to continue training; if yes, stop training and confirm that the training of the first prediction model has ended.

[0231] Here, the first accuracy range, the first precision range, the first recall range, and the first F1 score range in this embodiment of the invention are four preset numerical ranges.

[0232] Step 6: After model training is completed, receive multimodal MRI images of the brain of any subject as the current multimodal images; register various types of images of the current multimodal images according to the first brain atlas, and calculate the corresponding first SFC feature vector based on the registered images, including the corresponding structural matrix, functional matrix, and structural-functional coupling feature vector; input the first SFC feature vector into the first prediction model to obtain the corresponding first prediction vector; and take the classification type corresponding to the highest prediction probability in the first prediction vector as the current prediction result.

[0233] Specifically, this includes: Step 61, after the model training is completed, receiving multimodal MRI images of the brain of any subject as the current multimodal images;

[0234] Here, current multimodal imaging includes brain T1-weighted images, brain fMRI images, and brain DTI images;

[0235] Step 62, and register the various images of the current multimodal image according to the first brain map, and calculate the corresponding first SFC feature vector based on the registered image, the corresponding structure matrix, function matrix and structure-function coupling feature vector.

[0236] Specifically, this includes step 621, where the brain T1-weighted image, brain fMRI image, and brain DTI image of the current multimodal image are used as the corresponding current T1-weighted image, current fMRI image, and current DTI image.

[0237] In this context, each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence, the BOLD signal sequence consisting of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values;

[0238] Step 622: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector.

[0239] Specifically, this includes: Step 6221, calculating the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image;

[0240] Here, the current step 6221 is consistent with the processing flow of the previous step 431;

[0241] Specifically, this includes: step 62211, using the FSL tool to perform voxel space registration from the current fMRI image to the current T1-weighted image to obtain the corresponding first registration relationship; using the FSL tool to perform voxel space registration from the current DTI image to the current T1-weighted image to obtain the corresponding second registration relationship; using the FSL tool to perform voxel space registration from the current T1-weighted image to the MNI standard space corresponding to the first brain atlas to obtain the corresponding third registration relationship; and performing MNI standard space registration on the three-dimensional voxel coordinates of the current fMRI image based on the first and third registration relationships; and performing MNI standard space registration on the three-dimensional voxel coordinates of the current DTI image based on the second and third registration relationships.

[0242] Step 62212: Use MRtrix3 or FSL tools to identify white matter fiber streamlines in the whole brain based on the current DTI images to obtain the corresponding white matter fiber streamline set.

[0243] The white matter fiber streamline set includes multiple white matter fiber streamlines; each white matter fiber streamline is composed of multiple fiber voxel points; the voxel features of each fiber voxel point include three-dimensional voxel coordinates and FA values; the first and last fiber voxel points of each white matter fiber streamline are recorded as the corresponding start point and end point;

[0244] Step 62213: Combine the i-th and j-th brain region sub-maps of the first brain map to form the corresponding brain region group P. i,j ; and group each brain region P i,jThe current brain region group is defined as follows: the two sub-image spaces in the current DTI image corresponding to the current brain region group are defined as the first and second sub-spaces respectively; white matter fiber streamlines are grouped together, and white matter fiber streamlines whose origin and end point fall in the first and second sub-spaces are recorded as the corresponding current group fiber streamlines; and the fiber streamlines obtained from all current group fiber streamlines are combined to form the corresponding fiber streamline subset F. i,j ;

[0245] Step 62214, divide each fiber streamline subset F i,j As the corresponding current subset; the total number of white matter fiber streamlines in the current subset is statistically analyzed, and the statistical result is used as the corresponding fiber total number feature; the average volume of the first and second subspaces corresponding to the current subset is calculated, and the corresponding fiber density feature is calculated from the fiber total number feature and the average volume = fiber total number feature / average volume; the streamline length of each white matter fiber streamline in the current subset is calculated, and the average length of all streamlines is used as the corresponding fiber length feature; the minimum FA value of each white matter fiber streamline in the current subset is identified, and the average of all minimum FA values ​​is used as the corresponding cross-sectional area index; the average of all FA values ​​in the current subset is calculated, and the calculation result is used as the corresponding FA feature; and a corresponding structural feature s is composed of the fiber total number feature, fiber density feature, fiber length feature, cross-sectional area index, and FA feature corresponding to the current subset. i,j ; and from the obtained N×N structural features s i,j Form the corresponding SC matrix;

[0246] Step 62215: Take each brain region sub-map of the first brain atlas as the current atlas; take the sub-space image corresponding to the current atlas in the current fMRI image as the current image; calculate the mean of all BOLD signal sequences in the current image to obtain the corresponding mean signal sequence; delete the mean of the first M BOLD signals in the mean signal sequence, where M is a preset positive integer; and extract the low-frequency signal from the mean signal sequence according to the preset low-frequency signal frequency range to obtain the corresponding low-frequency signal sequence.

[0247] Step 62216: Calculate the Pearson correlation coefficient between the two low-frequency signal sequences corresponding to the i-th and j-th brain region sub-maps in the first brain map, and perform a Fisher z-transform on the calculation result to obtain the corresponding functional feature f. i,j ; and from the obtained N×N functional features f i,j Form the corresponding FC matrix;

[0248] Step 6222, and calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix;

[0249] Here, the current step 6222 is consistent with the processing flow of the previous step 432;

[0250] Specifically, this includes: step 62221, which involves dividing the structural features s of the SC matrix into... i,j As the current structural feature, the fiber total number, fiber density, fiber length, cross-sectional area index, and FA feature of the current structural feature are normalized to obtain the corresponding normalized total number, normalized density, normalized length, normalized cross-sectional area index, and normalized FA value. The six normalized data are then weighted according to six preset weighting coefficients to obtain six corresponding weighted values. The six weighted values ​​are then summed, and the result is used as the corresponding connection strength sd. i,j ; and the obtained N×N connection strengths sd i,j Form the corresponding intensity matrix SD;

[0251] Step 62222: Take the i-th row of the intensity matrix SD as the current row; calculate the corresponding cutoff position K=int(N×20%) based on the total number of brain regions N; and sort the N-1 connection intensities sd of the current row in descending order. i,j≠i Sort the connections and select the top K connections by their strength sd. i,j The index j forms the corresponding index vector ID. i ;

[0252] Step 62223: Take the i-th row of the FC matrix as the current row; and use the N-1 functional features f of the current row. i,j≠i Form the corresponding observation function vector And based on the index vector ID corresponding to the current row. i Calculate the corresponding predictive function vector ;

[0253] Among them, the predictive function vector It consists of the corresponding N-1 predictive feature features Composition; predictive functional characteristics The calculation method is as follows:

[0254] ;

[0255] Step 62224, for each observation function vector Its corresponding predictive function vector The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. i ; and based on each coefficient ρ i Calculate the corresponding brain region coupling features c i ; and obtain N brain region coupling features c iThis forms the corresponding brain region coupling feature sequence C;

[0256] Among them, brain region coupling features c i The calculation method is as follows:

[0257] ;

[0258] Step 62225, the N(N-1) / 2 connection strengths sd in the upper triangular matrix of the strength matrix SD (excluding diagonal matrix elements) i,j Form the corresponding vector V S ; and consists of N(N-1) / 2 functional features f, excluding diagonal matrix elements, in the upper triangular matrix of the FC matrix. i,j Form the corresponding vector V F ; and for vector V S sum vector V F The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. global ; and based on the coefficient ρ global Calculate the corresponding global coupling feature c global ;

[0259] Among them, the global coupling feature c global The calculation method is as follows:

[0260] ;

[0261] Step 6223, and use the current SFC feature vector as the corresponding first training SFC feature vector;

[0262] Step 63, and input the first SFC feature vector into the first prediction model to obtain the corresponding first prediction vector;

[0263] Step 64, and take the classification type corresponding to the highest predicted probability in the first prediction vector as the current prediction result.

[0264] Figure 3 This is a module structure diagram of a device for predicting early Alzheimer's disease based on SFC features, provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiments, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiments. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a brain region customization module 201, a model customization module 202, a sample acquisition module 203, a dataset preparation module 204, a model training module 205, and a model application module 206.

[0265] The brain region customization module 201 is used to set the first brain region set; and generate the first brain map corresponding to the first brain region set based on the preset standard brain map; and set the matrix and vector feature structure of the structure matrix, function matrix and structure-function coupling feature vector to obtain the corresponding SC matrix, FC matrix and SFC feature vector.

[0266] The model customization module 202 designs a neural network model for binary classification prediction of early Alzheimer's disease based on the vector structure of SFC feature vectors, denoted as the first prediction model. The first prediction model is used to perform binary classification prediction based on the SFC feature vectors input to the model and output the corresponding first prediction vector. The first prediction vector consists of first and second prediction probabilities. The classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively.

[0267] The sample acquisition module 203 is used to collect big data on multimodal MRI images of the brain of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample set; the multimodal MRI images of the brain include brain T1-weighted images, brain fMRI images, and brain DTI images, and the imaging time interval between any two types of images is lower than the preset interval duration threshold.

[0268] The dataset preparation module 204 is used to register the images of each sample in the original sample set according to the first brain map, and to calculate the structure matrix, function matrix and structure-function coupling feature vector of the sample based on the registered images. Based on the calculation results and sample type, the module generates corresponding training data to form the corresponding first dataset.

[0269] The model training module 205 trains the first prediction model based on the first dataset.

[0270] The model application module 206 is used to receive multimodal MRI images of the brain of any subject as the current multimodal images after the model training is completed; and to register various types of images of the current multimodal images according to the first brain atlas, and to calculate the corresponding first SFC feature vector based on the registered images, including the corresponding structural matrix, functional matrix and structural-functional coupling feature vector; and to input the first SFC feature vector into the first prediction model to predict and obtain the corresponding first prediction vector; and to take the classification type corresponding to the maximum prediction probability in the first prediction vector as the current prediction result.

[0271] The present invention provides a device for predicting early Alzheimer's disease based on SFC features, which can execute the method steps in the above method embodiments. Its implementation principle and technical effect are similar, and will not be repeated here.

[0272] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the brain region customization module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0273] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0274] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0275] Figure 4 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. Figure 4 As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the processing steps described in the foregoing embodiments. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0276] exist Figure 4The system bus 305 mentioned can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device.

[0277] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0278] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0279] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on SFC features. As described above, in this embodiment of the invention, a set of brain regions of interest (i.e., the first brain region set) is first customized based on multiple brain regions of interest related to early Alzheimer's disease, and a corresponding brain atlas (i.e., the first brain atlas) is customized for the first brain region set. The matrix and vector structures of the structural matrix (SC matrix), functional matrix (FC matrix), and SFC feature vectors are also customized. Then, a neural network model (i.e., the first prediction model) for binary classification prediction of early Alzheimer's disease is customized based on the vector structure of the SFC feature vectors. An original sample set is constructed by collecting large amounts of multimodal MRI images (T1-weighted images, fMRI images, and DTI images) of the brains of two groups (early AD patients and healthy individuals), and a first dataset is constructed based on the original sample set. Then, the first prediction model is trained based on the first dataset. After the model training is completed, multimodal MRI images of the brains of any subject are received, and the images of various types of the current multimodal images are registered according to the first brain atlas. The corresponding SFC feature vectors are calculated based on the registered images, and the first prediction model is used to perform binary classification prediction based on the current SFC feature vectors. The embodiments of the present invention can deeply explore the SFC relationship between SC network and FC network, and can perform binary classification prediction of AD in the early stage based on SFC features, which not only effectively makes up for the shortcomings of traditional schemes, but also improves the prediction accuracy.

[0280] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0281] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting early Alzheimer's disease based on SFC features, characterized in that, The method includes: Set up a first brain region set; generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and set up the matrix and vector feature structure of the structure matrix, function matrix, and structure-function coupling feature vector to obtain the corresponding SC matrix, FC matrix and SFC feature vector. Based on the vector structure of the SFC feature vector, a neural network model for binary classification prediction of early Alzheimer's disease is designed, denoted as the first prediction model. The first prediction model is used to perform binary classification prediction based on the SFC feature vector input to the model and output the corresponding first prediction vector. The first prediction vector is composed of a first and a second prediction probability. The classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively. The original sample set was obtained by collecting large data on multimodal MRI images of the brains of people with early Alzheimer's disease and healthy people; the multimodal MRI images of the brain include brain T1-weighted images, brain fMRI images, and brain DTI images, and the imaging time interval between any two types of images is less than a preset interval duration threshold. Based on the first brain map, the images of each sample in the original sample set are registered, and the structure matrix, function matrix and structure-function coupling feature vector of the sample are calculated based on the registered images. Based on the calculation results and the sample type, corresponding training data is generated to form the corresponding first dataset. The first prediction model is trained based on the first dataset; After model training is completed, the multimodal MRI images of the brain of any subject are received as the current multimodal images; and the images of various types of the current multimodal images are registered according to the first brain atlas, and the corresponding structural matrix, functional matrix and structural-functional coupling feature vector are calculated based on the registered images to obtain the corresponding first SFC feature vector; the first SFC feature vector is input into the first prediction model to predict to obtain the corresponding first prediction vector; and the classification type corresponding to the highest prediction probability in the first prediction vector is taken as the current prediction result.

2. The method for predicting early Alzheimer's disease based on SFC features according to claim 1, characterized in that, The first brain region set consists of multiple brain regions including the Default Mode Network (DMN), Sensorimotor Network (SMN), Visual Network (VN), Executive Control Network (ECN), and Dorsal Attention Network (DAN), with the total number of brain regions N being a positive integer. The standard brain atlas includes at least the Yeo-7 atlas, Yeo-17 atlas, Power Atlas atlas, BNA atlas, MMP atlas, and AAL atlas. The first brain atlas consists of N brain region sub-atlases. The voxel spaces of both the standard brain atlas and the first brain atlas are the MNI standard space. The SC matrix consists of N×N structural features s i,j Composition, 1 ≤ brain region index i, j ≤ N; the structural feature s i,j The characteristic components include total fiber count, fiber density, fiber length, cross-sectional area index, and fiber affinity (FA) characteristics. The total number of fibers is the total number of white matter fibers from brain region i to brain region j; the fiber density is the white matter fiber density from brain region i to brain region j; the fiber length is the average length of white matter fibers from brain region i to brain region j; the cross-sectional area index is the cross-sectional area index of white matter fibers from brain region i to brain region j; and the FA characteristic is the average FA value of white matter fibers from brain region i to brain region j. The FC matrix consists of N×N functional features f i,j Composition; the functional feature f i,j The correlation index of BOLD signals between brain region i and brain region j; The SFC feature vector consists of globally coupled feature c global It consists of a brain region coupling feature sequence C; the brain region coupling feature sequence C is composed of N brain region coupling features c i Composition; the global coupling feature c global The global SDI index of the SC matrix and the FC matrix; the brain region coupling feature c i The SDI index of the i-th brain region; The early Alzheimer's disease population includes multiple early Alzheimer's disease patients; The healthy population consists of multiple healthy individuals who do not have early-stage Alzheimer's disease; The original sample set includes multiple first sample records; The first sample record corresponds one-to-one with the early Alzheimer's disease population and the early Alzheimer's disease patients or healthy individuals in the healthy population; the first sample record includes the sample type, the brain T1-weighted image, the brain fMRI image, and the brain DTI image; the sample type includes patient samples and healthy samples; The first dataset includes multiple first data records; each first data record corresponds one-to-one with a first sample record; each first data record includes a first training SFC feature vector and a first label vector; the first label vector is composed of first and second label probabilities; the classification types of the first and second label probabilities correspond to disease type and healthy type, respectively; if the first sample type corresponding to the current data record is a patient sample, then the first label probability is 1 and the second label probability is 0; if the first sample type corresponding to the current data record is a healthy sample, then the first label probability is 0 and the second label probability is 1.

3. The method for predicting early Alzheimer's disease based on SFC features according to claim 2, characterized in that, The first and second model inputs of the first prediction model are respectively used to receive the global coupling feature c corresponding to the SFC feature vector. global and the brain region coupling feature sequence C{c i The model output is used to output the corresponding first prediction vector; The first prediction model includes a first fully connected layer, a first activation layer, a second fully connected layer, a first convolutional layer, a second activation layer, a first pooling layer, a third fully connected layer, a first fusion layer, a fourth fully connected layer, and a first classification layer; The input of the first fully connected layer is connected to the input of the first model, and its output is connected to the input of the first activation layer; the output of the first activation layer is connected to the input of the second fully connected layer; the output of the second fully connected layer is connected to the first input of the first fusion layer; the input of the first convolutional layer is connected to the input of the second model, and its output is connected to the input of the second activation layer; the output of the second activation layer is connected to the input of the first pooling layer; the output of the first pooling layer is connected to the input of the third fully connected layer; the output of the third fully connected layer is connected to the second input of the first fusion layer; the output of the first fusion layer is connected to the input of the fourth fully connected layer; the output of the fourth fully connected layer is connected to the input of the first classification layer; and the output of the first classification layer is connected to the model output. The first fully connected layer is used for the global coupling feature c global The first feature vector H1 obtained by feature encoding is sent to the first activation layer; The encoding method of the first feature vector H1 is as follows: ； W1 and B1 are the first weight matrix and the first bias vector of the model; the first weight matrix W1 and the first bias vector B1 both have the shape D1×1; D1 is the preset first feature dimension; the first feature vector H1 has the shape D1×1; The first activation layer is used to activate the first feature vector H1 using the ReLU activation function to obtain the corresponding second feature vector H2, which is then sent to the second fully connected layer. The activation method of the second feature vector H2 is as follows: ； The shape of the second feature vector H2 is D1×1; The second fully connected layer is used to perform feature mapping on the second feature vector H2 to obtain the corresponding third feature vector H3, which is then sent to the first fusion layer. The mapping method for the third feature vector H3 is as follows: ； W2 and B2 are the second weight matrix and the second bias vector of the model; the second weight matrix W2 has a shape of D2×D1, and the second bias vector B2 has a shape of D2×1; D2 is a preset second feature dimension; the third feature vector H3 has a shape of D2×1; The first convolutional layer is used to apply D3 1×1 convolutional kernels to each of the brain region coupling features c in the brain region coupling feature sequence C. i Performing a one-dimensional convolution operation yields a feature vector of length D3. ; and the resulting N feature vectors The fourth feature vector H4, which has a shape of N×D3, is sent to the second activation layer; Where D3 is the preset third feature dimension; The second activation layer is used to activate the fourth feature vector H4 using the ReLU activation function to obtain the corresponding fifth feature vector H5, which is then sent to the first pooling layer. The activation method of the fifth feature vector H5 is as follows: ； The shape of the fifth feature vector H5 is N×D3; The first pooling layer is used to perform global pooling on each feature channel of the fifth feature vector H5 to obtain the corresponding sixth feature vector H6, which is then sent to the third fully connected layer. The shape of the sixth feature vector H6 is 1×D3; The third fully connected layer is used to perform feature mapping processing on the sixth feature vector H6 to obtain the corresponding seventh feature vector H7 and send it to the first fusion layer. The mapping method for the seventh feature vector H7 is as follows: ； W3 and B3 are the third weight matrix and the third bias vector of the model; the third weight matrix W3 has a shape of D4×D3, and the third bias vector B3 has a shape of D4×1; D4 is a preset fourth feature dimension; the seventh feature vector H7 has a shape of D4×1. The first fusion layer is used to perform vector concatenation processing on the third feature vector H3 and the seventh feature vector H7 to obtain the corresponding eighth feature vector H8, which is then sent to the fourth fully connected layer. The shape of the eighth feature vector H8 is (D2+D4)×1; The fourth fully connected layer is used to perform binary classification feature mapping on the eighth feature vector H8 to obtain the corresponding ninth feature vector H9 and send it to the first classification layer. The mapping method for the ninth feature vector H9 is as follows: ； W4 and B4 are the fourth weight matrix and fourth bias vector of the model; the fourth weight matrix W4 has a shape of 2×(D2+D4), and the fourth bias vector B4 has a shape of 2×1; the ninth feature vector H9 has a shape of 2×1; the two vector data of the ninth feature vector H9 are denoted as corresponding... , ; The first classification layer is used to use the Softmax function to perform binary classification probability calculation based on the ninth feature vector H9 to obtain the corresponding first and second predicted probabilities, and outputs the corresponding first predicted vector composed of the two predicted probabilities. The first and second prediction probabilities are calculated as follows: ， 。 4. The method for predicting early Alzheimer's disease based on SFC features according to claim 2, characterized in that, The process of registering the images of each sample in the original sample set according to the first brain map, calculating the structure matrix, function matrix, and structure-function coupling feature vector of the sample based on the registered images, and generating corresponding training data based on the calculation results and sample type to form the corresponding first dataset specifically includes: Step 41: Take each of the first sample records in the original sample set as the current sample record; Step 42: The sample type, brain T1-weighted image, brain fMRI image, and brain DTI image recorded in the current sample are used as the corresponding current sample type, current T1-weighted image, current fMRI image, and current DTI image. Each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence, wherein the BOLD signal sequence consists of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values; Step 43: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector. Step 44: Identify the current sample type; if the current sample type is a patient sample, set the corresponding first label probability to 1 and the second label probability to 0; if the current sample type is a healthy sample, set the corresponding first label probability to 0 and the second label probability to 1; and form the corresponding first label vector from the current first label probability and the second label probability. Step 45: The first training SFC feature vector and the first label vector corresponding to the current sample record are combined to form a corresponding first data record; and all the obtained first data records are combined to form the corresponding first dataset.

5. The method for predicting early Alzheimer's disease based on SFC features according to claim 2, characterized in that, The step of training the first prediction model based on the first dataset specifically includes: Step 51: Based on a preset first segmentation ratio, the first dataset is divided into two sub-datasets, denoted as the first training set and the first evaluation set. Both the first training set and the first evaluation set consist of multiple first data records; the total number of records in the first training set is N. tr The total number of records N in the first evaluation set ev The ratio of the total number N tr :N ev The first segmentation ratio is satisfied; Step 52: Input the first training SFC feature vector of each first data record in the first training set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And record the first label vector of each of the first data records as the corresponding vector. ; and by each of the aforementioned vectors The corresponding vector Form a corresponding first prediction-label pair ( , ); Where 1 ≤ index u ≤ N tr ; Step 53, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L M The corresponding first loss value is obtained through calculation; Wherein, the model loss function L M Based on cross-entropy loss function L CE Implementation, specifically: ； Step 54: Identify whether the first loss value meets the preset first loss value range; if yes, proceed to step 55; if no, based on the preset first model optimizer, move towards making the model loss function L... M The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 52 after this round of modulation is completed; The first model optimizer includes the Adam optimizer and the SGD optimizer. Step 55: Input the first training SFC feature vector of each first data record in the first evaluation set into the first prediction model for processing to obtain the corresponding first prediction vector, denoted as the corresponding vector. And record the first label vector of each of the first data records as the corresponding vector. ; and by each of the aforementioned vectors The corresponding vector Form a corresponding second prediction-label pair ( , ); and based on all the obtained second prediction-label pairs ( , The accuracy, precision, recall, and F1 score are calculated to obtain the corresponding first accuracy, first precision, first recall, and first F1 score. Where 1 ≤ index q ≤ N ev ; Step 56: Identify whether the first accuracy, first precision, first recall, and first F1 score all satisfy their respective first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 51 to continue training; if yes, stop training and confirm that the training of the first prediction model has ended.

6. The method for predicting early Alzheimer's disease based on SFC features according to claim 2, characterized in that, The process of registering various images of the current multimodal imagery based on the first brain map and calculating the corresponding first SFC feature vector based on the registered images, including the structure matrix, function matrix, and structure-function coupling feature vector, specifically includes: Step 61: Use the brain T1-weighted image, brain fMRI image, and brain DTI image of the current multimodal image as the corresponding current T1-weighted image, current fMRI image, and current DTI image; Each voxel feature of the current T1-weighted image contains three-dimensional voxel coordinates; each voxel feature of the current fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence, wherein the BOLD signal sequence consists of a set of BOLD signal values ​​b ordered chronologically. t Composition, 1≤time index t≤T, where T is the time series length; each voxel feature of the current DTI image contains three-dimensional voxel coordinates and FA values; Step 62: Calculate the corresponding SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image; calculate the corresponding SFC feature vector based on the current SC matrix and FC matrix; and use the current SFC feature vector as the corresponding first training SFC feature vector.

7. The method for predicting early Alzheimer's disease based on SFC features according to any one of claims 4 or 6, characterized in that, The calculation of the SC matrix and FC matrix based on the first brain atlas, the current T1-weighted image, the current fMRI image, and the current DTI image specifically includes: Step 71: Use the FSL tool to perform voxel space registration from the current fMRI image to the current T1-weighted image to obtain the corresponding first registration relationship; use the FSL tool to perform voxel space registration from the current DTI image to the current T1-weighted image to obtain the corresponding second registration relationship; use the FSL tool to perform voxel space registration from the current T1-weighted image to the MNI standard space corresponding to the first brain atlas to obtain the corresponding third registration relationship; and perform MNI standard space registration on the three-dimensional voxel coordinates of the current fMRI image based on the first and third registration relationships; and perform MNI standard space registration on the three-dimensional voxel coordinates of the current DTI image based on the second and third registration relationships. Step 72: Use MRtrix3 or FSL tools to identify white matter fiber streamlines in the whole brain based on the current DTI images to obtain the corresponding white matter fiber streamline set. The white matter fiber streamline set includes multiple white matter fiber streamlines; each white matter fiber streamline is composed of multiple fiber voxel points; the voxel feature of each fiber voxel point includes three-dimensional voxel coordinates and FA value; the first and last fiber voxel points of each white matter fiber streamline are recorded as the corresponding start point and end point; Step 73: Combine the i-th and j-th brain region sub-maps of the first brain map to form a corresponding brain region group P. i,j ; and group each of the brain regions P i,j The current brain region group is defined as follows: the two sub-image spaces in the current DTI image corresponding to the current brain region group are defined as the corresponding first and second sub-spaces; the white matter fiber streamlines are grouped together, and the white matter fiber streamlines whose starting points and ending points fall on the first and second sub-spaces are recorded as the corresponding current group fiber streamlines; and all the current group fiber streamlines are combined to form the corresponding fiber streamline subset F. i,j ; Step 74, divide each of the fiber streamline subsets F i,j As the corresponding current subset; the total number of white matter fiber streamlines in the current subset is statistically analyzed, and the statistical result is used as the corresponding fiber total number feature; the average volume of the first and second subspaces corresponding to the current subset is calculated, and the fiber density feature is calculated from the fiber total number feature and the average volume = fiber total number feature / average volume; the streamline length of each white matter fiber streamline in the current subset is calculated, and the average length of all streamlines is used as the corresponding fiber length feature; the minimum FA value of each white matter fiber streamline in the current subset is identified, and the average of all minimum FA values ​​is used as the corresponding cross-sectional area index; the average of all FA values ​​in the current subset is calculated, and the calculation result is used as the corresponding FA feature; and the fiber total number feature, fiber density feature, fiber length feature, cross-sectional area index, and FA feature corresponding to the current subset are combined to form a corresponding structural feature s. i,j ; and the obtained N×N structural features s i,j Form the corresponding SC matrix; Step 75: Take each brain region sub-map of the first brain atlas as the current atlas; take the sub-space image corresponding to the current atlas in the current fMRI image as the current image; calculate the mean of all BOLD signal sequences in the current image to obtain the corresponding mean signal sequence; delete the mean of the first M BOLD signals in the mean signal sequence, where M is a preset positive integer; and extract the low-frequency signal from the mean signal sequence according to a preset low-frequency signal range to obtain the corresponding low-frequency signal sequence. Step 76: Calculate the Pearson correlation coefficient between the two low-frequency signal sequences corresponding to the i-th and j-th brain region sub-maps in the first brain map, and perform a Fisher z-transform on the calculation result to obtain the corresponding functional feature f. i,j ; and the obtained N×N functional features f i,j The corresponding FC matrix is ​​formed.

8. The method for predicting early Alzheimer's disease based on SFC features according to any one of claims 4 or 6, characterized in that, The calculation of the corresponding SFC feature vector based on the current SC matrix and FC matrix specifically includes: Step 81, convert each of the structural features s of the SC matrix i,j As the current structural feature; the total fiber count, fiber density, fiber length, cross-sectional area index, and FA feature of the current structural feature are normalized to obtain the corresponding normalized total count, normalized density, normalized length, normalized cross-sectional area index, and normalized FA value; the six normalized data are weighted according to six preset weighting coefficients to obtain the corresponding six weighted values; the six weighted values ​​are summed and the result is used as the corresponding connection strength sd. i,j ; and the obtained N×N connection strengths sd i,j Form the corresponding intensity matrix SD; Step 82: Take the i-th row of the intensity matrix SD as the current row; calculate the corresponding cutoff position K=int(N×20%) based on the total number of brain regions N, where int() is the floor function; and sort the N-1 connection intensities sd of the current row in descending order. i,j≠i Sort the connections by their top K strengths (sd). i,j The index j forms the corresponding index vector ID. i ; Step 83: Take the i-th row of the FC matrix as the current row; and use the N-1 functional features f of the current row. i,j≠i Form the corresponding observation function vector ; and based on the index vector ID corresponding to the current row i Calculate the corresponding predictive function vector ; Wherein, the prediction function vector It consists of the corresponding N-1 predictive feature features composition; The predictive functional features The calculation method is as follows: ； Step 84, for each of the observation function vectors The corresponding prediction function vector The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. i ; and based on each of the aforementioned coefficients ρ i Calculate the corresponding brain region coupling features c i ; and obtain N brain region coupling features c i This forms the corresponding brain region coupling feature sequence C; Among them, the brain region coupling feature c i The calculation method is as follows: ； Step 85, N(N-1) / 2 of the connection strengths sd, excluding diagonal matrix elements, are formed from the upper triangular matrix of the strength matrix SD. i,j Form the corresponding vector V S ; and consisting of N(N-1) / 2 functional features f, excluding diagonal matrix elements, in the upper triangular matrix of the FC matrix. i,j Form the corresponding vector V F ; and for the vector V S and the vector V F The corresponding coefficient ρ is obtained by calculating the Spearman rank correlation coefficient. global ; and based on the coefficient ρ global Calculate the corresponding global coupling feature c global ; Wherein, the global coupling feature c global The calculation method is as follows: 。 9. An apparatus for performing the method for predicting early Alzheimer's disease based on SFC features according to any one of claims 1-8, characterized in that, The device includes: a brain region customization module, a model customization module, a sample acquisition module, a dataset preparation module, a model training module, and a model application module; The brain region customization module is used to set up a first brain region set; and generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and set up the matrix and vector feature structure of the structure matrix, function matrix, and structure-function coupling feature vector to obtain the corresponding SC matrix, FC matrix and SFC feature vector. The model customization module designs a neural network model for binary classification prediction of early Alzheimer's disease based on the vector structure of the SFC feature vector, denoted as the first prediction model; the first prediction model is used to perform binary classification prediction based on the SFC feature vector input to the model and output the corresponding first prediction vector; the first prediction vector is composed of first and second prediction probabilities; the classification types of the first and second prediction probabilities correspond to disease type and healthy type, respectively; The sample acquisition module is used to collect large amounts of brain multimodal MRI images of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample set; the brain multimodal MRI images include brain T1-weighted images, brain fMRI images, and brain DTI images, and the imaging time interval between any two types of images is lower than a preset interval duration threshold. The dataset preparation module is used to register the images of each sample in the original sample set according to the first brain map, and to calculate the structure matrix, function matrix and structure-function coupling feature vector of the sample based on the registered images, and to generate corresponding training data based on the calculation results and sample type to form the corresponding first dataset. The model training module trains the first prediction model based on the first dataset; The model application module is used to receive the multimodal MRI images of the brain of any subject as the current multimodal images after the model training is completed; and to register various types of images of the current multimodal images according to the first brain atlas, and to calculate the corresponding first SFC feature vector based on the registered images by performing corresponding structural matrix, functional matrix and structural-functional coupling feature vector; and to input the first SFC feature vector into the first prediction model to predict and obtain the corresponding first prediction vector; and to take the classification type corresponding to the maximum prediction probability in the first prediction vector as the current prediction result.

10. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-8; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method described in any one of claims 1-8.

Images