A method and device for predicting early alzheimer's disease based on dynamic and static sfc features

Abstract

The embodiment of the present application relates to a kind of method and device for predicting early Alzheimer's disease based on dynamic and static SFC characteristics, the method comprises: setting brain region set, prediction model;The dynamic and static SFC characteristics of early Alzheimer's disease population and healthy population are carried out data set by big data acquisition;According to data set, prediction model is trained;After model training ends, the brain fMRI image and brain DTI image of arbitrary subject input by user are received, and dynamic and static SFC characteristics extraction processing is carried out based on subject image to obtain static SFC characteristic vector A, dynamic SFC characteristic matrix B;Characteristic vector A, feature matrix B are input into prediction model to obtain corresponding prediction vector, and the current user is fed back to feedback. The present application makes up the defect that conventional model cannot combine dynamic SFC characteristics, and can improve model prediction accuracy and generalization.

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 dynamic and static SFC features. Background Technology

[0002] Structural connectivity (SC) refers to the anatomical physical connections between brain regions, such as nerve fiber bundles between the cortex and subcortical nuclei, with white matter fiber bundles being a commonly observed object. SC matrices are commonly used to represent structural connectivity networks, with each matrix unit representing the strength of the structural connection, typically characterized by the density of white matter fiber bundles between brain regions. Because the two brain regions corresponding to the diagonal matrix units are the same, the values ​​of the diagonal units in the SC matrix are usually set to 0. SC matrices can be obtained by analyzing diffusion tensor imaging (DTI) images of the subjects using conventional image analysis tools (such as FSL and MRTrix3).

[0003] Functional connectivity (FC) between brain regions refers to the synchronicity or statistical dependence of blood oxygenation level dependent (BOLD) signals across different brain regions over time, reflecting dynamic collaborative patterns between these regions. FC matrices are commonly used to characterize functional connectivity networks, with each matrix unit representing the strength of the connectivity, typically characterized by the Pearson correlation coefficient of BOLD signals between brain regions. Because the two brain regions corresponding to the diagonal matrix units are the same, the values ​​of the diagonal units in the FC matrix are usually set to 0. FC matrices can be obtained by analyzing functional magnetic resonance imaging (fMRI) images of the subject using conventional image analysis tools (such as FSL).

[0004] The structure-function coupling (SFC) feature of a brain region is calculated by coupling the structural and functional connectivity strength vectors of that brain region. For brain region i, the non-zero values ​​in the i-th row of the SC / FC matrix are extracted to form the corresponding structure / functional connectivity strength vector, and the Pearson correlation coefficient between these two vectors is the SFC feature value of brain region i.

[0005] SFC features can be further divided into static SFC features and dynamic SFC features. Static SFC features can be obtained based on a set of SC / FC matrices using the calculation method described above, and the static SFC feature for each brain region is a scalar. Dynamic SFC features are relatively more complex. They require first dividing the BOLD signal sequence into multiple time segments based on a time-slicing mechanism, then calculating the FC matrix corresponding to each segment period, and then calculating a corresponding SFC feature scalar based on each set of SC / segment FC matrices while keeping the SC matrix unchanged. Finally, the dynamic SFC feature for each brain region is actually a feature sequence composed of multiple SFC feature scalars.

[0006] Numerous studies have shown that in the early stages of Alzheimer's Disease (AD), the SFC (Specialized Functional Context) characteristics of certain typical brain regions significantly decline. This weakening of SFC characteristics in these regions is closely related to early memory impairment and cognitive decline in AD. These typical brain regions include areas associated with various local brain networks such as the Default Mode Network (DMN), Sensorimotor Network (SMN), Visual Network (VN), Executive Control Network (ECN), and Dorsal Attention Network (DAN).

[0007] With the development and application of artificial intelligence technology, several neural network models based on static SFC features for early AD prediction have emerged in the field of early AD assistance technology. These models have made positive contributions to improving the efficiency of early AD diagnosis. However, technical analysis reveals that most of these conventional models only rely on static SFC features for prediction and analysis, neglecting dynamic SFC features and failing to make predictions based on the combined characteristics of static and dynamic SFC features. If dynamic and static SFC features could be used as a basis, the predictive accuracy and generalization ability of the models could be further improved. Summary of the Invention

[0008] 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 dynamic and static SFC features. This invention is based on a customized first brain region set of brain regions related to early AD prediction, and designs a neural network model capable of binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features as the first prediction model. A first dataset is obtained by collecting large amounts of dynamic and static SFC features from individuals with early Alzheimer's disease and healthy individuals. The first prediction model is trained based on this first dataset. After training, the first prediction model is used to make predictions based on the static SFC feature vector A and the dynamic SFC feature matrix B of any subject. This invention overcomes the technical deficiency of conventional models that cannot incorporate dynamic SFC features for prediction, and can effectively improve the prediction accuracy and generalization of the model.

[0009] To achieve the above objectives, a first aspect of the present invention provides a method for predicting early Alzheimer's disease based on dynamic and static SFC features, the method comprising: Set up a first set of brain regions; 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), and the total number of brain regions N is a positive integer; A neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features is set as the first prediction model. The first prediction model is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y. The static SFC feature vector A consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; the static feature a i The SFC feature value is the value of the i-th brain region; the dynamic SFC feature matrix B has an N×T shape and consists of N×T dynamic feature b values. i,j Composition, where 1 ≤ time index j ≤ T, and T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each of the dynamic features b i,j For the i-th brain region at time t j The corresponding SFC feature value; the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively; The first dataset was obtained by collecting dynamic and static SFC characteristics of early Alzheimer's disease patients and healthy individuals through big data analysis. The first prediction model is trained based on the first dataset; After model training is completed, the system receives fMRI and DTI images of the brain of any subject input by the user; and performs dynamic and static SFC feature extraction processing based on the subject images to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and inputs the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y, which is then fed back to the current user.

[0010] Preferably, the first dataset includes multiple first data records; each first data record includes a first training feature vector, a first training feature matrix, and a first label vector; the data format of the first training feature vector is consistent with that of the static SFC feature vector A, and the data format of the first training feature matrix is ​​consistent with that of the dynamic SFC feature matrix B; 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; one of the first and second label probabilities is 1 and the other is 0. Each voxel feature in the brain fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence S, which consists of a set of signal values ​​ordered chronologically; all BOLD signal sequences S in the brain fMRI image have the same sequence length. Each voxel feature in the brain DTI image contains three-dimensional voxel coordinates and anisotropy fraction (FA) values.

[0011] Preferably, the first model input terminal and the second model input terminal of the first prediction model are respectively used to receive the corresponding SFC feature vector A and the dynamic SFC feature matrix B, and the model output terminal is used to output the corresponding prediction vector Y; The first prediction model includes a first fully connected layer, a first activation layer, a second fully connected layer, a first encoder, a first self-attention layer, a first pooling layer, a first fusion layer, a third fully connected layer, a second activation 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 encoder is connected to the input of the second model, and its output is connected to the input of the first self-attention layer; the output of the first self-attention layer is connected to the first pooling layer; the output of the first pooling 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 third fully connected layer; the output of the third fully connected layer is connected to the input of the second activation layer; the output of the second activation 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 to perform feature encoding processing on the SFC feature vector A to obtain the corresponding feature vector H1 and send it to the first activation layer; The encoding method of the feature vector H1 is as follows: ; W1 and b1 are the first weight matrix and the first bias vector of the model; the shape of the SFC feature vector A is considered to be N×1; the shape of the first weight matrix W1 is D1×N; the shape of the first bias vector b1 is D1×1; the shape of the feature vector H1 is D1×1; D1 is a preset first feature dimension; The first activation layer is used to activate the feature vector H1 using the ReLU activation function to obtain the corresponding feature vector H2, which is then sent to the second fully connected layer. The activation method of the 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 processing on the feature vector H2 and send the corresponding feature vector H3 to the first fusion layer; The feature mapping method of the 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; the feature vector H3 has a shape of D2×1; D2 is a preset second feature dimension; The first encoder is implemented based on an LSTM model; the first encoder is used to process the T dynamic features b in each row of the dynamic SFC feature matrix B. i,j Extracted to form the corresponding feature sequence x i ; and use an LSTM model to process each of the aforementioned feature sequences x i Feature encoding is performed to obtain the corresponding encoding vector h. i ; and the resulting N encoded vectors h i The corresponding feature vector H4 is sent to the first self-attention layer; Wherein, the encoding vector h i The encoding method is as follows: ; LSTM() is the model inference function of the LSTM model, θ LSTM The model parameters of the LSTM model; the encoding vector h i The shape of the feature vector H4 is D3×1; the shape of the feature vector H4 is N×D3; D3 is a preset third feature dimension; The first self-attention layer is used to perform query, key, and value vector transformation on the feature vector H4 to obtain the corresponding query vector Q, key vector K, and value vector V; and to calculate the corresponding attention weight W based on the query vector Q and key vector K. att ; and based on the attention weight W att The value vector V is weighted to obtain the corresponding feature vector H5, which is then sent to the first pooling layer. The query vector Q, the key vector K, and the value vector V are calculated as follows: , , ; The attention weight W att The calculation method is as follows: ; The feature vector H5 is calculated as follows: ; W Q W K W V For the corresponding query vector weight matrix, key vector weight matrix, and value vector weight matrix, W Q W K W V The shapes of all vectors are D3×D4; the shapes of the query vector Q, the key vector K, and the value vector V are all N×D4; the attention weight W... att The shape of the feature vector H5 is N×N; the shape of the feature vector H5 is N×D4; D4 is a preset fourth feature dimension; The first pooling layer is used to perform average pooling on each feature dimension of the feature vector H5 to obtain the corresponding feature vector H6, which is then sent to the first fusion layer. The shape of the feature vector H6 is D4×1; The first fusion layer is used to perform feature concatenation processing on the feature vectors H3 and H6 to obtain the corresponding feature vector H7, which is then sent to the third fully connected layer. The shape of the feature vector H7 is (D2+D4)×1; The third fully connected layer is used to perform feature extraction processing on the feature vector H7 and send the corresponding feature vector H8 to the second activation layer; The feature extraction method for the feature vector H8 is as follows: ; 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 D5×(D2+D4), and the shape of the third bias vector b3 is D5×1; the shape of the feature vector H8 is D5×1; D5 is the preset fifth feature dimension; The second activation layer is used to activate the feature vector H8 using the ReLU activation function to obtain the corresponding feature vector H9, which is then sent to the fourth fully connected layer. The activation method of the feature vector H9 is as follows: ; The shape of the feature vector H9 is D5×1; The fourth fully connected layer is used to perform binary classification feature mapping on the feature vector H9, corresponding to the feature vector H. 10 Send to the first classification layer; Wherein, the feature vector H 10 The feature mapping method 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×D5, and the fourth bias vector b4 has a shape of 2×1; the feature vector H 10 The shape is 2×1, consisting of two characteristic scalars. , composition; The first classification layer is used to apply the Softmax function based on the feature vector H. 10 The corresponding prediction probabilities y1 and y2 are obtained by performing binary classification probability calculation, and the corresponding prediction vector Y is output. The prediction probabilities y1 and y2 are calculated as follows: , .

[0012] Preferably, the first dataset obtained by collecting large amounts of dynamic and static SFC features from early Alzheimer's disease patients and healthy individuals specifically includes: Step 41: Generate a corresponding dataset by collecting large datasets of brain fMRI images and brain DTI images from individuals in the early stages of Alzheimer's disease and healthy individuals during the same period. The dataset includes multiple acquisition records; each acquisition record includes an object type, the brain fMRI image, and the brain DTI image; the object type includes patient type and healthy type. Step 42: Generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and form a corresponding first brain region combination by any two brain region sub-maps in the first brain map. 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 is composed of N sub-maps of the brain regions; the voxel spaces of the standard brain atlas and the first brain atlas are both MNI standard spaces. Step 43: Take each of the acquisition records in the acquisition dataset as the current record; and take the object type, the brain fMRI image, and the brain DTI image of the current record as the current type, current fMRI image, and current DTI image; perform spatial alignment on the current fMRI image and the current DTI image, and perform MNI standard spatial registration on the spatially aligned current fMRI image and the current DTI image; and use a preset image analysis tool to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; The image analysis tools include at least the FSL tool and the MRtrix3 tool; 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 44: Construct the SC matrix corresponding to the first brain atlas based on the current DTI image, specifically as follows: A traversal is performed on all combinations of the first brain regions. During this traversal, the currently traversed combination of the first brain regions is taken as the current combination. The two brain region image spaces in the current DTI image corresponding to the current combination are taken as the corresponding first and second brain region spaces. The white matter fiber streamlines are grouped together, and the white matter fiber streamlines whose starting point and ending point fall on the first and second brain region spaces are recorded as the corresponding first fiber streamlines. The total number of the first fiber streamlines is counted to obtain the corresponding first total number. The average volume of the first and second brain region spaces is calculated. The corresponding white matter fiber bundle density is calculated from the first total number and the average volume = first total number / average volume. At the end of this traversal, the corresponding SC matrix is ​​constructed based on all the obtained white matter fiber bundle densities. Step 45: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI image, specifically as follows: Each brain region sub-map of the first brain atlas is taken as the current atlas; the brain region image corresponding to the current atlas in the current fMRI image is taken as the current image; and the mean signal sequence is obtained by averaging all the BOLD signal sequences S of the current image. ; and the mean signal sequence The mean of the first M BOLD signals is deleted, where M is a preset positive integer; and the mean signal sequence is processed according to a preset low-frequency band. Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence S. low ; and for each of the two low-frequency signal sequences S corresponding to the first brain region combination low The Pearson correlation coefficient is calculated and the corresponding functional connectivity strength is obtained by performing a Fisher z-transform on the coefficient; and an FC matrix is ​​constructed based on all the obtained functional connectivity strengths, and the constructed matrix is ​​denoted as the corresponding static FC matrix; Step 46: Based on the preset sliding step size and sequence segment length, slice the signal duration of the current fMRI image and divide the current fMRI image into corresponding T time-segment images P based on the slicing results. j Specifically: The time series of signal points corresponding to any BOLD signal sequence S in the current fMRI image is taken as the first time series; the first M signal point times of the first time series are deleted to obtain the corresponding second time series; the second time series is segmented into multiple corresponding first slice sequences according to the sliding step size and the sequence segment length; the total number of the first slice sequences is counted and the count result is taken as the corresponding total time T; and the corresponding time period of each first slice sequence is recorded as the time period seg.j ; and according to T time periods seg j The current fMRI image is processed to obtain T corresponding time-segment images P. j ; Images P for each of the aforementioned time periods j All BOLD signal sequences satisfy the corresponding time period seg j ; Step 47, based on T images P from the aforementioned time periods j The dynamic FC matrix sequence corresponding to the first brain map is constructed as follows: For all the images P of the aforementioned time periods j Perform one round of traversal; and during this round of traversal, the image P of the currently traversed time period is... j The image is used as the corresponding current time period image; and the i-th brain region sub-map in the first brain atlas is used as the corresponding current atlas i, and the brain region image in the current time period image corresponding to the current atlas i is used as the current brain region image, and the mean signal sequence is obtained by averaging all BOLD signal sequences of the current brain region image. and according to the low-frequency band, the mean signal sequence Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence. ; and for each of the two low-frequency signal sequences corresponding to the first brain region combination. The Pearson correlation coefficient is calculated, and the corresponding functional connectivity strength is obtained by performing a Fisher z-transform on the coefficient. Based on all functional connectivity strengths corresponding to the current time period image, an FC matrix is ​​constructed, and this constructed matrix is ​​denoted as the corresponding dynamic matrix FC. j ; and at the end of this round of traversal, the T dynamic matrices FC obtained are... j Form the corresponding dynamic FC matrix sequence; Step 48: Construct the corresponding first training feature vector and first training feature matrix based on the SC matrix, the static FC matrix, and the dynamic FC matrix sequence, specifically as follows: Extract the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and each of the dynamic matrices FC in the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connectivity strength vector. ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding first training feature vector is formed; and the obtained N×T dynamic features b i,j Form the corresponding first training feature matrix; Step 49: Identify the current type. If the current type is a patient type, set the corresponding first label probability to 1 and the second label probability to 0. If the current type is a healthy type, set the corresponding first label probability to 0 and the second label probability to 1. The first label vector is formed by the first label probability and the second label probability. The first data record is formed by the first training feature vector, the first training feature matrix, and the first label vector corresponding to the current record. Step 50: The first dataset is composed of all the first data records obtained.

[0013] Preferably, 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 satisfies the first segmentation ratio; Step 52: Take each of the first data records in the first training set as the corresponding current training record; and input the first training feature vector and the first training feature matrix of the current training record as the current static SFC feature vector A and the dynamic SFC feature matrix B into the first prediction model for processing; and record the prediction vector Y output by the model in this processing as the corresponding prediction vector. And record the first label vector of the current training record as the corresponding label vector. ; and the prediction vector corresponding to the current training record. and the label vector Form the corresponding first prediction-label pair ( , ); Where 1 ≤ index u ≤ total number of records N tr ; Step 53, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L tr The corresponding first loss value is obtained through calculation; Wherein, the model loss function L tr 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 it does, proceed to step 55; if not, based on the preset first model optimizer, move towards making the model loss function L... tr The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 52 when the modulation ends; The first model optimizer includes the Adam optimizer and the SGD optimizer. Step 55: Take each of the first data records in the first evaluation set as the corresponding current evaluation record; and input the first training feature vector and the first training feature matrix of the current evaluation record as the current static SFC feature vector A and the dynamic SFC feature matrix B into the first prediction model for processing; and form a corresponding second prediction-label pair with the prediction vector Y output by the model in this processing and the first label vector of the current evaluation record; and according to the obtained N ev The first precision, first accuracy, first recall, and first F1 score are calculated for each of the second prediction-label pairs. 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.

[0014] Preferably, the step of performing dynamic and static SFC feature extraction processing based on the subject's image to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B specifically includes: Step 61: The subject's brain fMRI image and brain DTI image are used as the corresponding current fMRI image and current DTI image; the current fMRI image and current DTI image are spatially aligned, and the spatially aligned current fMRI image and current DTI image are spatially registered using the MNI standard; the image analysis tool is used to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; Step 62: Construct the SC matrix corresponding to the first brain atlas based on the current DTI image; Step 63: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI image; Step 64: Based on the sliding step size and the sequence segment length, slice the signal duration of the current fMRI image and divide the current fMRI image into T corresponding time-segment images P based on the slicing results. j ; Step 65, based on T images P from the aforementioned time periods j Construct the dynamic FC matrix sequence corresponding to the first brain map; Step 66: Extract the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and each of the dynamic matrices FC in the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connection strength vector. ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding static SFC feature vector A is formed; and the resulting N×T dynamic features b i,j This forms the corresponding dynamic SFC feature matrix B.

[0015] A second aspect of the present invention provides an apparatus for implementing the method for predicting early Alzheimer's disease based on dynamic and static SFC features as described in the first aspect above. The apparatus includes: a brain region customization module, a model customization module, a data preparation module, a model training module, and a model application module. The brain region customization module is used to set the first brain region set; the first brain region set 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), and the total number of brain regions N is a positive integer; The model customization module is used to set up a neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features, denoted as the first prediction model; the first prediction model is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y; the static SFC feature vector A consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; the static feature a i The SFC feature value is the value of the i-th brain region; the dynamic SFC feature matrix B has an N×T shape and consists of N×T dynamic feature b values. i,j Composition, where 1 ≤ time index j ≤ T, and T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each of the dynamic features b i,j For the i-th brain region at time t j The corresponding SFC feature value; the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively; The data preparation module is used to collect large amounts of data on the dynamic and static SFC characteristics of early Alzheimer's disease patients and healthy people to obtain the corresponding first dataset. The model training module is used to train the first prediction model based on the first dataset; The model application module is used to receive fMRI and DTI images of the brain of any subject input by the user after model training is completed; and to perform dynamic and static SFC feature extraction processing based on the subject images to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and to input the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y, which is then fed back to the current user.

[0016] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver; 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; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0017] 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.

[0018] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on dynamic and static SFC features. As described above, this invention customizes a first set of brain regions related to early AD prediction and designs a neural network model capable of binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features as the first prediction model. A first dataset is obtained by collecting large amounts of dynamic and static SFC features from individuals with early Alzheimer's disease and healthy individuals. The first prediction model is trained based on this first dataset. After training, the first prediction model is used to make predictions based on the static SFC feature vector A and the dynamic SFC feature matrix B of any subject. This invention overcomes the technical deficiency of conventional models that cannot combine dynamic SFC features for prediction, improving the model's prediction accuracy and generalization. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a method for predicting early Alzheimer's disease based on dynamic and static SFC features provided in Embodiment 1 of the present invention; Figure 2 A schematic diagram of the modules of the first prediction model provided in Embodiment 1 of the present invention; Figure 3 This is a module structure diagram of a device for predicting early Alzheimer's disease based on dynamic and static SFC features, provided in Embodiment 2 of the present invention. Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0020] 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.

[0021] Embodiment 1 of this invention provides a method for predicting early Alzheimer's disease based on dynamic and static SFC features, such as... Figure 1 The schematic diagram of a method for predicting early Alzheimer's disease based on dynamic and static SFC features provided in Embodiment 1 of the present invention is shown. The method mainly includes the following steps: Step 1: Set up the first brain region set.

[0022] Here, the first brain region set in this embodiment of the invention 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.

[0023] Step 2: Set up a neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features, denoted as the first prediction model.

[0024] Here, the first prediction model in this embodiment of the invention is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y.

[0025] The static SFC feature vector A in this embodiment of the invention consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; static feature a i denoted as the SFC feature value of the i-th brain region.

[0026] The dynamic SFC feature matrix B in this embodiment of the invention has a shape of N×T, consisting of N×T dynamic features b. i,j Composition: 1 ≤ time index j ≤ T, where T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each dynamic feature b i,j For the i-th brain region at time t j The corresponding SFC feature value.

[0027] In this embodiment of the invention, the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively.

[0028] like Figure 2 As shown in the schematic diagram of the first prediction model provided in Embodiment 1 of the present invention, the first model input terminal and the second model input terminal of the first prediction model are respectively used to receive the corresponding SFC feature vector A and the dynamic SFC feature matrix B, and the model output terminal is used to output the corresponding prediction vector Y.

[0029] like Figure 2As shown, 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 encoder, a first self-attention layer, a first pooling layer, a first fusion layer, a third fully connected layer, a second activation layer, a fourth fully connected layer, and a first classification layer.

[0030] like Figure 2 As shown, the connection relationships of the model components of 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 encoder is connected to the input of the second model, and its output is connected to the input of the first self-attention layer; the output of the first self-attention layer is connected to the first pooling layer; the output of the first pooling 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 third fully connected layer; the output of the third fully connected layer is connected to the input of the second activation layer; the output of the second activation 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.

[0031] The model component functions of the first prediction model are shown below.

[0032] 1) First fully connected layer: In this embodiment of the invention, the first fully connected layer is used to perform feature encoding processing on the SFC feature vector A to obtain the corresponding feature vector H1, which is then sent to the first activation layer.

[0033] Here, the encoding method of feature vector H1 in this embodiment of the invention is as follows: ; Where W1 and b1 are the first weight matrix and the first bias vector of the model; the shape of the SFC feature vector A is considered to be N×1; the shape of the first weight matrix W1 is D1×N; the shape of the first bias vector b1 is D1×1; the shape of the feature vector H1 is D1×1; and D1 is the preset first feature dimension.

[0034] 2) First activation layer: In this embodiment of the invention, the first activation layer is used to activate the feature vector H1 using the ReLU activation function to obtain the corresponding feature vector H2, which is then sent to the second fully connected layer.

[0035] Here, the activation method of feature vector H2 in this embodiment of the invention is as follows: ; The shape of the second feature vector H2 is D1×1.

[0036] 3) Second fully connected layer: In this embodiment of the invention, the second fully connected layer is used to perform feature mapping processing on feature vector H2 and send the corresponding feature vector H3 to the first fusion layer.

[0037] Here, the feature mapping method of feature vector H3 in this embodiment of the invention is as follows: ; Where 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, the shape of the second bias vector b2 is D2×1; the shape of the feature vector H3 is D2×1; and D2 is the preset second feature dimension.

[0038] 4) First encoder: The first encoder in this embodiment of the invention is implemented based on the LSTM model.

[0039] The first encoder is used to process the T dynamic features b of each row of the dynamic SFC feature matrix B. i,j Extracted to form the corresponding feature sequence x i ; and use the LSTM model to process each feature sequence x i Feature encoding is performed to obtain the corresponding encoding vector h. i ; and the resulting N encoded vectors h i The corresponding feature vector H4 is then sent to the first self-attention layer.

[0040] Here, the encoding vector h in this embodiment of the invention i The encoding method is as follows: ; Where LSTM() is the model inference function of the LSTM model, θ LSTM The model parameters of the LSTM model; the encoding vector h i The shape of H4 is D3×1; the shape of H4 is N×D3; D3 is the preset third feature dimension.

[0041] 5) First layer of self-attention: The first self-attention layer in this embodiment of the invention is used to perform query, key, and value vector transformation on the feature vector H4 to obtain the corresponding query vector Q, key vector K, and value vector V; and to calculate the corresponding attention weight W based on the query vector Q and key vector K. att And based on attention weight W att The value vector V is weighted to obtain the corresponding feature vector H5, which is then sent to the first pooling layer.

[0042] Here, the calculation method for query vector Q, key vector K, and value vector V in this embodiment of the invention is as follows: , , ; Attention weight Watt The calculation method is as follows: ; The feature vector H5 is calculated as follows: ; Among them, W Q W K W V For the corresponding query vector weight matrix, key vector weight matrix, and value vector weight matrix, W Q W K W V The shapes of all vectors are D3×D4; the shapes of the query vector Q, key vector K, and value vector V are all N×D4; the attention weights W att The shape of H5 is N×N; the shape of the feature vector H5 is N×D4; D4 is the preset fourth feature dimension.

[0043] 6) First pooling layer: In this embodiment of the invention, the first pooling layer is used to perform average pooling on each feature dimension of the feature vector H5 to obtain the corresponding feature vector H6, which is then sent to the first fusion layer.

[0044] Here, the shape of the feature vector H6 in this embodiment of the invention is D4×1.

[0045] 7) First Fusion Layer: In this embodiment of the invention, the first fusion layer is used to perform feature concatenation processing on feature vectors H3 and H6 to obtain the corresponding feature vector H7, which is then sent to the third fully connected layer.

[0046] Here, the shape of the feature vector H7 in this embodiment of the invention is (D2+D4)×1.

[0047] 8) Third fully connected layer: In this embodiment of the invention, the third fully connected layer is used to perform feature extraction processing on feature vector H7 and send the corresponding feature vector H8 to the second activation layer.

[0048] Here, the feature extraction method for feature vector H8 in this embodiment of the invention is as follows: ; 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 D5×(D2+D4), and the shape of the third bias vector b3 is D5×1; the shape of the feature vector H8 is D5×1; D5 is the preset fifth feature dimension.

[0049] 9) Second activation layer: In this embodiment of the invention, the second activation layer is used to activate the feature vector H8 using the ReLU activation function to obtain the corresponding feature vector H9, which is then sent to the fourth fully connected layer.

[0050] Here, the activation method of feature vector H9 in this embodiment of the invention is as follows: ; The shape of the feature vector H9 is D5×1.

[0051] 10) Fourth fully connected layer: The fourth fully connected layer in this embodiment of the invention is used to perform binary classification feature mapping on the feature vector H9, corresponding to the feature vector H. 10 Send to the first classification layer.

[0052] Here, the feature vector H in the embodiment of the present invention 10 The feature mapping method is as follows: ; 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×D5, and the fourth bias vector b4 has a shape of 2×1; the feature vector H 10 The shape is 2×1, consisting of two characteristic scalars. , composition.

[0053] 11) First classification layer: The first classification layer in this embodiment of the invention is used to apply the Softmax function based on the feature vector H 10 The corresponding prediction probabilities y1 and y2 are obtained by performing binary classification probability calculation, and the corresponding prediction vector Y is output.

[0054] Here, the prediction probabilities y1 and y2 in this embodiment of the invention are calculated as follows: , .

[0055] Step 3: Collect big data on the dynamic and static SFC characteristics of people with early Alzheimer's disease and healthy people to obtain the corresponding first dataset; The first dataset includes multiple first data records; each first data record includes a first training feature vector, a first training feature matrix, and a first label vector; the data format of the first training feature vector is consistent with that of the static SFC feature vector A, and the data format of the first training feature matrix is ​​consistent with that of the dynamic SFC feature matrix B; 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; one of the first and second label probabilities is 1 and the other is 0. Specifically, this includes: Step 31, generating a corresponding dataset from large datasets of brain fMRI and brain DTI images of individuals from the early Alzheimer's disease population and healthy individuals during the same period; Here, the big data collection channels in this embodiment of the invention include multiple publicly available neuroimaging datasets, such as: Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, Open Access Series of Imaging Studies (OASIS) dataset, Dominantly Inherited Alzheimer Network Observational Study (DIAN-OBS) dataset, etc. The dataset acquired in this embodiment of the invention includes multiple acquisition records; the acquisition records include object types, brain fMRI images, and brain DTI images; the object types include patient types and healthy types; Each voxel feature of the brain fMRI image in this embodiment of the invention includes three-dimensional voxel coordinates and a BOLD signal sequence S. The BOLD signal sequence S consists of a set of signal values ​​ordered chronologically. All BOLD signal sequences S in the brain fMRI image have the same sequence length. Each voxel feature in the brain DTI image of this invention includes three-dimensional voxel coordinates and anisotropy fraction (FA) value. Step 32: Generate a first brain map corresponding to the first brain region set based on the preset standard brain map; and form a corresponding first brain region combination by any two brain region sub-maps in the first brain map; Here, 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 brain region sub-maps of the first brain atlas correspond one-to-one with the brain regions of the first brain region set; the voxel spaces of both the standard brain atlas and the first brain atlas are the MNI standard space; Step 33: Take each acquisition record of the dataset as the current record; and take the object type, brain fMRI image, and brain DTI image of the current record as the current type, current fMRI image, and current DTI image; perform spatial alignment on the current fMRI image and current DTI image, and perform MNI standard spatial registration on the spatially aligned current fMRI image and current DTI image; and use a preset image analysis tool to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; Here, the image analysis tools in this embodiment of the invention include at least the FSL tool and the MRtrix3 tool; The white matter fiber streamline set of this invention 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; Step 34: Construct the SC matrix corresponding to the first brain atlas based on the current DTI images; Specifically, this includes: performing a traversal of all first brain region combinations; during this traversal, the currently traversed first brain region combination is taken as the corresponding current combination; the image spaces of the two brain regions in the current DTI image corresponding to the current combination are taken as the corresponding first and second brain region spaces; white matter fiber streamlines are grouped together, and white matter fiber streamlines whose starting point and ending point fall on the first and second brain region spaces are recorded as the corresponding first fiber streamlines; the total number of first fiber streamlines is counted to obtain the corresponding first total number; the average volume of the first and second brain region spaces is calculated; and the corresponding white matter fiber bundle density is calculated from the first total number and the average volume = first total number / average volume; and at the end of this traversal, the corresponding SC matrix is ​​constructed based on all the obtained white matter fiber bundle densities. Step 35: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI images; Specifically, this includes: using each brain region sub-map of the first brain atlas as the current atlas; using the brain region image corresponding to the current atlas in the current fMRI image as the current image; and calculating the mean signal sequence by averaging all BOLD signal sequences S of the current image. ; and the mean signal sequence The mean of the first M BOLD signals is deleted, where M is a preset positive integer; and the mean signal sequence is processed according to a preset low-frequency band. Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence S. low ; and for each combination of the first brain regions, two low-frequency signal sequences S low The Pearson correlation coefficients are calculated and Fisher z-transforms are applied to the coefficients to obtain the corresponding functional connectivity strengths. Based on all the obtained functional connectivity strengths, FC matrices are constructed, and the constructed matrix is ​​denoted as the corresponding static FC matrix. Here, the low-frequency band in this embodiment of the invention is a pre-set low-frequency range, such as 0.01-0.1Hz, 0.01-0.08Hz, etc.; Step 36: Based on the preset sliding step size and sequence segment length, slice the signal duration of the current fMRI image and divide the current fMRI image into corresponding T time-segment images P based on the slicing results. j ; Specifically, this includes: taking the time series of signal points corresponding to any BOLD signal sequence S in the current fMRI image as the first time series; deleting the first M signal point times from the first time series to obtain the corresponding second time series; segmenting the second time series into multiple corresponding first slice sequences according to the sliding step size and sequence segment length; counting the total number of first slice sequences and using the result as the corresponding total time T; and recording the corresponding time period of each first slice sequence as the time period seg. j ; and divided into T time periods (seg) j The current fMRI image is cropped into T time-segment images P. j Images at various time periods P j All BOLD signal sequences satisfy the corresponding time period seg j ; Here, the sliding step size and sequence segment length are two preset step length parameters and sequence duration parameters; Step 37, based on the T time period images P j Construct the dynamic FC matrix sequence corresponding to the first brain map; Specifically, this includes: images P for all time periods. j Perform one round of traversal; and during this round of traversal, change the image P of the currently traversed time period. j The image is used as the corresponding current time period image; the i-th brain region sub-map in the first brain atlas is used as the corresponding current atlas i, and the brain region image corresponding to the current atlas i in the current time period image is used as the current brain region image. The mean signal sequence is obtained by averaging all BOLD signal sequences of the current brain region image. And according to the low frequency band, the mean signal sequence Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence. And for each combination of first brain regions, two low-frequency signal sequences were analyzed. The Pearson correlation coefficient was calculated, and the corresponding functional connectivity strength was obtained by performing a Fisher z-transform on the coefficient. Then, an FC matrix was constructed based on all functional connectivity strengths corresponding to the current time period image, and this constructed matrix was denoted as the corresponding dynamic matrix FC. j ; and at the end of this round of traversal, the T dynamic matrices FC obtained are... j This forms the corresponding dynamic FC matrix sequence; Step 38: Construct the corresponding first training feature vector and first training feature matrix based on the SC matrix, static FC matrix, and dynamic FC matrix sequence; Specifically, this includes extracting the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and the dynamic matrices FC of the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connectivity strength vector. ; and for vectors , The Pearson correlation coefficient was calculated and the result was used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient was calculated and the result was used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding first training feature vector is formed; and the resulting N×T dynamic features b i,j Form the corresponding first training feature matrix; Step 39: Identify the current type. If the current type is a patient type, set the corresponding first label probability to 1 and the second label probability to 0. If the current type is a health-related type, set the corresponding first label probability to 0 and the second label probability to 1. The first label vector is composed of the first label probability and the second label probability. The first data record is composed of the first training feature vector, the first training feature matrix and the first label vector corresponding to the current record. Step 40: The first dataset is composed of all the first data records obtained.

[0056] Step 4: Train the first prediction model based on the first dataset; Specifically, it includes: Step 41, 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; 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 satisfies the first division ratio; Step 42: Take each first data record of the first training set as the corresponding current training record; and input the first training feature vector and the first training feature matrix of the current training record as the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model for processing; and record the prediction vector Y output by the model in this processing as the corresponding prediction vector. And record the first label vector of the current training record as the corresponding label vector. ; and the prediction vector corresponding to the current training record. and label vector Form the corresponding first prediction-label pair ( , ); Where 1 ≤ index u ≤ total number of records N tr ; Step 43, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L tr The corresponding first loss value is obtained through calculation; Here, the model loss function L in this embodiment of the invention tr Based on cross-entropy loss function L CE Implementation, specifically: ; Step 44: Identify whether the first loss value meets the preset first loss value range; if it does, proceed to step 45; if not, based on the preset first model optimizer, move towards making the model loss function L... tr The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 42 when the current round of modulation ends; 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; Step 45: Take each first data record of the first evaluation set as the corresponding current evaluation record; and input the first training feature vector and first training feature matrix of the current evaluation record as the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model for processing; and form the corresponding second prediction-label pair with the prediction vector Y output by the model in this processing and the first label vector of the current evaluation record; and according to the obtained N ev The first precision, first accuracy, first recall, and first F1 score are calculated for each second prediction-label pair. Step 46: 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 41 to continue training; if yes, stop training and confirm that the training of the first prediction model has ended.

[0057] 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.

[0058] Step 5: After the model training is completed, receive the fMRI and DTI images of the brain of any subject input by the user; and perform dynamic and static SFC feature extraction processing based on the subject images to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and input the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y, which is then fed back to the current user. Specifically, this includes: Step 51, after the model training is completed, receiving the brain fMRI image and brain DTI image of any subject input by the user; Each voxel feature in a brain fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence S, which consists of a set of signal values ​​ordered chronologically. All BOLD signal sequences S in brain fMRI images have the same sequence length. Each voxel feature in a brain DTI image contains three-dimensional voxel coordinates and anisotropy fraction FA values. Step 52, and perform dynamic and static SFC feature extraction processing based on the subject's image to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; Specifically, this includes: Step 521, using the subject's brain fMRI image and brain DTI image as the corresponding current fMRI image and current DTI image; spatially aligning the current fMRI image and current DTI image, and performing MNI standard spatial registration on the spatially aligned current fMRI image and current DTI image; and using image analysis tools to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; Step 522: Construct the SC matrix corresponding to the first brain atlas based on the current DTI images; Here, the current step 522 is consistent with the processing flow of step 34 above, and will not be repeated here; Step 523: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI images; Here, the current step 523 is consistent with the processing flow of step 35 above, and will not be repeated here; Step 524: Slice the signal duration of the current fMRI image based on the sliding step size and sequence segment length, and divide the current fMRI image into corresponding T time-segment images P based on the slicing results. j ; Here, the current step 524 is consistent with the processing flow of step 36 above, and will not be repeated here; Step 525, based on the T time period images P jConstruct the dynamic FC matrix sequence corresponding to the first brain map; Here, the current step 525 is consistent with the processing flow of step 37 above, and will not be repeated here; Step 526: Extract the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and the dynamic matrices FC of the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connectivity strength vector. ; and for vectors , The Pearson correlation coefficient was calculated and the result was used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient was calculated and the result was used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding static SFC feature vector A is formed; and the resulting N×T dynamic features b i,j This forms the corresponding dynamic SFC feature matrix B; Step 53: Input the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to make a prediction and obtain the corresponding prediction vector Y, which is then fed back to the current user.

[0059] Figure 3 This is a module structure diagram of a device for predicting early Alzheimer's disease based on dynamic and static 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 data preparation module 203, a model training module 204, and a model application module 205.

[0060] The brain region customization module 201 is used to set the first brain region set; the first brain region set consists of multiple brain regions, including the default mode network DMN, the sensorimotor network SMN, the visual network VN, the executive control network ECN, and the dorsal attention network DAN, and the total number of brain regions N is a positive integer.

[0061] Model customization module 202 is used to set up a neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features, denoted as the first prediction model; the first prediction model is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y; the static SFC feature vector A consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; static feature a i The SFC feature value of the i-th brain region; the shape of the dynamic SFC feature matrix B is N×T, consisting of N×T dynamic feature bs. i,j Composition: 1 ≤ time index j ≤ T, where T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each dynamic feature b i,j For the i-th brain region at time t j The corresponding SFC feature value; the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively.

[0062] The data preparation module 203 is used to collect big data on the dynamic and static SFC characteristics of people with early Alzheimer's disease and healthy people to obtain the corresponding first dataset.

[0063] The model training module 204 is used to train the first prediction model based on the first dataset.

[0064] The model application module 205 is used to receive the brain fMRI image and brain DTI image of any subject input by the user after the model training is completed; and to perform dynamic and static SFC feature extraction processing based on the subject image to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and to input the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y and feed it back to the current user.

[0065] The present invention provides a device for predicting early Alzheimer's disease based on dynamic and static SFC features. It can execute the method steps in the above method embodiments, and its implementation principle and technical effect are similar, so they will not be repeated here.

[0066] 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.

[0067] 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).

[0068] 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)).

[0069] 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.

[0070] 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.

[0071] 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.

[0072] 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.

[0073] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on dynamic and static SFC features. As described above, this invention customizes a first set of brain regions related to early AD prediction and designs a neural network model capable of binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features as the first prediction model. A first dataset is obtained by collecting large amounts of dynamic and static SFC features from individuals with early Alzheimer's disease and healthy individuals. The first prediction model is trained based on this first dataset. After training, the first prediction model is used to make predictions based on the static SFC feature vector A and the dynamic SFC feature matrix B of any subject. This invention overcomes the technical deficiency of conventional models that cannot combine dynamic SFC features for prediction, improving the model's prediction accuracy and generalization.

[0074] 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.

[0075] 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 dynamic and static SFC features, characterized in that, The method includes: Set up a first set of brain regions; 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), and the total number of brain regions N is a positive integer; A neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features is set as the first prediction model. The first prediction model is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y. The static SFC feature vector A consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; the static feature a i The SFC feature value is the value of the i-th brain region; the dynamic SFC feature matrix B has an N×T shape and consists of N×T dynamic feature b values. i,j Composition, where 1 ≤ time index j ≤ T, and T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each of the dynamic features b i,j For the i-th brain region at time t j The corresponding SFC feature value; the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively; The first dataset was obtained by collecting dynamic and static SFC characteristics of early Alzheimer's disease patients and healthy individuals through big data analysis. The first prediction model is trained based on the first dataset; After model training is completed, the system receives fMRI and DTI images of the brain of any subject input by the user; and performs dynamic and static SFC feature extraction processing based on the subject images to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and inputs the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y, which is then fed back to the current user.

2. The method for predicting early Alzheimer's disease based on dynamic and static SFC features according to claim 1, characterized in that, The first dataset includes multiple first data records; each first data record includes a first training feature vector, a first training feature matrix, and a first label vector; the data format of the first training feature vector is consistent with that of the static SFC feature vector A, and the data format of the first training feature matrix is ​​consistent with that of the dynamic SFC feature matrix B; 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; one of the first and second label probabilities is 1 and the other is 0. Each voxel feature in the brain fMRI image contains three-dimensional voxel coordinates and a BOLD signal sequence S, which consists of a set of signal values ​​ordered chronologically; all BOLD signal sequences S in the brain fMRI image have the same sequence length. Each voxel feature in the brain DTI image contains three-dimensional voxel coordinates and anisotropy fraction (FA) values.

3. The method for predicting early Alzheimer's disease based on dynamic and static SFC features according to claim 2, characterized in that, The first model input terminal and the second model input terminal of the first prediction model are respectively used to receive the corresponding SFC feature vector A and the dynamic SFC feature matrix B, and the model output terminal is used to output the corresponding prediction vector Y. The first prediction model includes a first fully connected layer, a first activation layer, a second fully connected layer, a first encoder, a first self-attention layer, a first pooling layer, a first fusion layer, a third fully connected layer, a second activation 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 encoder is connected to the input of the second model, and its output is connected to the input of the first self-attention layer; the output of the first self-attention layer is connected to the first pooling layer; the output of the first pooling 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 third fully connected layer; the output of the third fully connected layer is connected to the input of the second activation layer; the output of the second activation 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 to perform feature encoding processing on the SFC feature vector A to obtain the corresponding feature vector H1 and send it to the first activation layer; The encoding method of the feature vector H1 is as follows: ； W1 and b1 are the first weight matrix and the first bias vector of the model; the shape of the SFC feature vector A is considered to be N×1; the shape of the first weight matrix W1 is D1×N; the shape of the first bias vector b1 is D1×1; the shape of the feature vector H1 is D1×1; D1 is a preset first feature dimension; The first activation layer is used to activate the feature vector H1 using the ReLU activation function to obtain the corresponding feature vector H2, which is then sent to the second fully connected layer. The activation method of the 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 processing on the feature vector H2 and send the corresponding feature vector H3 to the first fusion layer; The feature mapping method of the 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; the feature vector H3 has a shape of D2×1; D2 is a preset second feature dimension; The first encoder is implemented based on an LSTM model; the first encoder is used to process the T dynamic features b in each row of the dynamic SFC feature matrix B. i,j Extracted to form the corresponding feature sequence x i ; and use an LSTM model to process each of the aforementioned feature sequences x i Feature encoding is performed to obtain the corresponding encoding vector h. i ; and the resulting N encoded vectors h i The corresponding feature vector H4 is sent to the first self-attention layer; Wherein, the encoding vector h i The encoding method is as follows: ； LSTM() is the model inference function of the LSTM model, θ LSTM The model parameters of the LSTM model; the encoding vector h i The shape of the feature vector H4 is D3×1; the shape of the feature vector H4 is N×D3; D3 is a preset third feature dimension; The first self-attention layer is used to perform query, key, and value vector transformation on the feature vector H4 to obtain the corresponding query vector Q, key vector K, and value vector V; and to calculate the corresponding attention weight W based on the query vector Q and key vector K. att ; and based on the attention weight W att The value vector V is weighted to obtain the corresponding feature vector H5, which is then sent to the first pooling layer. The query vector Q, the key vector K, and the value vector V are calculated as follows: ， ， ； The attention weight W att The calculation method is as follows: ; The feature vector H5 is calculated as follows: ; W Q W K W V For the corresponding query vector weight matrix, key vector weight matrix, and value vector weight matrix, W Q W K W V The shapes of all vectors are D3×D4; the shapes of the query vector Q, the key vector K, and the value vector V are all N×D4; the attention weight W... att The shape of the feature vector H5 is N×N; the shape of the feature vector H5 is N×D4; D4 is a preset fourth feature dimension; The first pooling layer is used to perform average pooling on each feature dimension of the feature vector H5 to obtain the corresponding feature vector H6, which is then sent to the first fusion layer. The shape of the feature vector H6 is D4×1; The first fusion layer is used to perform feature concatenation processing on the feature vectors H3 and H6 to obtain the corresponding feature vector H7, which is then sent to the third fully connected layer. The shape of the feature vector H7 is (D2+D4)×1; The third fully connected layer is used to perform feature extraction processing on the feature vector H7 and send the corresponding feature vector H8 to the second activation layer; The feature extraction method for the feature vector H8 is as follows: ； 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 D5×(D2+D4), and the shape of the third bias vector b3 is D5×1; the shape of the feature vector H8 is D5×1; D5 is the preset fifth feature dimension; The second activation layer is used to activate the feature vector H8 using the ReLU activation function to obtain the corresponding feature vector H9, which is then sent to the fourth fully connected layer. The activation method of the feature vector H9 is as follows: ； The shape of the feature vector H9 is D5×1; The fourth fully connected layer is used to perform binary classification feature mapping on the feature vector H9, corresponding to the feature vector H. 10 Send to the first classification layer; Wherein, the feature vector H 10 The feature mapping method 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×D5, and the fourth bias vector b4 has a shape of 2×1; the feature vector H 10 The shape is 2×1, consisting of two characteristic scalars. , composition; The first classification layer is used to apply the Softmax function based on the feature vector H 10 The corresponding prediction probabilities y1 and y2 are obtained by performing binary classification probability calculation, and the corresponding prediction vector Y is output. The prediction probabilities y1 and y2 are calculated as follows: ， 。 4. The method for predicting early Alzheimer's disease based on dynamic and static SFC features according to claim 2, characterized in that, The first dataset obtained by collecting large amounts of dynamic and static SFC features from early Alzheimer's disease patients and healthy individuals specifically includes: Step 41: Generate a corresponding dataset by collecting large datasets of brain fMRI images and brain DTI images from individuals in the early stages of Alzheimer's disease and healthy individuals during the same period. The dataset includes multiple acquisition records; each acquisition record includes an object type, the brain fMRI image, and the brain DTI image; the object type includes patient type and healthy type. Step 42: Generate a first brain map corresponding to the first brain region set based on a preset standard brain map; and form a corresponding first brain region combination by any two brain region sub-maps in the first brain map. 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 is composed of N sub-maps of the brain regions; the voxel spaces of the standard brain atlas and the first brain atlas are both MNI standard spaces. Step 43: Take each of the acquisition records in the acquisition dataset as the current record; and take the object type, the brain fMRI image, and the brain DTI image of the current record as the current type, current fMRI image, and current DTI image; perform spatial alignment on the current fMRI image and the current DTI image, and perform MNI standard spatial registration on the spatially aligned current fMRI image and the current DTI image; and use a preset image analysis tool to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; The image analysis tools include at least the FSL tool and the MRtrix3 tool; 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 44: Construct the SC matrix corresponding to the first brain atlas based on the current DTI image, specifically as follows: A traversal is performed on all combinations of the first brain regions. During this traversal, the currently traversed combination of the first brain regions is taken as the current combination. The two brain region image spaces in the current DTI image corresponding to the current combination are taken as the corresponding first and second brain region spaces. The white matter fiber streamlines are grouped together, and the white matter fiber streamlines whose starting and ending points fall on the first and second brain region spaces, respectively, are recorded as the corresponding first fiber streamlines. The total number of the first fiber streamlines is counted to obtain the corresponding first total number. The average volume of the first and second brain region spaces is calculated. The white matter fiber bundle density is obtained by dividing the first total number by the average volume. At the end of this traversal, the corresponding SC matrix is ​​constructed based on all the obtained white matter fiber bundle densities. Step 45: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI image, specifically as follows: Each brain region sub-map of the first brain atlas is taken as the current atlas; the brain region image corresponding to the current atlas in the current fMRI image is taken as the current image; and the mean signal sequence is obtained by averaging all the BOLD signal sequences S of the current image. ; and the mean signal sequence The mean of the first M BOLD signals is deleted, where M is a preset positive integer; and the mean signal sequence is processed according to a preset low-frequency band. Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence S. low ; and for each of the two low-frequency signal sequences S corresponding to the first brain region combination low The Pearson correlation coefficient is calculated and the corresponding functional connectivity strength is obtained by performing a Fisher z-transform on the coefficient; and an FC matrix is ​​constructed based on all the obtained functional connectivity strengths, and the constructed matrix is ​​denoted as the corresponding static FC matrix; Step 46: Based on the preset sliding step size and sequence segment length, slice the signal duration of the current fMRI image and divide the current fMRI image into corresponding T time-segment images P based on the slicing results. j Specifically: The time series of signal points corresponding to any BOLD signal sequence S in the current fMRI image is taken as the first time series; the first M signal point times of the first time series are deleted to obtain the corresponding second time series; the second time series is segmented into multiple corresponding first slice sequences according to the sliding step size and the sequence segment length; the total number of the first slice sequences is counted and the count result is taken as the corresponding total time T; and the corresponding time period of each first slice sequence is recorded as the time period seg. j ; and according to T time periods seg j The current fMRI image is processed to obtain T corresponding time-segment images P. j ; each of the aforementioned time period images P j All BOLD signal sequences satisfy the corresponding time period seg j ; Step 47, based on T images P from the aforementioned time periods j The dynamic FC matrix sequence corresponding to the first brain map is constructed as follows: For all the time period images P j Perform one round of traversal; and during this round of traversal, the image P of the currently traversed time period is... j The image is used as the corresponding current time period image; and the i-th brain region sub-map in the first brain atlas is used as the corresponding current atlas i, and the brain region image in the current time period image corresponding to the current atlas i is used as the current brain region image, and the mean signal sequence is obtained by averaging all BOLD signal sequences of the current brain region image. and according to the low-frequency band, the mean signal sequence Low-frequency signal extraction is performed to obtain the corresponding low-frequency signal sequence. ; and for each of the two low-frequency signal sequences corresponding to the first brain region combination. The Pearson correlation coefficient is calculated, and the corresponding functional connectivity strength is obtained by performing a Fisher z-transform on the coefficient. Based on all functional connectivity strengths corresponding to the current time period image, an FC matrix is ​​constructed, and this constructed matrix is ​​denoted as the corresponding dynamic matrix FC. j ; and at the end of this round of traversal, the T dynamic matrices FC obtained are... j Form the corresponding dynamic FC matrix sequence; Step 48: Construct the corresponding first training feature vector and first training feature matrix based on the SC matrix, the static FC matrix, and the dynamic FC matrix sequence, specifically as follows: Extract the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and each of the dynamic matrices FC in the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connectivity strength vector. ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding first training feature vector is formed; and the obtained N×T dynamic features b i,j Form the corresponding first training feature matrix; Step 49: Identify the current type. If the current type is a patient type, set the corresponding first label probability to 1 and the second label probability to 0. If the current type is a healthy type, set the corresponding first label probability to 0 and the second label probability to 1. The first label vector is formed by the first label probability and the second label probability. The first data record is formed by the first training feature vector, the first training feature matrix, and the first label vector corresponding to the current record. Step 50: The first dataset is composed of all the first data records obtained.

5. The method for predicting early Alzheimer's disease based on dynamic and static 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 satisfies the first segmentation ratio; Step 52: Take each of the first data records in the first training set as the corresponding current training record; and input the first training feature vector and the first training feature matrix of the current training record as the current static SFC feature vector A and the dynamic SFC feature matrix B into the first prediction model for processing; and record the prediction vector Y output by the model in this processing as the corresponding prediction vector. And record the first label vector of the current training record as the corresponding label vector. ; and the prediction vector corresponding to the current training record. and the label vector Form the corresponding first prediction-label pair ( , ); Where 1 ≤ index u ≤ total number of records N tr ; Step 53, obtain N tr The first prediction-label pair ( , Substitute the preset model loss function L tr The corresponding first loss value is obtained through calculation; Wherein, the model loss function L tr 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 it does, proceed to step 55; if not, based on the preset first model optimizer, move towards making the model loss function L... tr The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 52 when the modulation ends; The first model optimizer includes the Adam optimizer and the SGD optimizer. Step 55: Take each of the first data records in the first evaluation set as the corresponding current evaluation record; and input the first training feature vector and the first training feature matrix of the current evaluation record as the current static SFC feature vector A and the dynamic SFC feature matrix B into the first prediction model for processing; and form a corresponding second prediction-label pair with the prediction vector Y output by the model in this processing and the first label vector of the current evaluation record; and according to the obtained N ev The first precision, first accuracy, first recall, and first F1 score are calculated for each of the second prediction-label pairs. 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 dynamic and static SFC features according to claim 4, characterized in that, The step of extracting static and dynamic SFC features from the subject's image to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B specifically includes: Step 61: The subject's brain fMRI image and brain DTI image are used as the corresponding current fMRI image and current DTI image; the current fMRI image and current DTI image are spatially aligned, and the spatially aligned current fMRI image and current DTI image are spatially registered using the MNI standard; the image analysis tool is used to identify the white matter fiber streamlines of the whole brain based on the current DTI image to obtain the corresponding white matter fiber streamline set; Step 62: Construct the SC matrix corresponding to the first brain atlas based on the current DTI image; Step 63: Construct the static FC matrix corresponding to the first brain atlas based on the current fMRI image; Step 64: Based on the sliding step size and the sequence segment length, slice the signal duration of the current fMRI image and divide the current fMRI image into T corresponding time-segment images P based on the slicing results. j ; Step 65, based on T images P from the aforementioned time periods j Construct the dynamic FC matrix sequence corresponding to the first brain map; Step 66: Extract the non-zero values ​​in the i-th row of the SC matrix to form the corresponding structural connection strength vector. The non-zero values ​​in the i-th row of the static FC matrix are extracted to form the corresponding functional connectivity strength vector. ; and each of the dynamic matrices FC in the dynamic FC matrix sequence j The non-zero values ​​in the i-th row are extracted to form the corresponding functional connection strength vector. ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding static feature a. i ; and for vectors , The Pearson correlation coefficient is calculated and the result is used as the corresponding dynamic feature b. i,j ; and from the obtained N static features a i The corresponding static SFC feature vector A is formed; and the resulting N×T dynamic features b i,j This forms the corresponding dynamic SFC feature matrix B.

7. An apparatus for performing the method for predicting early Alzheimer's disease based on dynamic and static SFC features according to any one of claims 1-6, characterized in that, The device includes: a brain region customization module, a model customization module, a data preparation module, a model training module, and a model application module; The brain region customization module is used to set the first brain region set; the first brain region set 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), and the total number of brain regions N is a positive integer; The model customization module is used to set up a neural network model for binary classification prediction of early Alzheimer's disease based on dynamic and static SFC features, denoted as the first prediction model; the first prediction model is used to perform binary classification prediction based on the static SFC feature vector A and the dynamic SFC feature matrix B input to the model and output the corresponding prediction vector Y; the static SFC feature vector A consists of N static features a i Sort by, 1 ≤ brain region index i ≤ N; the static feature a i The SFC feature value is the value of the i-th brain region; the dynamic SFC feature matrix B has an N×T shape and consists of N×T dynamic feature b values. i,j Composition, where 1 ≤ time index j ≤ T, and T is the total number of times; each row of the dynamic SFC feature matrix B corresponds to a brain region, and each column corresponds to a time t. j Each of the dynamic features b i,j For the i-th brain region at time t j The corresponding SFC feature value; the prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 correspond to disease type and healthy type, respectively; The data preparation module is used to collect large amounts of data on the dynamic and static SFC characteristics of early Alzheimer's disease patients and healthy people to obtain the corresponding first dataset. The model training module is used to train the first prediction model based on the first dataset; The model application module is used to receive fMRI and DTI images of the brain of any subject input by the user after model training is completed; and to perform dynamic and static SFC feature extraction processing based on the subject images to obtain the corresponding static SFC feature vector A and dynamic SFC feature matrix B; and to input the current static SFC feature vector A and dynamic SFC feature matrix B into the first prediction model to predict and obtain the corresponding prediction vector Y, which is then fed back to the current user.

8. 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 1-6; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

9. 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 according to any one of claims 1-6.

Images