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

Abstract

The embodiment of the present application relates to a kind of method and device based on early Alzheimer's disease prediction of lymphoid system characteristics, the method comprises: gathering original sample set, based on 3D-UNet model, first segmentation model is constructed, based on MLP model, first prediction model is constructed;Based on original sample set, first, second data set is constructed;First segmentation model is trained based on first data set;First prediction model is trained based on second data set;After model training ends, the brain DTI image and brain 3D-T2 weighted image of subject are received, and according to brain DTI image, left and right brain ALPS index analysis is carried out based on DTI-ALPS technique, and based on first segmentation model, semantic segmentation is carried out to brain 3D-T2 weighted image, and according to segmentation result, four brain area perivascular space volume fraction analysis is carried out, and using first prediction model, according to two kinds of analysis results, prediction is carried out.The present application can improve processing efficiency, improve prediction accuracy.

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 lymphoid system characteristics. Background Technology

[0002] The glymphatic system (GS) is a unique fluid transport system in the brain, named for its functional similarity to the peripheral lymphatic system. It removes metabolic waste and harmful substances from the brain through the exchange of cerebrospinal fluid (CSF) and interstitial fluid (ISF), playing a crucial role in maintaining brain homeostasis and neurological health. Common characteristic parameters used to assess GS function include the ALPS index and the perivascular space volume fraction (PVS-VF).

[0003] The ALPS index characterizes the ability of water molecules to diffuse in the perivascular space (PG). It is obtained through diffusion tensor image analysis along the perivascular space (DTI-ALPS). In the early stages of Alzheimer's disease (AD), increased Aβ deposition in the brain leads to impaired PG function and reduced water molecule diffusion in the perivascular space, resulting in a decreased ALPS index. Brain DTI images are essential for obtaining the ALPS index.

[0004] Perivascular space volume fraction (PVS-VF) is the proportion of perivascular space volume to the volume of a specific brain region or the whole brain. In the early stages of Alzheimer's disease (AD), increased Aβ deposition in the brain leads to impaired sigmoid syndrome (GS) function and an increase in perivascular space volume, resulting in an increased PVS-VF. The essential imaging data for obtaining PVS-VF is T2-weighted MRI of the brain, specifically including 2D / 3D T2-weighted images.

[0005] Based on the significant correlation between the ALPS index, PVS-VF score, and early AD pathological features, researchers have developed a series of auxiliary technologies for early AD screening. In simple terms, these conventional technologies involve: 1) analyzing the ALPS index using mature DTI-ALPS technology based on the subject's brain DTI images; 2) estimating the PVS-VF score of a single brain region using manual or semi-manual statistical methods based on the subject's T2-weighted brain images; 3) constructing a multimodal GS feature from the ALPS index and the PVS-VF score of a single brain region, and statistically analyzing the correlation between the multimodal GS feature and early AD features. Finally, based on statistical thresholds, binary classification is performed according to the analysis results to obtain the corresponding early AD screening result.

[0006] This conventional technique has some problems in practical applications: 1) PVS-VF analysis requires human intervention, which leads to long analysis time and low efficiency; 2) Due to human resource limitations, it is difficult to conduct simultaneous PVS-VF analysis on multiple brain regions of interest, resulting in insufficient PVS-VF feature richness and affecting prediction accuracy; 3) Data analysis of multimodal GS features is based solely on statistical thresholds, resulting in poor prediction flexibility and generalization.

[0007] To address the aforementioned issues, we propose an improved solution combining DTI-ALPS and artificial intelligence technologies: ALPS index analysis is performed based on DTI-ALPS technology using brain DTI images of subjects; multi-region PVS-VF score analysis is performed based on 3D-T2 weighted brain images of subjects using a visual semantic segmentation model; multimodal GS features are constructed from ALPS indices and brain region PVS-VF sequences; and early AD prediction is performed based on multimodal GS features using a binary classification neural network model. This solution effectively reduces manual intervention, improves processing efficiency, increases the abundance of multi-region PVS-VF features, improves prediction accuracy, and enhances the flexibility and generalization of prediction through model neural networks. The technical problem this invention aims to solve is how to implement this solution. 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 lymphoid system characteristics. This invention first collects large-scale data from DTI and 3D-T2 weighted brain images of individuals with early Alzheimer's disease and healthy individuals to obtain an original sample set. Then, based on the 3D-UNet model, a first segmentation model is constructed to semantically segment the perivascular spaces of four brain regions (midbrain, hippocampus, basal ganglia, and centrum semiovale) using 3D-T2 weighted brain images. Next, based on the MLP model, a first prediction model is constructed to perform binary classification prediction of early Alzheimer's disease based on lymphoid system characteristics (ALPS indices of the left and right hemispheres + PVS-VF scores of the four brain regions). Then, based on the original sample set, training datasets required for the segmentation and prediction models are constructed to obtain corresponding first and second datasets. The first segmentation model is trained based on the first dataset, and the first prediction model is trained based on the second dataset. After the training of both models, for each set of subjects receiving brain DTI images and 3D-T2 weighted images, the following steps are performed: first, ALPS index analysis of the left and right hemispheres is conducted based on DTI-ALPS technology using the brain DTI images; then, semantic segmentation of the brain 3D-T2 weighted images is performed based on the first segmentation model; and finally, volume fraction analysis of the perivascular spaces in the four brain regions is conducted based on the segmentation results. Then, the first prediction model is used to make predictions based on the results of both analyses. This invention completely eliminates the need for manual intervention in the PVS-VF analysis process, increases multi-brain region PVS-VF features, and allows for classification and prediction based on the nonlinear coupling characteristics of multimodal features. This invention improves analysis and processing efficiency, increases the abundance of multimodal features, enhances prediction accuracy, and improves the flexibility and generalization of predictions.

[0009] To achieve the above objectives, a first aspect of the present invention provides a method for predicting early Alzheimer's disease based on lymphoid system characteristics, the method comprising: We collected large amounts of brain DTI images and 3D-T2 weighted images of the brains of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample sets. A first segmentation model is constructed based on the 3D-UNet model to perform semantic segmentation of perivascular spaces in multiple specified brain regions using the aforementioned 3D-T2 weighted brain images. The specified brain regions include the midbrain, hippocampus, basal ganglia, and centrum semiovale. The first segmentation model is used to perform semantic segmentation of brain regions and perivascular spaces on the input 3D-T2 weighted brain images and output corresponding semantic segmentation maps. The 3D dimensions and 3D resolution of the semantic segmentation maps are consistent with the corresponding 3D-T2 weighted brain images. The voxel semantics of the semantic segmentation maps include brain region semantics and PVS semantics. A first prediction model is constructed based on the MLP model to perform binary classification prediction of early Alzheimer's disease based on lymphoid system characteristics; the first prediction model is used to perform binary classification prediction based on the lymphoid system characteristics X input to the model and output the corresponding prediction vector Y; Based on the original sample set, the training dataset required for the segmentation and prediction model is constructed to obtain the corresponding first dataset and second dataset; The first segmentation model is trained based on the first dataset; The first prediction model is trained based on the second dataset; After the training of the two types of models is completed, the brain DTI images and the brain 3D-T2 weighted images of any subject are received; and based on DTI-ALPS technology, the left and right brain ALPS indexes are analyzed according to the brain DTI images, and semantic segmentation is performed on the brain 3D-T2 weighted images based on the first segmentation model. The volume fraction of the perivascular space in the four brain regions is analyzed according to the segmentation results, and the first prediction model is used to make predictions based on the two types of analysis results.

[0010] Preferably, the big data acquisition channel for the original sample set is a publicly available neuroimaging dataset; the publicly available neuroimaging dataset includes at least the ADNI dataset; The original sample set includes multiple first sample records; each first sample record corresponds one-to-one with an individual subject from the early Alzheimer's disease population or a healthy population; each first sample record includes the subject type, the brain DTI image, and the brain 3D-T2 weighted image; the subject type includes patient type and healthy type; The brain DTI image includes multiple first voxel points; the voxel features of each first voxel point consist of a set of three-dimensional voxel coordinates and anisotropy fraction FA values; The brain 3D-T2 weighted image includes multiple second voxel points; the voxel features of each second voxel point consist of a set of three-dimensional voxel coordinates and signal intensity; The semantic segmentation map includes multiple third voxel points; the voxel features of each third voxel point consist of a set of three-dimensional voxel coordinates, brain region semantic type, and PVS semantic type; the brain region semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the PVS semantic type includes PVS points and non-PVS points. The lymphoid system feature X includes the left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score. The prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 are the corresponding disease type and health type, respectively. The first dataset includes multiple first data records; each first data record includes a first training image, a first-label semantic map, and a second-label semantic map; the first training image is a 3D-T2 weighted image of the brain; the first and second-label semantic maps are each three-dimensional semantic maps, and the three-dimensional dimensions of both semantic maps are consistent with the first training image; the first-label semantic map includes multiple first-label voxel points; the voxel features of each first-label voxel point consist of a set of three-dimensional voxel coordinates and a first semantic type; the first semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the second-label semantic map includes multiple second-label voxel points; the voxel features of each second-label voxel point consist of a set of three-dimensional voxel coordinates and a second semantic type; the second semantic type includes PVS points and non-PVS points; The second dataset includes multiple second data records; the second data record includes a first training feature and a first label vector; the first training feature is a lymphoid system feature X; the first label vector consists of first and second label probabilities; the first and second label probabilities correspond one-to-one with the prediction probabilities y1 and y2; one of the first and second label probabilities is 1 and the other is 0.

[0011] Preferably, the model input terminal of the first segmentation model is used to receive the 3D-T2 weighted brain image, and the model output terminal is used to output the corresponding semantic segmentation map; The first segmentation model includes the 3D-UNet model, a first convolutional layer, a second convolutional layer, a first mask layer, a third convolutional layer, and a first fusion layer; The input terminal of the 3D-UNet model is connected to the model input terminal, and the output terminal is connected to the input terminal of the first convolutional layer and the first input terminal of the first mask layer, respectively. The output terminal of the first convolutional layer is connected to the input terminal of the second convolutional layer. The output terminal of the second convolutional layer is connected to the second input terminal of the first mask layer and the first input terminal of the first fusion layer, respectively. The input terminal of the first mask layer is connected to the input terminal of the third convolutional layer. The output terminal of the third convolutional layer is connected to the second input terminal of the first fusion layer. The output terminal of the first fusion layer is connected to the model output terminal. The 3D-UNet model is used to take the brain 3D-T2 weighted image as the initial feature tensor Z0; and to perform feature extraction processing on the feature tensor Z0 to obtain the corresponding feature tensor Z1, which is then sent to the first convolutional layer and the first mask layer. Wherein, the shape of the feature tensor Z0 is C0×D×H×W; C0 is the corresponding initial feature dimension, C0=1; D, H, and W are the three-dimensional image depth, height, and width of the brain 3D-T2 weighted image, respectively; The shape of the feature tensor Z1 is C1×D×H×W; C1 is the output feature dimension of the 3D-UNet model, and C1 > C0; The feature tensor Z1 consists of D×H×W voxel feature vectors of length C1. Composition, 1≤Depth index d≤D, 1≤Height index h≤H, 1≤Width index w≤W; The first convolutional layer is used to perform feature channel dimensionality reduction on the feature tensor Z1 using C2 pre-set convolutional kernels of shape C1×3×3×3 to obtain the corresponding feature tensor Z2; and uses the ReLU activation function to perform feature activation on the feature tensor Z2 to obtain the corresponding feature tensor Z3, which is then sent to the second convolutional layer. Wherein, the shape of the feature tensor Z2 and the feature tensor Z3 is C2×D×H×W, where C2 is a preset second feature dimension, and C2<C1; The second convolutional layer is used to perform a five-class feature transformation on the feature tensor Z3 using C3 pre-defined convolutional kernels of shape C2×1×1×1 to obtain the corresponding feature tensor Z4, which is then sent to the first mask layer and the first fusion layer. Wherein, the shape of the feature tensor Z4 is C3×D×H×W, where C3 is a preset third feature dimension, C3<C2, C3=5; The feature tensor Z4 is composed of D×H×W voxel feature vectors. composition; The voxel feature vector The data consists of five vectors that correspond one-to-one with the semantic types of the brain regions mentioned above: the midbrain, hippocampus, basal ganglia, centrum semiovale, and others. Composition, 1 ≤ data index i ≤ 5; The first mask layer is used to match the voxel feature vectors of the three-dimensional voxel coordinates in the feature tensor Z1 and the feature tensor Z4. and the voxel feature vector Form a corresponding feature vector pair; and perform a traversal of all the feature vector pairs; during this traversal, take the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest median value The corresponding semantic type of the brain region is taken as the current brain region type; and it is identified whether the current brain region type is other; if so, the corresponding voxel feature vector is set. For the preset mask feature vector The mask feature vector The vector length is C1; otherwise, the corresponding voxel feature vector is set. The voxel feature vector of the current vector pair At the end of this round of traversal, the obtained D×H×W voxel feature vectors are used to... The corresponding feature tensor Z5 is sent to the third convolutional layer; The feature tensor Z5 has a shape of C1×D×H×W, consisting of D×H×W voxel feature vectors of length C1. composition; The voxel feature vector The setup method is as follows: ; The voxel feature vectors in the feature tensor Z5 that are unrelated to the four brain regions of the midbrain, hippocampus, basal ganglia, and centrum semiovale. All are set as the mask feature vectors ; The third convolutional layer is used to process all the voxel feature vectors of the feature tensor Z5. Perform one round of traversal; and during this round of traversal, the feature vector of the voxel being traversed is... As the current voxel feature vector; and whether the current voxel feature vector matches the mask feature vector. Matching is used for identification; if so, the corresponding voxel feature vector is set. For the preset mask feature vector The mask feature vector If the vector length is C4, then C4 pre-defined convolutional kernels of shape C1×1×1×1 are used to perform binary classification feature transformation on the current voxel feature vector to obtain the corresponding voxel feature vector. At the end of this round of traversal, the obtained D×H×W voxel feature vectors are used to... The corresponding feature tensor Z6 is sent to the first fusion layer; The feature tensor Z6 has a shape of C4×D×H×W, where C4 is a preset fourth feature dimension and C4=2; the feature tensor Z6 consists of D×H×W voxel feature vectors of length C4. Composition; the voxel feature vector Two vector data points, each corresponding one-to-one with a PVS point and a non-PVS point of the PVS semantic type. Composition, 1 ≤ data index j ≤ 2; The mask feature vector Of the two preset vector data, the preset vector data corresponding to the PVS point is greater than the preset vector data corresponding to the non-PVS point. The first fusion layer is used to match the voxel feature vectors of the three-dimensional voxel coordinates in the feature tensor Z4 and the feature tensor Z6. and the voxel feature vector Form a corresponding feature vector pair; and perform a traversal of all the feature vector pairs; during this traversal, take the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest median value The corresponding semantic type of the brain region is taken as the current brain region type; and the voxel feature vector of the current vector pair is used as the current brain region type. The vector data with the largest median value The corresponding PVS semantic type is used as the current PVS type; and the current vector pair, the corresponding three-dimensional voxel coordinates, the current brain region type, and the current PVS type are combined to form a voxel feature of the corresponding third voxel point; and all the obtained third voxel points are combined to form the corresponding semantic segmentation map and output.

[0012] Preferably, the model input terminal of the first prediction model is used to receive the lymphoid system-like feature X, and the model output terminal is used to output the corresponding prediction vector Y; The first prediction model is formed by sequentially connecting the MLP model and the Softmax function layer; The MLP model consists of a hidden layer network and an output layer connected sequentially; the hidden layer network consists of multiple hidden layers connected sequentially; the hidden layer consists of a linear layer and an activation layer connected sequentially; the output layer is a linear layer; the MLP model is used to extract features from the lymphoid system feature X to obtain the corresponding two-dimensional feature vector X1, which is then sent to the Softmax function layer. The two-dimensional feature vector X1 has a length of 2 and consists of two vector data. , composition; The Softmax function layer is used to calculate the corresponding prediction probabilities y1 and y2 based on the two-dimensional feature vector X1 using the Softmax function, and then outputs the corresponding prediction vector Y. The prediction probabilities y1 and y2 are calculated as follows: , .

[0013] Preferably, the step of constructing the corresponding first and second datasets based on the training datasets required for the segmentation and prediction models using the original sample set specifically includes: Each of the first sample records in the original sample set is taken as the current sample record; and the object type, the brain DTI image, and the brain 3D-T2 weighted image of the current sample record are taken as the corresponding current object type, current DTI image, and current T2 image. And the current T2 image is used as the corresponding first training image; Based on DTI-ALPS technology, the left and right brain ALPS indices are analyzed according to the current DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index. Based on manual annotation, the first and second label semantic maps corresponding to the current T2 image are labeled with tags. Based on manual statistical methods, the volumes of the four brain regions—midbrain, hippocampus, basal ganglia, and centrum semiovale—are statistically analyzed according to the first label semantic map corresponding to the current T2 image to obtain the corresponding first, second, third, and fourth volumes. Furthermore, based on the second label semantic map corresponding to the current T2 image, the total volume of the perivascular spaces in these four brain regions is statistically analyzed to obtain the corresponding fifth, sixth, seventh, and eighth volumes. Based on these eight statistical volumes, the corresponding PVS-VF scores for the midbrain, hippocampus, basal ganglia, and centrum semiovale are calculated: midbrain PVS-VF score = fifth volume / first volume; hippocampus PVS-VF score = sixth volume / second volume; basal ganglia PVS-VF score = seventh volume / third volume; centrum semiovale PVS-VF score = eighth volume / fourth volume. The first training feature is composed of the left brain ALPS index, the right brain ALPS index, the midbrain PVS-VF score, the hippocampus PVS-VF score, the basal ganglia PVS-VF score, and the centrum semiovale PVS-VF score corresponding to the current sample record. The current object type is identified; if the current object type is a patient, the corresponding first label probability is set to 1 and the second label probability is set to 0; if the current object type is a healthy type, the corresponding first label probability is set to 0 and the second label probability is set to 1; and the first label vector is composed of the first and second label probabilities. The first data record is composed of the first training image, the first label semantic map, and the second label semantic map corresponding to the current sample record; and the second data record is composed of the first training feature and the first label vector corresponding to the current sample record. The first dataset is composed of all the first data records obtained; and the second dataset is composed of all the second data records obtained.

[0014] Preferably, training the first segmentation model based on the first dataset specifically includes: Step 61: Take each of the first data records in the first dataset as the current record; and input the first training image of the current record as the current brain 3D-T2 weighted image into the first segmentation model for processing; and take the feature tensor Z4 and feature tensor Z6 generated during the current processing of the model as the corresponding predicted semantic graph. , And the currently recorded first tag semantic graph and second tag semantic graph are recorded as the corresponding tag semantic graphs. , ; and from the predicted semantic graph and label semantic graph Form the corresponding first prediction-label pair ( , ), from the predicted semantic graph and label semantic graph Form the corresponding second prediction-label pair ( , ); Where 1 ≤ record index u ≤ N1, and N1 is the total number of records in the first dataset; Step 62: Substitute all the obtained first and second prediction-label pairs into the preset first model loss function L. M1 The corresponding first loss value is obtained through calculation; Wherein, the first model loss function L M1 Based on cross-entropy loss function L CE The implementation of the Dice loss function is as follows: ; Step 63: Identify whether the first loss value meets the preset first loss value range; if not, then based on the preset first model optimizer, move towards making the first model loss function L... M1The model parameters of the first segmentation model are modulated once in the direction that reaches the minimum value, and the process returns to step 61 when the modulation ends; if the condition is met, training is stopped and the training of the first segmentation model is confirmed to be complete. The first model optimizer includes the Adam optimizer and the SGD optimizer.

[0015] Preferably, training the first prediction model based on the second dataset specifically includes: Step 71: Take each of the second data records in the second dataset as the current record; and input the first training feature of the current record as the current lymphoid system feature X into the first prediction model for processing; and record the prediction vector Y output by the model this time as the prediction vector. And the first tag vector currently recorded is recorded as the corresponding tag vector. ; and by the prediction vector and the label vector Form the corresponding third prediction-label pair ( , ); Where 1 ≤ record index v ≤ N2, and N2 is the total number of records in the second dataset; Step 72: Substitute all the obtained third prediction-label pairs into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation; Wherein, the second model loss function L M2 Based on cross-entropy loss function L CE Implementation, specifically: ; Step 73: Identify whether the second loss value meets the preset second loss value range; if not, then based on the preset second model optimizer, move towards making the second model loss function L... M2 The model parameters of the first prediction model are modulated once in the direction that reaches the minimum value, and the process returns to step 71 when the modulation ends; if the condition is met, training is stopped and the training of the first prediction model is confirmed to be complete. The second model optimizer includes the Adam optimizer and the SGD optimizer.

[0016] Preferably, the step of performing left and right brain ALPS index analysis based on DTI-ALPS technology using the brain DTI images, performing semantic segmentation on the brain 3D-T2 weighted images based on the first segmentation model, performing perivascular space volume fraction analysis in the four brain regions based on the segmentation results, and using the first prediction model to make predictions based on the two types of analysis results specifically includes: Step 81: Based on DTI-ALPS technology, perform left and right brain ALPS index analysis according to the current brain DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index. Step 82: Input the current brain DTI image into the first segmentation model for processing to obtain the corresponding semantic segmentation map; Step 83: Cluster the third voxel points on the semantic segmentation map whose semantic type is midbrain, hippocampus, basal ganglia, or centrum semiovale into a class to form corresponding first, second, third, or fourth point sets; and assign maximum and minimum depth coordinates d to each point set. max d min Maximum and minimum height coordinates h max h min Maximum and minimum width coordinates w max w min The system performs identification and generates eight vertex coordinates based on the six identified extreme coordinates. The cube space formed by the eight vertex coordinates is used as the first, second, third, or fourth brain region space corresponding to the current point set. Step 84: Cluster the third voxel points on the semantic segmentation map whose PVS semantic type is PVS points and whose three-dimensional voxel coordinates are located in the first, second, third or fourth brain region space into a class to form the corresponding fifth, sixth, seventh or eighth point set; Step 85: Count the total number of voxel points in the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets respectively to obtain the corresponding totals for the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets; and calculate the corresponding PVS-VF scores for the midbrain, hippocampus, basal ganglia, and centrum semiovale based on the obtained totals for the eight point sets. Among them, the midbrain PVS-VF score = the fifth total score / the first total score, the hippocampus PVS-VF score = the sixth total score / the second total score, the basal ganglia PVS-VF score = the seventh total score / the third total score, and the centrum semiovale PVS-VF score = the eighth total score / the fourth total score; Step 86: The lymphoid system feature X, composed of the obtained left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score, is input into the first prediction model for prediction processing to obtain the corresponding prediction vector Y. Step 87: Take the classification type corresponding to the highest probability in the prediction vector Y as the prediction result for the current subject and save it.

[0017] A second aspect of the present invention provides an apparatus for implementing the method for predicting early Alzheimer's disease based on lymphoid system features as described in the first aspect above. The apparatus includes: a data acquisition module, a segmentation model construction module, a prediction model construction module, a dataset preparation module, a segmentation model training module, a prediction model training module, and a prediction task processing module. The data acquisition module is used to collect large amounts of brain DTI images and 3D-T2 weighted images of the brains of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample sets. The segmentation model construction module constructs a first segmentation model based on the 3D-UNet model to perform semantic segmentation of perivascular spaces in multiple specified brain regions using the 3D-T2 weighted brain images. The multiple specified brain regions include the midbrain, hippocampus, basal ganglia, and centrum semiovale. The first segmentation model performs semantic segmentation of brain regions and perivascular spaces on the input 3D-T2 weighted brain images and outputs corresponding semantic segmentation maps. The three-dimensional dimensions and resolution of the semantic segmentation maps are consistent with the corresponding 3D-T2 weighted brain images. The voxel semantics of the semantic segmentation maps include brain region semantics and PVS semantics. The prediction model construction module constructs a first prediction model based on the MLP model to perform binary classification prediction of early Alzheimer's disease according to lymphoid system characteristics; the first prediction model is used to perform binary classification prediction based on the lymphoid system characteristics X input to the model and output the corresponding prediction vector Y; The dataset preparation module constructs the corresponding first dataset and second dataset based on the original sample set to obtain the training dataset required for the segmentation and prediction models. The segmentation model training module trains the first segmentation model based on the first dataset; The prediction model training module trains the first prediction model based on the second dataset; The prediction task processing module is used to receive the brain DTI images and the brain 3D-T2 weighted images of any subject after the training of the two types of models is completed; and to perform left and right brain ALPS index analysis based on the brain DTI images using DTI-ALPS technology, and to perform semantic segmentation on the brain 3D-T2 weighted images based on the first segmentation model and to perform perivascular space volume fraction analysis of the four brain regions based on the segmentation results, and to make predictions using the first prediction model based on the two types of analysis results.

[0018] 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.

[0019] 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.

[0020] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on lymphoid system features. As described above, this invention first collects large-scale data from DTI and 3D-T2 weighted images of the brains of individuals with early Alzheimer's disease and healthy individuals to obtain an original sample set. Then, based on the 3D-UNet model, a first segmentation model is constructed to semantically segment the perivascular spaces of four brain regions (midbrain, hippocampus, basal ganglia, and centrum semiovale) using 3D-T2 weighted images. Next, based on the MLP model, a first prediction model is constructed to perform binary classification prediction of early Alzheimer's disease based on lymphoid system features (ALPS indices of the left and right hemispheres + PVS-VF scores of the four brain regions). Then, based on the original sample set, training datasets required for the segmentation and prediction models are constructed to obtain corresponding first and second datasets. The first segmentation model is trained based on the first dataset, and the first prediction model is trained based on the second dataset. After the training of both models is completed, for each set of subjects receiving brain DTI images and 3D-T2 weighted images, the following steps are performed: first, ALPS index analysis of the left and right hemispheres is conducted based on DTI-ALPS technology using the brain DTI images; then, semantic segmentation of the brain 3D-T2 weighted images is performed based on the first segmentation model; and finally, volume fraction analysis of the perivascular space in the four brain regions is conducted based on the segmentation results. Then, the first prediction model is used to make predictions based on the results of both analyses. This invention completely eliminates the need for manual intervention in the PVS-VF analysis process, increases multi-brain region PVS-VF features, and allows for classification and prediction based on the nonlinear coupling characteristics of multimodal features. This invention improves analysis and processing efficiency, increases the abundance of multimodal features, improves prediction accuracy, and enhances the flexibility and generalization of prediction. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a method for predicting early Alzheimer's disease based on lymphoid system characteristics provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the modules of the first segmentation model provided in Embodiment 1 of the present invention; Figure 3 A schematic diagram of the modules of the first prediction model provided in Embodiment 1 of the present invention; Figure 4 This is a module structure diagram of a device for predicting early Alzheimer's disease based on lymphoid system characteristics, provided in Embodiment 2 of the present invention. Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0022] 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.

[0023] Embodiment 1 of this invention provides a method for predicting early Alzheimer's disease based on lymphoid system characteristics, such as... Figure 1 The schematic diagram of a method for predicting early Alzheimer's disease based on lymphoid system characteristics provided in Embodiment 1 of the present invention is shown. The method mainly includes the following steps: Step 1: Collect large-scale data from brain DTI images and 3D-T2 weighted images of individuals with early-stage Alzheimer's disease and healthy individuals to obtain the corresponding raw sample set.

[0024] Here, the big data collection channel in this embodiment of the invention is a publicly available neuroimaging dataset; the publicly available neuroimaging dataset includes at least the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.

[0025] The original sample set of this invention includes multiple first sample records; each first sample record corresponds one-to-one with an individual subject in the early Alzheimer's disease population or a healthy population; each first sample record includes the subject type, brain DTI images, and brain 3D-T2 weighted images; the subject type includes patient type and healthy type.

[0026] The brain DTI image of this embodiment includes multiple first voxel points; the voxel features of each first voxel point consist of a set of three-dimensional voxel coordinates and anisotropy fraction (FA) values. The brain 3D-T2 weighted image of this embodiment includes multiple second voxel points; the voxel features of each second voxel point consist of a set of three-dimensional voxel coordinates and signal intensity.

[0027] Step 2: Construct a first segmentation model based on the 3D-UNet model to perform semantic segmentation of the perivascular spaces of multiple specified brain regions based on 3D-T2 weighted images of the brain.

[0028] Here, the multiple designated brain regions in this embodiment of the invention are specifically four brain regions, including the midbrain, hippocampus, basal ganglia, and centrum semiovale.

[0029] The first segmentation model in this embodiment of the invention is used to perform semantic segmentation of brain regions and perivascular spaces on the brain 3D-T2 weighted images of the brain input to the model and output the corresponding semantic segmentation map.

[0030] The three-dimensional size and resolution of the semantic segmentation map in this embodiment of the invention are consistent with the corresponding 3D-T2 weighted brain image. The voxel semantics of the semantic segmentation map include brain region semantics and PVS semantics. Specifically, the semantic segmentation map includes multiple third voxel points; the voxel features of each third voxel point are composed of a set of three-dimensional voxel coordinates, brain region semantic type, and PVS semantic type; the brain region semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the PVS semantic type includes PVS points and non-PVS points.

[0031] like Figure 2 As shown in the schematic diagram of the first segmentation model provided in Embodiment 1 of the present invention, the model input end of the first segmentation model of the present invention is used to receive 3D-T2 weighted images of the brain, and the model output end is used to output the corresponding semantic segmentation map.

[0032] like Figure 2 As shown, the model components of the first segmentation model include: a 3D-UNet model, a first convolutional layer, a second convolutional layer, a first mask layer, a third convolutional layer, and a first fusion layer.

[0033] like Figure 2 As shown, the connection relationships of the model components in the first segmentation model are as follows: the input end of the 3D-UNet model is connected to the model input end, and the output end is connected to the input end of the first convolutional layer and the first input end of the first mask layer, respectively; the output end of the first convolutional layer is connected to the input end of the second convolutional layer; the output end of the second convolutional layer is connected to the second input end of the first mask layer and the first input end of the first fusion layer, respectively; the input end of the first mask layer is connected to the input end of the third convolutional layer; the output end of the third convolutional layer is connected to the second input end of the first fusion layer; and the output end of the first fusion layer is connected to the model output end.

[0034] The functions of the model components in the first segmentation model are shown below.

[0035] 1) 3D-UNet model: The 3D-UNet model in this embodiment of the invention is used to take the brain 3D-T2 weighted image as the initial feature tensor Z0; and to perform feature extraction processing on the feature tensor Z0 to obtain the corresponding feature tensor Z1, which is then sent to the first convolutional layer and the first mask layer.

[0036] Here, the shape of the feature tensor Z0 in this embodiment of the invention is C0×D×H×W; C0 is the corresponding initial feature dimension, C0=1; D, H, and W are the three-dimensional image depth, height, and width of the brain 3D-T2 weighted image, respectively.

[0037] In this embodiment of the invention, the feature tensor Z1 has a shape of C1×D×H×W; C1 is the output feature dimension of the 3D-UNet model, and C1 > C0. The feature tensor Z1 consists of D×H×W voxel feature vectors of length C1. Composition: 1 ≤ depth index d ≤ D, 1 ≤ height index h ≤ H, 1 ≤ width index w ≤ W.

[0038] 2) First convolutional layer: In this embodiment of the invention, the first convolutional layer is used to perform feature channel dimensionality reduction on the feature tensor Z1 using C2 pre-set convolutional kernels of shape C1×3×3×3 to obtain the corresponding feature tensor Z2; and to perform feature activation on the feature tensor Z2 using the ReLU activation function to obtain the corresponding feature tensor Z3, which is then sent to the second convolutional layer.

[0039] Here, the feature tensor Z2 and feature tensor Z3 in this embodiment of the invention both have the shape C2×D×H×W, where C2 is a preset second feature dimension and C2 < C1.

[0040] 3) Second convolutional layer: In this embodiment of the invention, the second convolutional layer is used to perform five-class feature transformation on the feature tensor Z3 using C3 pre-set convolutional kernels of shape C2×1×1×1 to obtain the corresponding feature tensor Z4, which is then sent to the first mask layer and the first fusion layer.

[0041] Here, the feature tensor Z4 in this embodiment of the invention has a shape of C3×D×H×W, where C3 is a preset third feature dimension, C3 < C2, and C3 = 5. The feature tensor Z4 consists of D×H×W voxel feature vectors. Composition; Voxel feature vector The data consists of five vectors that correspond one-to-one with the semantic types of brain regions: the midbrain, hippocampus, basal ganglia, centrum semiovale, and others. Composition, 1 ≤ data index i ≤ 5.

[0042] 4) First mask layer: The first mask layer in this embodiment of the invention is used to match the voxel feature vectors of the three-dimensional voxel coordinates in feature tensors Z1 and Z4. and voxel feature vector Form a corresponding feature vector pair; perform a traversal of all feature vector pairs; during this traversal, use the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest value The corresponding semantic type of the brain region is used as the current brain region type; and it is determined whether the current brain region type is other; if so, the corresponding voxel feature vector is set. For the preset mask feature vector Mask feature vector The vector length is C1; otherwise, the corresponding voxel feature vector is set. The voxel feature vector of the current vector pair At the end of this round of traversal, the obtained D×H×W individual feature vectors are used to... The corresponding feature tensor Z5 is sent to the third convolutional layer.

[0043] Here, the feature tensor Z5 of this embodiment has a shape of C1×D×H×W, consisting of D×H×W voxel feature vectors of length C1. composition; voxel feature vector The setup method is as follows: .

[0044] It is important to emphasize that the voxel feature vectors in the feature tensor Z5 are unrelated to the four brain regions of the midbrain, hippocampus, basal ganglia, and centrum semiovale. All are set as mask feature vectors .

[0045] 5) Third convolutional layer: The third convolutional layer in this embodiment of the invention is used to process all voxel feature vectors of the feature tensor Z5. Perform one round of traversal; and during this round of traversal, change the feature vector of the voxel being traversed. As the current voxel feature vector; and check whether the current voxel feature vector matches the mask feature vector. Matching is used for identification; if so, the corresponding voxel feature vector is set. For the preset mask feature vector Mask feature vector If the vector length is C4, then C4 pre-defined convolutional kernels of shape C1×1×1×1 are used to perform binary classification feature transformation on the current voxel feature vector to obtain the corresponding voxel feature vector. At the end of this round of traversal, the obtained D×H×W individual feature vectors are used to... The corresponding feature tensor Z6 is sent to the first fusion layer.

[0046] Here, the feature tensor Z6 in this embodiment of the invention has a shape of C4×D×H×W, where C4 is a preset fourth feature dimension and C4=2; the feature tensor Z6 consists of D×H×W voxel feature vectors of length C4. Composition; Voxel feature vector It consists of two vector data that correspond one-to-one with PVS points and non-PVS points of the PVS semantic type. Composition, 1 ≤ data index j ≤ 2.

[0047] Mask feature vector of the present invention Of the two preset vector data, the preset vector data corresponding to the PVS point is greater than the preset vector data corresponding to the non-PVS point.

[0048] 6) First Fusion Layer: The first fusion layer in this embodiment of the invention is used to match the voxel feature vectors of the three-dimensional voxel coordinates in feature tensor Z4 and feature tensor Z6. and voxel feature vector Form a corresponding feature vector pair; perform a traversal of all feature vector pairs; during this traversal, use the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest value The corresponding brain region semantic type is used as the current brain region type; and the voxel feature vector of the current vector pair is used as the current vector region type. The vector data with the largest value The corresponding PVS semantic type is used as the current PVS type; and the current vector pair is composed of the corresponding three-dimensional voxel coordinates, the current brain region type, and the current PVS type to form a voxel feature of a corresponding third voxel point; and all the obtained third voxel points are used to form the corresponding semantic segmentation map and output it.

[0049] Step 3: Construct a first prediction model based on the MLP model to perform binary classification prediction of early Alzheimer's disease based on lymphoid system characteristics.

[0050] Here, the first prediction model in this embodiment of the invention is used to perform binary classification prediction based on the lymphoid system-like features X input to the model and output the corresponding prediction vector Y. The lymphoid system-like features X include the left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score. The prediction vector Y consists of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 are the corresponding disease type and healthy type, respectively.

[0051] like Figure 3As shown in the schematic diagram of the first prediction model provided in Embodiment 1 of the present invention, the model input end of the first prediction model is used to receive lymphatic system-like features X, and the model output end is used to output the corresponding prediction vector Y.

[0052] like Figure 3 As shown, the first prediction model is composed of an MLP model and a Softmax function layer connected sequentially.

[0053] The functions of each model component in the first prediction model are shown below.

[0054] 1) MLP model The MLP model in this embodiment of the invention consists of a hidden layer network and an output layer connected sequentially; the hidden layer network consists of multiple hidden layers connected sequentially; the hidden layer consists of a linear layer and an activation layer connected sequentially; and the output layer is a linear layer.

[0055] The MLP model in this embodiment of the invention is used to extract features from the lymphoid system feature X to obtain the corresponding two-dimensional feature vector X1, which is then sent to the Softmax function layer.

[0056] Here, in this embodiment of the invention, the length of the two-dimensional feature vector X1 is 2, consisting of two vector data. , composition.

[0057] 2) Softmax function layer: The Softmax function layer in this embodiment of the invention is used to calculate the binary classification probability based on the two-dimensional feature vector X1 using the Softmax function, obtain the corresponding prediction probabilities y1 and y2, form the corresponding prediction vector Y, and output it.

[0058] Here, the prediction probabilities y1 and y2 in this embodiment of the invention are calculated as follows: , .

[0059] Step 4: Based on the original sample set, construct the training dataset required for the segmentation and prediction models to obtain the corresponding first dataset and second dataset; Specifically, this includes: Step 41, taking each first sample record of the original sample set as the current sample record; and taking the object type, brain DTI image, and brain 3D-T2 weighted image of the current sample record as the corresponding current object type, current DTI image, and current T2 image; Step 42, and use the current T2 image as the corresponding first training image; Step 43, and based on DTI-ALPS technology, perform left and right brain ALPS index analysis according to the current DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index. Specifically, this includes: Step 431, setting a pair of symmetrical projection fiber regions of interest and a pair of symmetrical association fiber regions of interest for the left and right hemispheres at the level of the left and right ventricles in the current DTI image, respectively. Step 432: Based on the principle of DTI-ALPS technology, the diffusion rate of the region of interest of the projection fibers in the left brain along the x, y, and z axes is calculated to obtain the corresponding diffusion rates of the left brain projection fibers along the x, y, and z axes. , , The diffusion rates of the regions of interest of the right brain projection fibers along the x, y, and z axes were calculated to obtain the corresponding diffusion rates of the right brain projection fibers along the x, y, and z axes. , , ; Step 433: Based on the principle of DTI-ALPS technology, the diffusion rate of the region of interest of the left brain association fibers along the x, y, and z axes is calculated to obtain the corresponding diffusion rates of the left brain association fibers along the x, y, and z axes. , , The diffusion rates of the regions of interest of the right brain's associative fibers along the x, y, and z axes were calculated to obtain the corresponding diffusion rates of the right brain's associative fibers along the x, y, and z axes, respectively. , , ; Step 434, calculate based on the diffusion rate according to the DTI-ALPS index principle. , , , Calculate the corresponding left brain ALPS index, based on the diffusion rate. , , , Calculate the corresponding right brain ALPS index; , , Where mean() is the function for calculating the average value; Step 44, and based on manual annotation, label the first and second label semantic maps corresponding to the current T2 image; Step 45: Based on manual statistical methods, the volumes of the four brain regions—midbrain, hippocampus, basal ganglia, and centrum semiovale—are statistically analyzed according to the first label semantic map corresponding to the current T2 image to obtain the corresponding first, second, third, and fourth volumes. Then, based on the second label semantic map corresponding to the current T2 image, the total volume of the perivascular spaces in these four brain regions is statistically analyzed to obtain the corresponding fifth, sixth, seventh, and eighth volumes. Finally, based on the eight obtained statistical volumes, the corresponding PVS-VF scores for the midbrain, hippocampus, basal ganglia, and centrum semiovale are calculated. Here, the midbrain PVS-VF score = fifth volume / first volume, the hippocampus PVS-VF score = sixth volume / second volume, the basal ganglia PVS-VF score = seventh volume / third volume, and the centrum semiovale PVS-VF score = eighth volume / fourth volume; Step 46, and the corresponding first training feature is composed of the left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score of the current sample record. Step 47, and identify the current object type; if the current object type is a patient type, set the corresponding first label probability to 1 and the second label probability to 0; if the current object type is a healthy type, set the corresponding first label probability to 0 and the second label probability to 1; and form the corresponding first label vector from the first and second label probabilities; Step 48, and a corresponding first data record is formed by the first training image, the first label semantic map and the second label semantic map corresponding to the current sample record; and a corresponding second data record is formed by the first training feature and the first label vector corresponding to the current sample record; Step 49, and the first dataset is composed of all the first data records obtained; and the second dataset is composed of all the second data records obtained.

[0060] Here, the first dataset obtained in this embodiment of the invention includes multiple first data records; the first data records include a first training image, a first-label semantic map, and a second-label semantic map; the first training image is a 3D-T2 weighted image of the brain; the first and second-label semantic maps are each a three-dimensional semantic map, and the three-dimensional dimensions of the two semantic maps are consistent with the first training image; the first-label semantic map includes multiple first-label voxel points; the voxel features of each first-label voxel point are composed of a set of three-dimensional voxel coordinates and a first semantic type; the first semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the second-label semantic map includes multiple second-label voxel points; the voxel features of each second-label voxel point are composed of a set of three-dimensional voxel coordinates and a second semantic type; the second semantic type includes PVS points and non-PVS points.

[0061] The second dataset obtained in this embodiment of the invention includes multiple second data records; the second data record includes a first training feature and a first label vector; the first training feature is a lymphatic system-like feature X; the first label vector is composed of first and second label probabilities; the first and second label probabilities correspond one-to-one with the prediction probabilities y1 and y2; one of the first and second label probabilities is 1 and the other is 0.

[0062] Step 5: Train the first segmentation model based on the first dataset; Specifically, this includes: Step 51, taking each first data record of the first dataset as the current record; inputting the first training image of the current record as the current brain 3D-T2 weighted image into the first segmentation model for processing; and using the feature tensor Z4 and feature tensor Z6 generated during this processing as the corresponding predicted semantic graph. , And record the first and second tag semantic graphs currently being recorded as the corresponding tag semantic graphs. , ; and from the predicted semantic graph and tag semantic graph Form the corresponding first prediction-label pair ( , ), from the predicted semantic graph and tag semantic graph Form the corresponding second prediction-label pair ( , ); Where 1 ≤ record index u ≤ N1, and N1 is the total number of records in the first dataset; Step 52: Substitute all the obtained first and second prediction-label pairs into the preset first model loss function L. M1 The corresponding first loss value is obtained through calculation; Here, the first model loss function L in this embodiment of the invention M1 Based on cross-entropy loss function L CE The implementation of the Dice loss function is as follows: ; Step 53: Identify whether the first loss value meets the preset first loss value range; if not, then based on the preset first model optimizer, move towards making the first model loss function L... M1 The direction that reaches the minimum value modulates the model parameters of the first segmentation model in one round, and returns to step 51 when the modulation ends; if satisfied, training stops and the training of the first segmentation model is confirmed to be completed. 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.

[0063] Step 6: Train the first prediction model based on the second dataset; Specifically, this includes: Step 61, taking each second data record of the second dataset as the current record; inputting the first training feature of the current record as the current lymphoid system feature X into the first prediction model for processing; and recording the prediction vector Y output by the model this time as the prediction vector. And record the first label vector of the current record as the corresponding label vector. ; and by the prediction vector and label vector Form the corresponding third prediction-label pair ( , ); Where 1 ≤ record index v ≤ N2, and N2 is the total number of records in the second dataset; Step 62: Substitute all the obtained third prediction-label pairs into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation; Here, the second model loss function L in this embodiment of the invention M2 Based on cross-entropy loss function L CE Implementation, specifically: ; Step 63: Identify whether the second loss value meets the preset range of the second loss value; if not, then based on the preset second model optimizer, move towards making the second model loss function L... M2 The direction that reaches the minimum value modulates the model parameters of the first prediction model in one round, and returns to step 61 when the modulation ends; if satisfied, training stops and the training of the first prediction model is confirmed to be completed. Here, the second loss value range in this embodiment of the invention is a pre-set numerical range; the second model optimizer includes the Adam optimizer and the SGD optimizer.

[0064] Step 7: After the training of the two types of models is completed, receive the brain DTI images and brain 3D-T2 weighted images of any subject; and perform left and right brain ALPS index analysis based on the brain DTI images using DTI-ALPS technology, and perform semantic segmentation on the brain 3D-T2 weighted images based on the first segmentation model, and perform volume fraction analysis of the perivascular space in the four brain regions based on the segmentation results, and use the first prediction model to make predictions based on the two types of analysis results. Specifically, this includes: Step 71, after the training of the two types of models is completed, receiving brain DTI images and brain 3D-T2 weighted images of any subject; Step 72, and based on DTI-ALPS technology, perform left and right brain ALPS index analysis based on brain DTI images, and based on the first segmentation model, perform semantic segmentation on brain 3D-T2 weighted images and perform volume fraction analysis of perivascular space in four brain regions based on the segmentation results, and use the first prediction model to make predictions based on the two types of analysis results. Specifically, it includes: Step 721, based on DTI-ALPS technology, performing left and right brain ALPS index analysis based on the current brain DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index; Here, the processing flow of the current step 721 is similar to that of step 43 above, and will not be repeated here; Step 722: Input the current brain DTI image into the first segmentation model for processing to obtain the corresponding semantic segmentation map; Step 723: Cluster the third voxel points on the semantic segmentation map whose semantic region semantic type is midbrain, hippocampus, basal ganglia, or centrum semiovale into a class to form the corresponding first, second, third, or fourth point set; and assign the maximum and minimum depth coordinates d of each point set. max d min Maximum and minimum height coordinates h max h min Maximum and minimum width coordinates w max w min The system performs identification and generates eight vertex coordinates based on the six identified extreme coordinates. The cube space formed by the eight vertex coordinates is used as the first, second, third, or fourth brain region space corresponding to the current point set. Step 724: Cluster the third voxel points on the semantic segmentation map that have PVS semantic type and whose three-dimensional voxel coordinates are located in the first, second, third or fourth brain region space into a class to form the corresponding fifth, sixth, seventh or eighth point set. Step 725: Count the total number of voxels in the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets respectively to obtain the corresponding totals for the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets; and calculate the corresponding midbrain PVS-VF scores, hippocampal PVS-VF scores, basal ganglia PVS-VF scores, and centrum semiovale PVS-VF scores based on the obtained totals for the eight point sets. Among them, the midbrain PVS-VF score = the fifth total score / the first total score, the hippocampus PVS-VF score = the sixth total score / the second total score, the basal ganglia PVS-VF score = the seventh total score / the third total score, and the centrum semiovale PVS-VF score = the eighth total score / the fourth total score; Step 726: The corresponding lymphoid system feature X, composed of the left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score, is input into the first prediction model for prediction processing to obtain the corresponding prediction vector Y. Step 727: Take the classification type corresponding to the highest probability in the prediction vector Y as the prediction result for the current subject and save it.

[0065] Figure 4 This is a module structure diagram of a device for predicting early Alzheimer's disease based on lymphoid system characteristics, 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 4 As shown, the device includes: a data acquisition module 201, a segmentation model construction module 202, a prediction model construction module 203, a dataset preparation module 204, a segmentation model training module 205, a prediction model training module 206, and a prediction task processing module 207.

[0066] The data acquisition module 201 is used to collect large data from brain DTI images and brain 3D-T2 weighted images of people with early Alzheimer's disease and healthy people to obtain the corresponding raw sample set.

[0067] The segmentation model construction module 202 constructs a first segmentation model based on the 3D-UNet model to perform semantic segmentation of the perivascular spaces of multiple specified brain regions based on 3D-T2 weighted images of the brain. The multiple specified brain regions include the midbrain, hippocampus, basal ganglia, and centrum semiovale. The first segmentation model is used to perform semantic segmentation of brain regions and perivascular spaces on the 3D-T2 weighted images of the brain input to the model and output the corresponding semantic segmentation map. The three-dimensional size and three-dimensional resolution of the semantic segmentation map are consistent with the corresponding 3D-T2 weighted images of the brain. The voxel semantics of the semantic segmentation map include brain region semantics and PVS semantics.

[0068] The prediction model building module 203 constructs a first prediction model based on the MLP model to perform binary classification prediction of early Alzheimer's disease according to the lymphoid system characteristics; the first prediction model is used to perform binary classification prediction based on the lymphoid system characteristics X input to the model and output the corresponding prediction vector Y.

[0069] The dataset preparation module 204 constructs the training dataset required for the segmentation and prediction models based on the original sample set to obtain the corresponding first dataset and second dataset.

[0070] The segmentation model training module 205 trains the first segmentation model based on the first dataset.

[0071] The prediction model training module 206 trains the first prediction model based on the second dataset.

[0072] The prediction task processing module 207 is used to receive brain DTI images and brain 3D-T2 weighted images of any subject after the training of the two types of models; and to perform left and right brain ALPS index analysis based on DTI-ALPS technology according to the brain DTI images, and to perform semantic segmentation of brain 3D-T2 weighted images based on the first segmentation model and to perform volume fraction analysis of perivascular space in the four brain regions according to the segmentation results, and to make predictions based on the two types of analysis results using the first prediction model.

[0073] The present invention provides a device for predicting early Alzheimer's disease based on lymphoid system characteristics. 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.

[0074] 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 data acquisition 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.

[0075] 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).

[0076] 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)).

[0077] Figure 5 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 5 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.

[0078] exist Figure 5The 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.

[0079] 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.

[0080] 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.

[0081] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for predicting early Alzheimer's disease based on lymphoid system features. As described above, this invention first collects large-scale data from DTI and 3D-T2 weighted images of the brains of individuals with early Alzheimer's disease and healthy individuals to obtain an original sample set. Then, based on the 3D-UNet model, a first segmentation model is constructed to semantically segment the perivascular spaces of four brain regions (midbrain, hippocampus, basal ganglia, and centrum semiovale) using 3D-T2 weighted images. Next, based on the MLP model, a first prediction model is constructed to perform binary classification prediction of early Alzheimer's disease based on lymphoid system features (ALPS indices of the left and right hemispheres + PVS-VF scores of the four brain regions). Then, based on the original sample set, training datasets required for the segmentation and prediction models are constructed to obtain corresponding first and second datasets. The first segmentation model is trained based on the first dataset, and the first prediction model is trained based on the second dataset. After the training of both models is completed, for each set of subjects receiving brain DTI images and 3D-T2 weighted images, the following steps are performed: first, ALPS index analysis of the left and right hemispheres is conducted based on DTI-ALPS technology using the brain DTI images; then, semantic segmentation of the brain 3D-T2 weighted images is performed based on the first segmentation model; and finally, volume fraction analysis of the perivascular space in the four brain regions is conducted based on the segmentation results. Then, the first prediction model is used to make predictions based on the results of both analyses. This invention completely eliminates the need for manual intervention in the PVS-VF analysis process, increases multi-brain region PVS-VF features, and allows for classification and prediction based on the nonlinear coupling characteristics of multimodal features. This invention improves analysis and processing efficiency, increases the abundance of multimodal features, improves prediction accuracy, and enhances the flexibility and generalization of prediction.

[0082] 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.

[0083] 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 lymphoid system characteristics, characterized in that, The method includes: We collected large amounts of brain DTI images and 3D-T2 weighted images of the brains of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample sets. A first segmentation model is constructed based on the 3D-UNet model to perform semantic segmentation of perivascular spaces in multiple specified brain regions using the aforementioned 3D-T2 weighted brain images. The specified brain regions include the midbrain, hippocampus, basal ganglia, and centrum semiovale. The first segmentation model is used to perform semantic segmentation of brain regions and perivascular spaces on the input 3D-T2 weighted brain images and output corresponding semantic segmentation maps. The 3D dimensions and 3D resolution of the semantic segmentation maps are consistent with the corresponding 3D-T2 weighted brain images. The voxel semantics of the semantic segmentation maps include brain region semantics and PVS semantics. A first prediction model is constructed based on the MLP model to perform binary classification prediction of early Alzheimer's disease based on lymphoid system characteristics; the first prediction model is used to perform binary classification prediction based on the lymphoid system characteristics X input to the model and output the corresponding prediction vector Y; Based on the original sample set, the training dataset required for the segmentation and prediction model is constructed to obtain the corresponding first dataset and second dataset; The first segmentation model is trained based on the first dataset; The first prediction model is trained based on the second dataset; After the training of the two types of models is completed, the brain DTI images and the brain 3D-T2 weighted images of any subject are received; and based on DTI-ALPS technology, the left and right brain ALPS indexes are analyzed according to the brain DTI images, and semantic segmentation is performed on the brain 3D-T2 weighted images based on the first segmentation model. The volume fraction of the perivascular space in the four brain regions is analyzed according to the segmentation results, and the first prediction model is used to make predictions based on the two types of analysis results.

2. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 1, characterized in that, The original sample set was collected from publicly available neuroimaging datasets; these publicly available neuroimaging datasets include at least the ADNI dataset. The original sample set includes multiple first sample records; each first sample record corresponds one-to-one with an individual subject from the early Alzheimer's disease population or a healthy population; each first sample record includes the subject type, the brain DTI image, and the brain 3D-T2 weighted image; the subject type includes patient type and healthy type; The brain DTI image includes multiple first voxel points; the voxel features of each first voxel point consist of a set of three-dimensional voxel coordinates and anisotropy fraction FA values; The brain 3D-T2 weighted image includes multiple second voxel points; the voxel features of each second voxel point consist of a set of three-dimensional voxel coordinates and signal intensity; The semantic segmentation map includes multiple third voxel points; the voxel features of each third voxel point consist of a set of three-dimensional voxel coordinates, brain region semantic type, and PVS semantic type; the brain region semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the PVS semantic type includes PVS points and non-PVS points. The lymphoid system feature X includes the left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score. The prediction vector Y is composed of prediction probabilities y1 and y2; the classification types of prediction probabilities y1 and y2 are the corresponding patient type and healthy type, respectively. The first dataset includes multiple first data records; each first data record includes a first training image, a first-label semantic map, and a second-label semantic map; the first training image is a 3D-T2 weighted image of the brain; the first and second-label semantic maps are each three-dimensional semantic maps, and the three-dimensional dimensions of both semantic maps are consistent with the first training image; the first-label semantic map includes multiple first-label voxel points; the voxel features of each first-label voxel point consist of a set of three-dimensional voxel coordinates and a first semantic type; the first semantic type includes midbrain, hippocampus, basal ganglia, centrum semiovale, and others; the second-label semantic map includes multiple second-label voxel points; the voxel features of each second-label voxel point consist of a set of three-dimensional voxel coordinates and a second semantic type; the second semantic type includes PVS points and non-PVS points; The second dataset includes multiple second data records; the second data record includes a first training feature and a first label vector; the first training feature is a lymphoid system feature X; the first label vector consists of first and second label probabilities; the first and second label probabilities correspond one-to-one with the prediction probabilities y1 and y2; one of the first and second label probabilities is 1 and the other is 0.

3. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 2, characterized in that, The input end of the first segmentation model is used to receive the 3D-T2 weighted brain image, and the output end is used to output the corresponding semantic segmentation map. The first segmentation model includes the 3D-UNet model, a first convolutional layer, a second convolutional layer, a first mask layer, a third convolutional layer, and a first fusion layer; The input terminal of the 3D-UNet model is connected to the model input terminal, and the output terminal is connected to the input terminal of the first convolutional layer and the first input terminal of the first mask layer, respectively. The output terminal of the first convolutional layer is connected to the input terminal of the second convolutional layer. The output terminal of the second convolutional layer is connected to the second input terminal of the first mask layer and the first input terminal of the first fusion layer, respectively. The output terminal of the first mask layer is connected to the input terminal of the third convolutional layer. The output terminal of the third convolutional layer is connected to the second input terminal of the first fusion layer. The output terminal of the first fusion layer is connected to the model output terminal. The 3D-UNet model is used to take the brain 3D-T2 weighted image as the initial feature tensor Z0; and to perform feature extraction processing on the feature tensor Z0 to obtain the corresponding feature tensor Z1, which is then sent to the first convolutional layer and the first mask layer. Wherein, the shape of the feature tensor Z0 is C0×D×H×W; C0 is the corresponding initial feature dimension, C0=1; D, H, and W are the three-dimensional image depth, height, and width of the brain 3D-T2 weighted image, respectively; The shape of the feature tensor Z1 is C1×D×H×W; C1 is the output feature dimension of the 3D-UNet model, and C1 > C0; The feature tensor Z1 consists of D×H×W voxel feature vectors of length C1. Composition, 1≤Depth index d≤D, 1≤Height index h≤H, 1≤Width index w≤W; The first convolutional layer is used to perform feature channel dimensionality reduction on the feature tensor Z1 using C2 pre-set convolutional kernels of shape C1×3×3×3 to obtain the corresponding feature tensor Z2; and uses the ReLU activation function to perform feature activation on the feature tensor Z2 to obtain the corresponding feature tensor Z3, which is then sent to the second convolutional layer. Wherein, the shape of the feature tensor Z2 and the feature tensor Z3 is C2×D×H×W, where C2 is a preset second feature dimension, and C2<C1; The second convolutional layer is used to perform a five-class feature transformation on the feature tensor Z3 using C3 pre-defined convolutional kernels of shape C2×1×1×1 to obtain the corresponding feature tensor Z4, which is then sent to the first mask layer and the first fusion layer. Wherein, the shape of the feature tensor Z4 is C3×D×H×W, where C3 is a preset third feature dimension, C3<C2, C3=5; The feature tensor Z4 is composed of D×H×W voxel feature vectors. composition; The voxel feature vector The data consists of five vectors that correspond one-to-one with the semantic types of the brain regions mentioned above: the midbrain, hippocampus, basal ganglia, centrum semiovale, and others. Composition, 1 ≤ data index i ≤ 5; The first mask layer is used to match the voxel feature vectors of the three-dimensional voxel coordinates in the feature tensor Z1 and the feature tensor Z4. and the voxel feature vector Form a corresponding feature vector pair; and perform a traversal of all the feature vector pairs; during this traversal, take the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest median value The corresponding semantic type of the brain region is taken as the current brain region type; and it is identified whether the current brain region type is other; if so, the corresponding voxel feature vector is set. For the preset mask feature vector The mask feature vector The vector length is C1; otherwise, the corresponding voxel feature vector is set. The voxel feature vector of the current vector pair At the end of this round of traversal, the obtained D×H×W voxel feature vectors are used to... The corresponding feature tensor Z5 is sent to the third convolutional layer; The feature tensor Z5 has a shape of C1×D×H×W, consisting of D×H×W voxel feature vectors of length C1. composition; The voxel feature vector The setup method is as follows: ； The voxel feature vectors in the feature tensor Z5 that are unrelated to the four brain regions of the midbrain, hippocampus, basal ganglia, and centrum semiovale. All are set as the mask feature vectors ; The third convolutional layer is used to process all the voxel feature vectors of the feature tensor Z5. Perform one round of traversal; and during this round of traversal, the feature vector of the voxel being traversed is... As the current voxel feature vector; and whether the current voxel feature vector matches the mask feature vector. Matching is used for identification; if so, the corresponding voxel feature vector is set. For the preset mask feature vector The mask feature vector If the vector length is C4, then C4 pre-defined convolutional kernels of shape C1×1×1×1 are used to perform binary classification feature transformation on the current voxel feature vector to obtain the corresponding voxel feature vector. At the end of this round of traversal, the obtained D×H×W voxel feature vectors are used to... The corresponding feature tensor Z6 is sent to the first fusion layer; The feature tensor Z6 has a shape of C4×D×H×W, where C4 is a preset fourth feature dimension and C4=2; the feature tensor Z6 consists of D×H×W voxel feature vectors of length C4. Composition; the voxel feature vector Two vector data points, each corresponding one-to-one with a PVS point and a non-PVS point of the PVS semantic type. Composition, 1 ≤ data index j ≤ 2; The mask feature vector Of the two preset vector data, the preset vector data corresponding to the PVS point is greater than the preset vector data corresponding to the non-PVS point. The first fusion layer is used to match the voxel feature vectors of the three-dimensional voxel coordinates in the feature tensor Z4 and the feature tensor Z6. and the voxel feature vector Form a corresponding feature vector pair; and perform a traversal of all the feature vector pairs; during this traversal, take the currently traversed feature vector pair as the current vector pair; and set the voxel feature vector of the current vector pair as the current vector pair. The vector data with the largest median value The corresponding semantic type of the brain region is taken as the current brain region type; and the voxel feature vector of the current vector pair is used as the current brain region type. The vector data with the largest median value The corresponding PVS semantic type is used as the current PVS type; and the current vector pair, the corresponding three-dimensional voxel coordinates, the current brain region type, and the current PVS type are combined to form a voxel feature of the corresponding third voxel point; and all the obtained third voxel points are combined to form the corresponding semantic segmentation map and output.

4. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 2, characterized in that, The input terminal of the first prediction model is used to receive the lymphoid system-like feature X, and the output terminal is used to output the corresponding prediction vector Y. The first prediction model is formed by sequentially connecting the MLP model and the Softmax function layer; The MLP model consists of a hidden layer network and an output layer connected sequentially; the hidden layer network consists of multiple hidden layers connected sequentially; the hidden layer consists of a linear layer and an activation layer connected sequentially; the output layer is a linear layer; the MLP model is used to extract features from the lymphoid system feature X to obtain the corresponding two-dimensional feature vector X1, which is then sent to the Softmax function layer. The two-dimensional feature vector X1 has a length of 2 and consists of two vector data. , composition; The Softmax function layer is used to calculate the corresponding prediction probabilities y1 and y2 based on the two-dimensional feature vector X1 using the Softmax function, and then outputs the corresponding prediction vector Y. The prediction probabilities y1 and y2 are calculated as follows: ， 。 5. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 2, characterized in that, The construction of the first and second datasets based on the original sample set for the training datasets required by the segmentation and prediction models specifically includes: Each of the first sample records in the original sample set is taken as the current sample record; and the object type, the brain DTI image, and the brain 3D-T2 weighted image of the current sample record are taken as the corresponding current object type, current DTI image, and current T2 image. And the current T2 image is used as the corresponding first training image; Based on DTI-ALPS technology, the left and right brain ALPS indices are analyzed according to the current DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index. Based on manual annotation, the first and second label semantic maps corresponding to the current T2 image are labeled with tags. Based on manual statistical methods, the volumes of the four brain regions—midbrain, hippocampus, basal ganglia, and centrum semiovale—are statistically analyzed according to the first label semantic map corresponding to the current T2 image to obtain the corresponding first, second, third, and fourth volumes. Furthermore, based on the second label semantic map corresponding to the current T2 image, the total volume of the perivascular spaces in these four brain regions is statistically analyzed to obtain the corresponding fifth, sixth, seventh, and eighth volumes. Based on these eight statistical volumes, the corresponding PVS-VF scores for the midbrain, hippocampus, basal ganglia, and centrum semiovale are calculated: Midbrain PVS-VF score = fifth volume / first volume; Hippocampus PVS-VF score = sixth volume / second volume; basal ganglia PVS-VF score = seventh volume / third volume; and centrum semiovale PVS-VF score = eighth volume / fourth volume. The first training feature is composed of the left brain ALPS index, the right brain ALPS index, the midbrain PVS-VF score, the hippocampus PVS-VF score, the basal ganglia PVS-VF score, and the centrum semiovale PVS-VF score corresponding to the current sample record. The current object type is identified; if the current object type is a patient, the corresponding first label probability is set to 1 and the second label probability is set to 0; if the current object type is a healthy type, the corresponding first label probability is set to 0 and the second label probability is set to 1; and the first label vector is composed of the first and second label probabilities. The first data record is composed of the first training image, the first label semantic map, and the second label semantic map corresponding to the current sample record; and the second data record is composed of the first training feature and the first label vector corresponding to the current sample record. The first dataset is composed of all the first data records obtained; and the second dataset is composed of all the second data records obtained.

6. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 3, characterized in that, Training the first segmentation model based on the first dataset specifically includes: Step 61: Take each of the first data records in the first dataset as the current record; and input the first training image of the current record as the current brain 3D-T2 weighted image into the first segmentation model for processing; and take the feature tensor Z4 and feature tensor Z6 generated during the current processing of the model as the corresponding predicted semantic graph. , And the currently recorded first tag semantic graph and second tag semantic graph are recorded as the corresponding tag semantic graphs. , ; and from the predicted semantic graph and tag semantic graph Form the corresponding first prediction-label pair ( , ), from the predicted semantic graph and tag semantic graph Form the corresponding second prediction-label pair ( , ); Where 1 ≤ record index u ≤ N1, and N1 is the total number of records in the first dataset; Step 62: Substitute all the obtained first and second prediction-label pairs into the preset first model loss function L. M1 The corresponding first loss value is obtained through calculation; Wherein, the first model loss function L M1 Based on cross-entropy loss function L CE The implementation of the Dice loss function is as follows: ； Step 63: Identify whether the first loss value meets the preset first loss value range; if not, then based on the preset first model optimizer, move towards making the first model loss function L... M1 The model parameters of the first segmentation model are modulated once in the direction that reaches the minimum value, and the process returns to step 61 when the modulation ends; if the condition is met, training is stopped and the training of the first segmentation model is confirmed to be complete. The first model optimizer includes the Adam optimizer and the SGD optimizer.

7. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 2, characterized in that, Training the first prediction model based on the second dataset specifically includes: Step 71: Take each of the second data records in the second dataset as the current record; and input the first training feature of the current record as the current lymphoid system feature X into the first prediction model for processing; and record the prediction vector Y output by the model this time as the prediction vector. And the first tag vector currently recorded is recorded as the corresponding tag vector. ; and by the prediction vector and the label vector Form the corresponding third prediction-label pair ( , ); Where 1 ≤ record index v ≤ N2, and N2 is the total number of records in the second dataset; Step 72: Substitute all the obtained third prediction-label pairs into the preset second model loss function L. M2 The corresponding second loss value is obtained through calculation; Wherein, the second model loss function L M2 Based on cross-entropy loss function L CE Implementation, specifically: ； Step 73: Identify whether the second loss value meets the preset second loss value range; if not, then based on the preset second model optimizer, move towards making the second model loss function L... M2 The model parameters of the first prediction model are modulated once in the direction that reaches the minimum value, and the process returns to step 71 when the modulation ends; if the condition is met, training is stopped and the training of the first prediction model is confirmed to be complete. The second model optimizer includes the Adam optimizer and the SGD optimizer.

8. The method for predicting early Alzheimer's disease based on lymphoid system characteristics according to claim 2, characterized in that, The method involves performing left and right brain ALPS index analysis based on DTI-ALPS technology using the brain DTI images, semantic segmentation of the brain 3D-T2 weighted images based on the first segmentation model, and volume fraction analysis of the perivascular space in the four brain regions based on the segmentation results. The method also includes using the first prediction model to make predictions based on the two types of analysis results. Specifically, this includes: Step 81: Based on DTI-ALPS technology, perform left and right brain ALPS index analysis according to the current brain DTI images to obtain the corresponding left brain ALPS index and right brain ALPS index. Step 82: Input the current brain DTI image into the first segmentation model for processing to obtain the corresponding semantic segmentation map; Step 83: Cluster the third voxel points on the semantic segmentation map whose semantic type is midbrain, hippocampus, basal ganglia, or centrum semiovale into a class to form corresponding first, second, third, or fourth point sets; and assign maximum and minimum depth coordinates d to each point set. max d min Maximum and minimum height coordinates h max h min Maximum and minimum width coordinates w max w min The system performs identification and generates eight vertex coordinates based on the six identified extreme coordinates. The cube space formed by the eight vertex coordinates is used as the first, second, third, or fourth brain region space corresponding to the current point set. Step 84: Cluster the third voxel points on the semantic segmentation map whose PVS semantic type is PVS points and whose three-dimensional voxel coordinates are located in the first, second, third or fourth brain region space into a class to form the corresponding fifth, sixth, seventh or eighth point set; Step 85: Count the total number of voxel points in the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets respectively to obtain the corresponding totals for the first, second, third, fourth, fifth, sixth, seventh, and eighth point sets; and calculate the corresponding PVS-VF scores for the midbrain, hippocampus, basal ganglia, and centrum semiovale based on the obtained totals for the eight point sets. Among them, the midbrain PVS-VF score = the fifth total score / the first total score, the hippocampus PVS-VF score = the sixth total score / the second total score, the basal ganglia PVS-VF score = the seventh total score / the third total score, and the centrum semiovale PVS-VF score = the eighth total score / the fourth total score; Step 86: The lymphoid system feature X, composed of the obtained left brain ALPS index, right brain ALPS index, midbrain PVS-VF score, hippocampal PVS-VF score, basal ganglia PVS-VF score, and centrum semiovale PVS-VF score, is input into the first prediction model for prediction processing to obtain the corresponding prediction vector Y. Step 87: Take the classification type corresponding to the highest probability in the prediction vector Y as the prediction result for the current subject and save it.

9. An apparatus for performing the method for predicting early Alzheimer's disease based on lymphoid system characteristics according to any one of claims 1-8, characterized in that, The device includes: a data acquisition module, a segmentation model construction module, a prediction model construction module, a dataset preparation module, a segmentation model training module, a prediction model training module, and a prediction task processing module. The data acquisition module is used to collect large amounts of brain DTI images and 3D-T2 weighted images of the brains of people with early Alzheimer's disease and healthy people to obtain the corresponding original sample sets. The segmentation model construction module constructs a first segmentation model based on the 3D-UNet model to perform semantic segmentation of perivascular spaces in multiple specified brain regions using the 3D-T2 weighted brain images. The multiple specified brain regions include the midbrain, hippocampus, basal ganglia, and centrum semiovale. The first segmentation model performs semantic segmentation of brain regions and perivascular spaces on the input 3D-T2 weighted brain images and outputs corresponding semantic segmentation maps. The three-dimensional dimensions and resolution of the semantic segmentation maps are consistent with the corresponding 3D-T2 weighted brain images. The voxel semantics of the semantic segmentation maps include brain region semantics and PVS semantics. The prediction model construction module constructs a first prediction model based on the MLP model to perform binary classification prediction of early Alzheimer's disease according to lymphoid system characteristics; the first prediction model is used to perform binary classification prediction based on the lymphoid system characteristics X input to the model and output the corresponding prediction vector Y; The dataset preparation module constructs the corresponding first dataset and second dataset based on the original sample set to obtain the training dataset required for the segmentation and prediction models. The segmentation model training module trains the first segmentation model based on the first dataset; The prediction model training module trains the first prediction model based on the second dataset; The prediction task processing module is used to receive the brain DTI images and the brain 3D-T2 weighted images of any subject after the training of the two types of models is completed; and to perform left and right brain ALPS index analysis based on the brain DTI images using DTI-ALPS technology, and to perform semantic segmentation on the brain 3D-T2 weighted images based on the first segmentation model and to perform perivascular space volume fraction analysis of the four brain regions based on the segmentation results, and to make predictions using the first prediction model based on the two types of analysis results.

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

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

Images