Alzheimer's disease prediction method and system based on dual-omic feature topology adjustment

Abstract

The application relates to the technical field of deep learning and image data processing, in particular to an Alzheimer disease prediction method and system based on double-omics feature topology adjustment, which divides brain regions, respectively extracts image omics features and morphological features, respectively obtains an image omics feature matrix and a morphological node feature matrix, performs morphological correlation calculation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix, performs graph convolution processing on the adjacency matrix to obtain an enhanced topology matrix of each subject, performs sparse processing on the enhanced topology matrix to obtain a sparse topology matrix, performs feature aggregation processing on the image omics feature matrix based on the sparse topology matrix for several times to obtain a fusion feature matrix, performs feature extraction processing, layer normalization processing and classification processing on the fusion feature matrix in sequence after flat processing, and finally obtains an Alzheimer disease prediction result, so that the accuracy of Alzheimer disease prediction can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep learning and image data processing, in particular to an Alzheimer's disease prediction method and system based on double-omics feature topology adjustment. BACKGROUND

[0002] Structural brain network research, as an important research direction in the field of neuroscience, aims to explore the anatomical connections of various regions of the brain and their roles in cognitive function, emotional regulation, motor control, and other aspects. In recent years, with the rapid development of neuroimaging technology, the research on structural brain network has made significant progress. Structural magnetic resonance imaging (sMRI) has become an important tool for studying the anatomical connections of the brain due to its high resolution and non-invasiveness. In particular, the brain structural connectivity network constructed using sMRI data is widely used to explore the anatomical basis of the brain and provides an important theoretical basis for the prediction and evaluation of diseases such as Alzheimer's disease.

[0003] In the construction of structural brain network, morphological features, as an important biological feature, have become an important tool for studying the anatomical structure of the brain. Morphological features directly reflect the size, shape, and local structural changes of brain regions. They provide a more intuitive representation of anatomical connections between different regions of the brain, helping to understand the structural associations between brain regions and providing a solid foundation for the construction of brain networks. In early brain network research, brain networks were constructed based on morphological features.

[0004] In recent years, with the introduction of imageomics features, the construction of brain networks based on imageomics has gradually become a new research direction. Imageomics can reveal the microscopic changes and organizational heterogeneity of brain structure by extracting a large number of quantitative features such as texture and intensity from sMRI images. These details may be difficult to capture directly in morphological features. Therefore, imageomics features have unique advantages in reflecting more detailed physiological connections between brain regions.

[0005] Morphological features and imageomics features have their own advantages in brain network construction and certain complementarity. However, current research on brain networks is mostly based on a single omics, i.e., only based on morphological features or only based on imageomics features to construct and optimize brain networks, without combining the advantages of both omics to construct and optimize brain networks. SUMMARY

[0006] To solve the problems raised in the background art, the present application provides an Alzheimer's disease prediction method and system based on double-omics feature topology adjustment.

[0007] The technical solution of the present application is as follows:

[0008] An Alzheimer's disease prediction method based on double-omics feature topology adjustment, comprising the following steps:

[0009] S1, segmenting the obtained T1-weighted imaging data of the subject to obtain a gray matter volume graph, dividing the preprocessed gray matter volume graph into a plurality of brain regions, extracting a plurality of image omics features of each brain region, and obtaining an image omics feature matrix based on the number of brain regions and the image omics features;

[0010] The T1-weighted imaging data is divided into brain regions, a plurality of morphological features of each brain region are extracted, and a morphological node feature matrix is obtained based on the number of brain regions and the morphological features;

[0011] S2, calculating the morphological correlation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix; performing graph convolution processing on the adjacency matrix to obtain an enhanced topology matrix for each subject, and each element value in the enhanced topology matrix is a connection strength value;

[0012] S3, performing sparse processing on the enhanced topology matrix, and the sparse processing process is:

[0013] Calculate the average value of each connection strength value in the enhanced topology matrix of all subjects, obtain a plurality of distance values according to the difference between different connection strength values and the corresponding average value, sort the plurality of distance values and set different contribution weights respectively, calculate the weighted average value of each connection strength value in the enhanced topology matrix based on different contribution weights, and perform sparse processing on the enhanced topology matrix based on the weighted average value. The value less than the weighted average value in the enhanced topology matrix is set to 0 to obtain a sparse topology matrix;

[0014] S4, performing a plurality of times of feature aggregation processing on the image omics feature matrix based on the sparse topology matrix to obtain a fusion feature matrix, performing feature extraction processing, layer normalization processing and classification processing on the flattened fusion feature matrix in turn, and obtaining an Alzheimer's disease prediction result.

[0015] Specifically, in S3, after sorting the plurality of distance values and setting different contribution weights respectively, the specific process is: after sorting the plurality of distance values from small to large, constructing a first data set based on the first number of distance values in the order, and setting a first contribution weight for all distance values in the first data set; a second data set is constructed based on the remaining number of distance values, and a second contribution weight is set for all distance values in the second data set.

[0016] Further, in S3, the weighted average value of each connection strength value in the enhanced topology matrix is calculated based on different contribution weights, and the formula is:

[0017] ,

[0018] wherein, is a weighted average value, is the n th subject about the connection strength value on the enhanced topology matrix , is the first data set, is the second data set, is the first contribution weight, is the second contribution weight.

[0019] The S3 in the weighted average value is based on the sparse processing of the enhanced topology matrix, the value less than the weighted average value in the enhanced topology matrix is set to 0, and the sparse topology matrix is obtained, and the formula is expressed as:

[0020] ,

[0021] wherein, is the connection strength value in the sparse topology matrix, is a weighted average value.

[0022] Specifically, the S4 in the sparse topology matrix is based on the imageomics feature matrix for a plurality of times of feature aggregation processing to obtain a fusion feature matrix, and the process of feature aggregation processing is:

[0023] Based on the connection strength value in the sparse topology matrix, the adjacent node features of each node feature in the imageomics feature matrix are aggregated, and the updated node feature is obtained. The updated node feature is sequentially subjected to feature normalization, nonlinear activation and loss processing to obtain a first fusion feature matrix as the input of the next feature aggregation processing.

[0024] Specifically, the S2 in the adjacency matrix is subjected to graph convolution processing, and the process of graph convolution processing is:

[0025] The adjacency matrix is subjected to normalization processing to obtain a first adjacency matrix, and the adjacency matrix is sequentially subjected to activation function processing and loss layer processing based on the first adjacency matrix and the set weight matrix to obtain an enhanced topology matrix.

[0026] Further, the S2 in the morphological node feature matrix is based on the calculation of morphological correlation between all brain regions, and the formula is expressed as:

[0027] ,

[0028] wherein, is the morphological correlation between brain region i and brain region j, and k is the serial number of the morphological feature, ; is the numerical value of the k th morphological feature on the i th brain region, is the average value of all morphological features on the i-th brain region, is the value of the k-th morphological feature on the j-th brain region, is the average value of all morphological features on the j-th brain region.

[0029] Specifically, the several image features in S1 include: brain region texture, intensity, shape.

[0030] Specifically, the several morphological features in S1 include: brain region volume, surface area, thickness, curvature.

[0031] The application also provides an Alzheimer's disease prediction system based on double-omics feature topology adjustment, comprising:

[0032] The feature extraction module is used for segmenting the obtained T1-weighted imaging data of the subject, obtaining a gray matter volume graph, dividing the gray matter volume graph into several brain regions after preprocessing, extracting several image features of each brain region, and obtaining an image feature matrix based on the number of brain regions and the image features.

[0033] The T1-weighted imaging data is divided into brain regions, and several morphological features of each brain region are extracted, and a morphological node feature matrix is obtained based on the number of brain regions and the morphological features.

[0034] The feature enhancement module is used for calculating the morphological correlation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix; and performing graph convolution processing on the adjacency matrix to obtain an enhanced topology matrix of each subject, and each element value in the enhanced topology matrix is a connection strength value.

[0035] The sparse processing module is used for sparse processing of the enhanced topology matrix, and the sparse processing process is as follows:

[0036] The average value of each connection strength value in the enhanced topology matrix of all subjects is calculated, the difference between different connection strength values and the corresponding average value is obtained to obtain several distance values, the several distance values are sorted and different contribution weights are set respectively, the weighted average value of each connection strength value in the enhanced topology matrix is calculated based on the different contribution weights, the enhanced topology matrix is sparse processed based on the weighted average value, the values less than the weighted average value in the enhanced topology matrix are set to 0, and a sparse topology matrix is obtained.

[0037] The prediction module is used for performing several times of feature aggregation processing on the image feature matrix based on the sparse topology matrix to obtain a fusion feature matrix, performing feature extraction processing, layer normalization processing and classification processing on the fusion feature matrix in turn after flattening processing, and obtaining an Alzheimer's disease prediction result.

[0038] The present application has the beneficial effects of:

[0039] 1、The present application extracts image genomics features of different brain regions to obtain an image genomics feature matrix, extracts morphological features to obtain a morphological node feature matrix, obtains an adjacency matrix based on the morphological node feature matrix, and then performs graph convolution processing to obtain an enhanced topology matrix, further performs sparse processing on the enhanced topology matrix, calculates the weighted average value of each connection strength value in the enhanced topology matrix based on different contribution weights, sets the values less than the weighted average value in the enhanced topology matrix to 0, which helps to highlight important connections of brain regions and weaken irrelevant connections, and further captures the key connections between brain regions, making the topology structure of brain network more clear and further improving the accuracy of Alzheimer's disease prediction.

[0040] 2、The present application performs several times of feature aggregation processing on the image genomics feature matrix based on the sparse topology matrix to obtain a fusion feature matrix, combines the morphological features and the image genomics features, which can make up for the deficiency of morphological features in fine representation and provide more comprehensive brain region connection and functional features; by fusing the image genomics feature matrix and the topology relationship between brain regions, the complex relationship between brain regions is captured, high-level feature representation is extracted, and the Alzheimer's disease prediction result is further obtained, which significantly improves the analysis ability of complex structure and functional connection of brain, and has important scientific research and clinical application prospect. DETAILED DESCRIPTION

[0041] The exemplary embodiments of the present disclosure are described in more detail below.

[0042] The present embodiment provides an Alzheimer's disease prediction method based on dual-omics feature topology adjustment, comprising the following steps:

[0043] S1, the obtained T1 weighted imaging data of the subject is segmented and processed to obtain a gray matter volume graph, the gray matter volume graph is preprocessed and divided into several brain regions, several image genomics features of each brain region are extracted, and an image genomics feature matrix is obtained based on the number of brain regions and the image genomics features;

[0044] The T1 weighted imaging data is divided into brain regions, several morphological features of each brain region are extracted, and a morphological node feature matrix is obtained based on the number of brain regions and the morphological features.

[0045] In step S1, the obtained T1-weighted imaging data of the subject is subjected to segmentation processing to obtain a gray matter volume graph, and after deviation correction processing and registration processing of the gray matter volume graph, the registered gray matter volume graph is divided into a plurality of brain regions according to automatic anatomical markers, and in this embodiment, the gray matter volume graph is divided into 68 brain regions. Then, a plurality of image features of each brain region are extracted, for example, 25 image features are extracted. The plurality of image features include: brain region texture, intensity, shape.

[0046] Further based on the number of brain regions and the image features, an image feature matrix is obtained, and in this embodiment, a 68*25 image feature matrix is obtained.

[0047] It also includes brain region division according to automatic anatomical markers based on T1-weighted imaging data, and the brain is also divided into 68 brain regions. After skull removal and motion correction series processing, a plurality of morphological features of each brain region are extracted, for example, 9 morphological features are extracted. The plurality of morphological features include: brain region volume, surface area, thickness, curvature.

[0048] Further based on the number of brain regions and the morphological features, a morphological node feature matrix is obtained, and in this embodiment, a 68*9 morphological node feature matrix is obtained.

[0049] S2, based on the morphological node feature matrix, the morphological correlation between all brain regions is calculated to obtain an adjacency matrix; the adjacency matrix is subjected to graph convolution processing to obtain an enhanced topology matrix of each subject, and each element value in the enhanced topology matrix is a connection strength value.

[0050] Based on the morphological node feature matrix, the morphological correlation between all brain regions is calculated, and the formula is represented as:

[0051] ,

[0052] Wherein, is the morphological correlation between brain region i and brain region j, k is the serial number of the morphological feature, ; is the value of the kth morphological feature on the ith brain region, is the average value of all morphological features on the ith brain region, is the value of the kth morphological feature on the jth brain region, is the average value of all morphological features on the jth brain region.

[0053] The adjacency matrix is subjected to graph convolution processing, and the process of graph convolution processing is:

[0054] The adjacency matrix is normalized to obtain a first adjacency matrix. The adjacency matrix is sequentially subjected to an activation function processing and a loss layer processing based on the first adjacency matrix and a set weight matrix to obtain an enhanced topology matrix. The process is represented by the formula:

[0055] ,

[0056] wherein, is the output of the graph convolution processing, is the first adjacency matrix, is the adjacency matrix, is the set weight matrix, is the used activation function, is the loss layer processing.

[0057] S3, the enhanced topology matrix is subjected to a sparsification processing, and the sparsification processing process is:

[0058] The average value of each connection strength value in the enhanced topology matrix of all subjects is calculated. A plurality of distance values are obtained according to the difference between different connection strength values and the corresponding average value. After sorting the plurality of distance values, different contribution weights are respectively set. The weighted average value of each connection strength value in the enhanced topology matrix is calculated based on the different contribution weights. The enhanced topology matrix is subjected to a sparsification processing based on the weighted average value. The values less than the weighted average value in the enhanced topology matrix are set to 0 to obtain a sparse topology matrix.

[0059] In step S3, the average value of each connection strength value in the enhanced topology matrix of all subjects is first calculated , and the formula is represented as:

[0060] ,

[0061] wherein, is the number of subjects, is the connection strength value of the i-th subject on the enhanced topology matrix n .

[0062] Then, a plurality of distance values are obtained according to the difference between different connection strength values and the corresponding average value , and the formula is represented as:

[0063] ,

[0064] wherein, is the connection strength value of the i-th subject on the enhanced topology matrix n . ​​is the average value of each connection strength in the augmented topology matrix of all subjects.

[0065] After sorting several distance values, different contribution weights are set respectively. Specifically, after sorting several distance values ​​from small to large, a first data set is constructed based on the first number of distance values ​​in the sorting order. Indicates that a certain number of distance values ​​in the first order can be set to the distance values ​​in the first 20% after ordering. The first contribution weight is set for all distance values ​​in the first data set, and express;

[0066] Construct a second dataset based on the remaining number of distance values, using Indicates that the second contribution weight is set for all distance values ​​in the second data set, and express.

[0067] The weighted average value of each connection strength value in the enhanced topology matrix is ​​calculated based on different contribution weights. The formula is expressed as:

[0068] ,

[0069] in, is the weighted average, For the nth subject in the enhanced topology matrix The connection strength value on For the first data set, For the second data set, is the first contribution weight, is the second contribution weight.

[0070] With this weighting method, 20% of the distance values ​​in the first dataset contribute 80% of the weight, and the remaining 80% of the distance values ​​contribute 20% of the weight.

[0071] The enhanced topology matrix is ​​sparsely processed based on the weighted average value, and the values ​​in the enhanced topology matrix that are less than the weighted average value are set to 0 to obtain a sparse topology matrix. The formula is expressed as:

[0072] ,

[0073] in, is the connection strength value in the sparse topological moment, is the weighted average.

[0074] S4. Based on the sparse topological matrix, the imaging omics feature matrix is ​​subjected to several feature aggregation processes to obtain a fused feature matrix. After flattening the fused feature matrix, feature extraction, layer normalization and classification are performed in sequence to obtain the Alzheimer's disease prediction results.

[0075] Specifically, the imageomics feature matrix is subjected to several times of feature aggregation processing based on the sparse topology matrix to obtain a fusion feature matrix, and the process of the feature aggregation processing is as follows:

[0076] Based on the connection strength value in the sparse topology matrix, the adjacent node features of each node feature in the imageomics feature matrix are aggregated to obtain an updated node feature, and the updated node feature is subjected to feature normalization, nonlinear activation and loss processing in sequence to obtain a first fusion feature matrix as the input of the next time of feature aggregation processing.

[0077] The above process can be expressed by a formula as follows:

[0078] ,

[0079] wherein, is the output after the feature aggregation processing, is the sparse topology matrix, is the imageomics feature matrix, , is a different weight matrix, is a used nonlinear activation function, represents loss processing.

[0080] The application further provides an Alzheimer's disease prediction system based on dual-omics feature topology adjustment, comprising:

[0081] a feature extraction module: used for performing segmentation processing on the acquired T1-weighted imaging data of a subject to obtain a gray matter volume graph, dividing the gray matter volume graph into a plurality of brain regions after preprocessing, extracting a plurality of imageomics features of each brain region, and obtaining an imageomics feature matrix based on the number of brain regions and the imageomics features;

[0082] performing brain region division on the T1-weighted imaging data, extracting a plurality of morphological features of each brain region, and obtaining a morphological node feature matrix based on the number of brain regions and the morphological features;

[0083] a feature enhancement module: used for performing morphological correlation calculation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix; performing graph convolution processing on the adjacency matrix to obtain an enhanced topology matrix of each subject, and each element value in the enhanced topology matrix being a connection strength value;

[0084] a sparse processing module: used for performing sparse processing on the enhanced topology matrix, and the sparse processing process being as follows:

[0085] Calculate the average value of each connection strength value in the enhanced topology matrix of all subjects, obtain several distance values ​​based on the difference between different connection strength values ​​and the corresponding average value, sort the several distance values ​​and set different contribution weights respectively, calculate the weighted average value of each connection strength value in the enhanced topology matrix based on the different contribution weights, perform sparse processing on the enhanced topology matrix based on the weighted average value, set the values ​​in the enhanced topology matrix that are less than the weighted average value to 0, and obtain a sparse topology matrix;

[0086] Prediction module: It is used to perform several feature aggregation processes on the imaging omics feature matrix based on the sparse topological matrix to obtain a fused feature matrix. After flattening the fused feature matrix, it performs feature extraction, layer normalization and classification in sequence to obtain the Alzheimer's disease prediction results.

[0087] Experimental content

[0088] Experimental setup: The dataset was randomly divided into training, validation, and test sets in a ratio of 8:1:1. The experimental development environment used was pytorch 2.0.1. Training was performed on an NVIDIA GeForce RTX 4060 Laptop GPU using the Adam optimizer with an initial learning rate of 3e. -4 , the batch size is set to 4.

[0089] The experimental results are shown in Table 1.

[0090] Table 1 Experimental results

[0091]

[0092] Among them, TP represents the correctly predicted diseased individuals, TN represents the correctly predicted disease-free individuals, FP represents the incorrectly predicted diseased individuals (actually disease-free), and FN represents the incorrectly predicted disease-free individuals (actually diseased).

[0093] The accuracy in Table 1 is the proportion of correctly predicted diseased and disease-free individuals, and the formula is as follows:

[0094] ,

[0095] Sensitivity is the proportion of individuals correctly predicted to be sick, also known as the true positive rate, and the formula is as follows:

[0096] ,

[0097] Specificity is the proportion of individuals who are correctly predicted to be disease-free, also known as the true negative rate, and is expressed as follows:

[0098] .

[0099] The specific operation process is as follows:

[0100] When only morphological features are considered, the T1 weighted imaging data of the subject is divided into brain regions according to the automatic anatomical markers as described in step S1, and the brain is divided into 68 brain regions, 9 morphological features are extracted from each brain region, a morphological node feature matrix of 68*9 is obtained, the morphological correlation between each two brain regions is calculated, and a 68*68 adjacency matrix is obtained. Using the morphological node feature matrix of 68*9 and the 68*68 adjacency matrix, a plurality of times of feature fusion processing are performed to obtain a fusion feature matrix, and the fusion feature matrix is further flattened and processed, and then feature extraction processing, layer normalization processing and classification processing are sequentially performed, to obtain an Alzheimer's disease prediction result, and the parameters that perform best on the validation set are saved for actual Alzheimer's disease prediction.

[0101] When only image features are considered, the acquired T1 weighted imaging is segmented into a gray matter volume graph, the gray matter volume graph is subjected to deviation correction processing and registration processing, and then the registered gray matter volume graph is divided into 68 brain regions according to the automatic anatomical markers, 25 corresponding image features are extracted from each brain region, an image feature matrix of 68*25 is obtained, and then the image feature correlation between each two brain regions is calculated based on the image feature matrix, to obtain a 68*68 adjacency matrix. Using the image feature matrix of 68*25 and the corresponding 68*68 adjacency matrix, a plurality of times of feature fusion processing are performed to obtain a fusion feature matrix, and the fusion feature matrix is further flattened and processed, and then feature extraction processing, layer normalization processing and classification processing are sequentially performed, to obtain an Alzheimer's disease prediction result, and the parameters that perform best on the validation set are saved.

[0102] It can be seen that the Alzheimer's disease prediction method provided by the application has high accuracy, sensitivity and specificity in predicting Alzheimer's disease.

Claims

1.A method for Alzheimer's disease prediction based on biomic feature topology adjustment, characterized in that, The method comprises the following steps: S1, segmenting the acquired T1-weighted imaging data of the subject to obtain a gray matter volume graph, dividing the gray matter volume graph into a plurality of brain regions after preprocessing, extracting a plurality of image feature characteristics of each brain region, and obtaining an image feature characteristic matrix based on the number of brain regions and the image feature characteristics; S2, calculating the morphological correlation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix; S3, performing sparse processing on the enhanced topology matrix, the sparse processing process being: S4, performing a plurality of times of feature aggregation processing on the image feature characteristic matrix based on the sparse topology matrix to obtain a fusion feature matrix, performing flattening processing on the fusion feature matrix, and then sequentially performing feature extraction processing, layer normalization processing and classification processing to obtain an Alzheimer's disease prediction result. In the S3, after sorting the plurality of distance values from small to large, a first data set is constructed based on the first number of distance values in the front of the sorting, and a first contribution weight is set for all distance values in the first data set; a second data set is constructed based on the remaining number of distance values, and a second contribution weight is set for all distance values in the second data set. In the S3, the weighted average value of each connection strength value in the enhanced topology matrix is calculated based on the different contribution weights, and the formula is: In the S3, the enhanced topology matrix is sparsified based on the weighted average value, the values less than the weighted average value in the enhanced topology matrix are set to 0, and the sparse topology matrix is obtained, and the formula is: 2.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, In the S4, a plurality of times of feature aggregation processing are performed on the image feature characteristic matrix based on the sparse topology matrix to obtain a fusion feature matrix, and the process of feature aggregation processing is: 3.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 2, wherein, In the S2, the graph convolution processing of the adjacency matrix is: ， wherein, is a weighted average, is the n-th subject's value for the connection strength value on the augmented topology matrix , is the first data set, is the second data set, is the first contribution weight, is the second contribution weight. 4.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, In the S2, the graph convolution processing of the adjacency matrix is: ， wherein, is the connection strength value in the sparse topology matrix, is the weighted average. 5.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, ​ ​ 6.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, ​ The adjacency matrix is normalized to obtain a first adjacency matrix, and the adjacency matrix is sequentially subjected to an activation function processing and a loss layer processing based on the first adjacency matrix and a set weight matrix to obtain an enhanced topology matrix. 7.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, The morphological correlation calculation between all brain regions is performed based on the morphological node feature matrix in S2, and the formula is represented as: ， wherein, is the morphological correlation between brain region i and brain region j, k is the index of the morphological feature, ; is the value of the kth morphological feature on the ith brain region, is the average of all morphological features on the ith brain region, is the value of the kth morphological feature on the jth brain region, is the average of all morphological features on the jth brain region. 8.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, The several imageomics features in S1 include: brain region texture, intensity, shape. 9.The Alzheimer's disease prediction method based on dual-omics feature topology adjustment of claim 1, wherein, The several morphological features in S1 include: brain region volume, surface area, thickness, curvature. 10.A system for Alzheimer's disease prediction based on biomics feature topology adjustment, characterized in that, Comprise: The feature extraction module is used for segmenting the obtained T1 weighted imaging data of the subject to obtain a gray matter volume graph, and the gray matter volume graph is divided into several brain regions after preprocessing, and several imageomics features of each brain region are extracted, and an imageomics feature matrix is obtained based on the number of brain regions and imageomics features; The T1 weighted imaging data is divided into brain regions, and several morphological features of each brain region are extracted, and a morphological node feature matrix is obtained based on the number of brain regions and morphological features; The feature enhancement module is used for calculating the morphological correlation between all brain regions based on the morphological node feature matrix to obtain an adjacency matrix; The adjacency matrix is subjected to graph convolution processing to obtain an enhanced topology matrix of each subject, and each element value in the enhanced topology matrix is a connection strength value; The sparse processing module is used for sparse processing of the enhanced topology matrix, and the sparse processing process is: The average value of each connection strength value in the enhanced topology matrix of all subjects is calculated, the difference between different connection strength values and the corresponding average value is obtained to obtain several distance values, the several distance values are sorted and different contribution weights are set respectively, the weighted average value of each connection strength value in the enhanced topology matrix is calculated based on different contribution weights, and the enhanced topology matrix is subjected to sparse processing based on the weighted average value. The value less than the weighted average value in the enhanced topology matrix is set to 0 to obtain a sparse topology matrix; The prediction module is used for performing several times of feature aggregation processing on the imageomics feature matrix based on the sparse topology matrix to obtain a fusion feature matrix, and the fusion feature matrix is subjected to flattening processing, and then subjected to feature extraction processing, layer normalization processing and classification processing in sequence to obtain an Alzheimer's disease prediction result.

Images