A method, device, medium and product for determining a comprehensive brain aging index

Abstract

The present application discloses a method, device, medium and product for determining a comprehensive brain aging index, relating to the field of image data processing. The method includes: preprocessing images such as T1 and TOF-MRA to obtain brain tissue segmentation images; post-processing TOF-MRA magnetic resonance images to extract cerebrovascular imaging features; using a method combining modules such as asymmetric convolution to perform convolution feature extraction on T1 and T2-FLAIR images, and compressing the features to generate a convolution feature group; after connecting demographic indicators, cerebrovascular imaging features and the convolution feature group, using a biological age prediction model to obtain a brain aging age prediction result, and combining with the actual age to obtain a comprehensive brain aging index. The present application can identify changes in large intracranial blood vessels, and can also reflect the state of earlier microvascular aging, improve the sensitivity to microvascular aging, solve the limitations such as low efficiency and poor consistency of manual feature extraction, and achieve comprehensive quantitative evaluation of brain aging.

Description

Technical Field

[0001] This application relates to the field of image data processing, and particularly to a method, device, medium, and product for determining a comprehensive brain aging index. Background Art

[0002] Although certain progress has been made in evaluating peripheral vascular aging, the quantitative assessment techniques for intracranial vascular aging still face many challenges. Among them:

[0003] (1) Limitations of single-modal imaging: Traditional imaging methods usually rely on imaging data of a single modality, such as Computed Tomography (CT), conventional Magnetic Resonance Imaging (MRI), etc. Although these methods can provide certain vascular morphological information, they cannot comprehensively capture the multi-dimensional characteristics of vascular aging. Especially for the aging state of microvessels, it is often difficult for single-modal imaging to reflect its early changes. For example, many structural features in T1-weighted imaging and Fluid Attenuated Inversion Recovery (T2-FLAIR) images of magnetic resonance imaging are actually another manifestation of vascular aging and may reflect the earlier aging state of microvessels. Therefore, it is difficult to comprehensively evaluate the degree of cerebrovascular aging relying only on single-modal imaging data.

[0004] (2) Lack of sensitivity to microvascular aging: Existing imaging techniques and assessment methods mainly focus on the structural changes of large blood vessels (such as wall thickening, plaque formation, etc.), while relatively neglecting the aging state of microvessels. Microvascular aging is one of the important risk factors for cerebrovascular diseases. Especially in the early stage, the changes in microvessels may be the earliest pathological features. However, existing assessment means cannot effectively capture these subtle changes, resulting in an incomplete and inaccurate assessment of cerebrovascular aging.

[0005] (3) Limitations of manual feature extraction: Traditional methods for vascular segmentation and feature extraction of Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) usually rely on manually designed algorithms. This method is not only time-consuming and laborious but also easily affected by subjective factors, making it difficult to comprehensively capture complex vascular aging features. Especially when dealing with multi-modal image data, the method of manual feature extraction often cannot fully utilize the complementary information between different modalities, limiting the performance and reliability of the model.

[0006] (4)Lack of quantitative assessment tools: Currently, although there are some imaging-based methods for evaluating vascular aging, most of them remain at the qualitative or semi-quantitative level, lacking a unified standard that can quantitatively and objectively measure the degree of cerebrovascular aging. This makes it lack a reliable reference index for clinicians to evaluate the vascular health status of patients, affecting the early screening and risk assessment of diseases. Summary of the Invention

[0007] To solve the above problems, the present application provides a method, device, medium and product for determining a comprehensive brain aging index.

[0008] To achieve the above object, the present application provides the following solutions:

[0009] In a first aspect, the present application provides a method for determining a comprehensive brain aging index, including:

[0010] Obtain T1-weighted imaging images, T2-FLAIR images, and TOF-MRA magnetic resonance images;

[0011] Preprocess the T1-weighted imaging images and TOF-MRA magnetic resonance images to obtain preprocessed images;

[0012] Obtain cerebrovascular tissue segmentation images based on the preprocessed images;

[0013] Post-process the cerebrovascular tissue segmentation images based on the TOF-MRA magnetic resonance images to obtain post-processed images;

[0014] Extract cerebrovascular imaging features based on the post-processed images;

[0015] Use a method combining an asymmetric convolution module and a Squeeze-and-Excitation module to perform convolution feature extraction on the T1-weighted imaging images and T2-FLAIR images respectively to obtain T1-weighted imaging features and T2-FLAIR image features;

[0016] Compress the T1-weighted imaging features and T2-FLAIR image features through global pooling to generate T1-weighted imaging and T2-FLAIR image convolution feature groups with each feature dimension being a set value;

[0017] Obtain demographic indicators; the demographic indicators include gender and age;

[0018] Connect the demographic indicators, cerebrovascular imaging features, and T1-weighted imaging and T2-FLAIR image convolution feature groups to generate a cerebrovascular aging imaging characterization group;

[0019] Construct a biological age prediction model;

[0020] Input the imaging features of cerebral vascular aging into the biological age prediction model to obtain the prediction result of brain aging age;

[0021] Based on the prediction result of brain aging age and the actual age of the evaluation object, obtain the comprehensive index of brain aging.

[0022] Optionally, obtain the cerebral vascular tissue segmentation image based on the preprocessed image, including:

[0023] Construct a training and testing data set; the training and testing data set includes paired data of T1-weighted imaging images and TOF-MRA magnetic resonance images;

[0024] Based on the 3D-UNet convolutional neural network, combine the attention gating module to construct a deep neural network model;

[0025] Use the training and testing data set to train and test the deep neural network model. When the output error of the deep neural network model after training and testing reaches the set threshold, obtain the trained deep neural network model;

[0026] Take the trained deep neural network model as the cerebral vascular tissue segmentation model, and input the preprocessed image into the cerebral vascular tissue segmentation model to obtain the cerebral vascular tissue segmentation image.

[0027] Optionally, based on the Tversky function, use the weighted Tversky loss to construct the loss function of the cerebral vascular tissue segmentation model.

[0028] Optionally, preprocess the T1-weighted imaging image and the TOF-MRA magnetic resonance image to obtain the preprocessed image, including:

[0029] Perform brain tissue-specific recognition and segmentation on the TOF-MRA magnetic resonance image to obtain the brain tissue structure image;

[0030] Using the method of linear rigid body transformation and spline interpolation, register the T1-weighted imaging image and the TOF-MRA magnetic resonance image based on the brain tissue structure image, and resample the registered structure to the same image space to obtain the transformed image;

[0031] Use the adaptive non-local mean algorithm to perform image denoising processing on the TOF-MRA magnetic resonance image to obtain the denoised image;

[0032] Use the signal intensity remapping method after Gaussian deconvolution to remove the magnetic resonance image artifacts caused by the B0 field inhomogeneity in the denoised image to obtain the intermediate image;

[0033] Use the method of restricted contrast adaptive histogram equalization to process the intermediate image to obtain the first signal enhancement image;

[0034] Using the Sobel convolution method, identify the edge positions where signal intensity mutations occur in the first signal-enhanced image, and obtain a second signal-enhanced image based on the identified edge positions;

[0035] Process the second signal-enhanced image using the Gamma correction method to obtain the preprocessed image.

[0036] Optionally, post-process the cerebrovascular tissue segmentation image to obtain a post-processed image, including:

[0037] Using a three-dimensional conditional random field model, perform penalty processing on adjacent points with similar signal intensities in the cerebrovascular tissue segmentation image to obtain a three-dimensional space segmentation image; the adjacent points with similar signal intensities refer to pixel points whose signal intensity difference is within a set range;

[0038] Find the three-dimensional maximum connected component in the three-dimensional space segmentation image, and remove unconnected signal noise in the three-dimensional space segmentation image based on the three-dimensional maximum connected component to obtain the post-processed image.

[0039] Optionally, use a method combining an asymmetric convolution module and a Squeeze-and-Excitation module to perform convolution feature extraction on the T1-weighted imaging image and the T2-FLAIR image respectively to obtain T1-weighted imaging features and T2-FLAIR image features, including:

[0040] In the convolutional neural network encoder, set two connections of the asymmetric convolution module and the Squeeze-and-Excitation module to obtain a new convolutional neural network encoder;

[0041] Use the new convolutional neural network encoder to perform convolution feature extraction on the T1-weighted imaging image and the T2-FLAIR image respectively to obtain the T1-weighted imaging features and the T2-FLAIR image features.

[0042] Optionally, the biological age prediction model is constructed using a fully connected neural network architecture with a single hidden layer.

[0043] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the brain aging comprehensive index determination method provided above.

[0044] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-provided method for determining the comprehensive brain aging index are implemented.

[0045] In a fourth aspect, the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the above-provided method for determining the comprehensive brain aging index are implemented.

[0046] According to the specific embodiments provided by the present application, the present application has the following technical effects:

[0047] The present application provides a method, device, medium, and product for determining a comprehensive brain aging index. By extracting aging structures using T1-weighted imaging images and T2-FLAIR images, the limitations of single-modal imaging are avoided. It can not only provide more abundant vascular aging information but also capture structural features that are not fully emphasized in other modalities, can reflect the earlier microvascular aging state, and improve the sensitivity to microvascular aging. Moreover, by using a neural network model for feature extraction in the present application, the problem of limitations in manual feature extraction can be solved. The present application obtains the comprehensive brain aging index through the predicted brain aging age and the actual age of the evaluation object, and can realize the quantitative evaluation of cerebral vascular aging. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or in the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1 It is a schematic flowchart of a method for determining a comprehensive brain aging index provided by an embodiment of the present application;

[0050] Figure 2 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0052] Based on magnetic resonance angiography image segmentation and the extraction of multi-level cerebrovascular-specific standardized imaging features, the method for determining the comprehensive brain aging index provided by this application performs the fusion of multi-modal brain imaging features to construct a global imaging index of cerebrovascular aging. Through this index, the overall level of cerebrovascular aging at the individual level can be quantitatively and objectively evaluated to assist in the actual work of relevant clinical scientific research.

[0053] To make the above objects, features, and advantages of this application more obvious and understandable, the following further details this application in conjunction with the accompanying drawings and specific embodiments.

[0054] In an exemplary embodiment, this application provides a method for determining a comprehensive brain aging index. This method is executed by a computer device, which can specifically be executed alone by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of this application, this method is described by taking its application to a server as an example. As Figure 1 shown, this method includes:

[0055] Step 100: Obtain T1-weighted imaging images, T2-FLAIR images, and TOF-MRA magnetic resonance images.

[0056] Step 101: Preprocess the T1-weighted imaging images and TOF-MRA magnetic resonance images to obtain preprocessed images. In the actual application process, in order to ensure the accuracy of the data, it is also necessary to preprocess the T2-FLAIR images, including noise reduction processing and other methods.

[0057] Step 102: Obtain cerebrovascular tissue segmentation images based on the preprocessed images.

[0058] Step 103: Post-process the cerebrovascular tissue segmentation images based on the TOF-MRA magnetic resonance images to obtain post-processed images.

[0059] Step 104: Extract cerebrovascular imaging features based on the post-processed images. The cerebrovascular imaging features include: basic cerebrovascular features, cerebrovascular geometric features, cerebrovascular topological features, and cerebrovascular morphological features. The basic cerebrovascular features include: the length and volume of cerebrovascular vessels, and the diameter and length of local brain regions. The local brain regions are determined based on voxels. The cerebrovascular geometric features include: the curvature radius of cerebrovascular vessels and the degree of curvature of cerebrovascular vessels. The cerebrovascular topological features include: the angle of cerebrovascular bifurcation points and the cerebrovascular bifurcation density. The cerebrovascular morphological features include the distribution density of cerebrovascular vessels in the standard atlas space.

[0060] Step 105: Use the method of combining the asymmetric convolution module and the Squeeze-and-Excitation module to perform convolution feature extraction on the T1-weighted imaging image and the T2-FLAIR image respectively, and obtain the T1-weighted imaging features and the T2-FLAIR image features.

[0061] Step 106: Compress the T1-weighted imaging features and the T2-FLAIR image features through global pooling to generate T1-weighted imaging and T2-FLAIR image convolution feature groups with each feature dimension being a set value.

[0062] Step 107: Obtain demographic indicators. The demographic indicators include gender and age.

[0063] Step 108: Connect the demographic indicators, cerebrovascular imaging features with the T1-weighted imaging and T2-FLAIR image convolution feature groups to generate a cerebrovascular aging imaging characterization group.

[0064] Step 109: Construct a biological age prediction model.

[0065] Step 110: Input the cerebrovascular aging imaging characterization group into the biological age prediction model to obtain the brain aging age prediction result.

[0066] Step 111: Obtain a comprehensive brain aging index based on the brain aging age prediction result and the actual age of the evaluation object. The evaluation object refers to the user who needs to determine the comprehensive brain aging index.

[0067] Based on the above description, by implementing the above steps 100 to 11, the present application can overcome the limitations of the prior art and achieve a comprehensive and quantitative evaluation of cerebrovascular aging. In particular, through the early evaluation of microvascular aging, the present application can provide strong support for the early screening, risk assessment and personalized treatment of cerebrovascular diseases, and has broad application prospects.

[0068] In another exemplary embodiment of the present application, the implementation process of step 101 may include:

[0069] 11. Perform brain tissue-specific recognition and segmentation on the T1-weighted imaging image and the TOF-MRA magnetic resonance image, and remove non-brain tissue structures such as the scalp, skull, facial features and neck to obtain a brain tissue structure image. Implementing this step can eliminate the interference of non-brain tissue structure image signals on subsequent processing.

[0070] 12. Using the methods of linear rigid body transformation and spline interpolation, based on brain tissue structure images, register the T1-weighted imaging images and TOF-MRA magnetic resonance images, and resample the registered structures to the same image space to obtain transformed images. For example, use the methods of linear rigid body transformation with 6 degrees of freedom and spline interpolation to register the T1-weighted imaging and TOF-MRA magnetic resonance images of the same individual, so as to transform each sequence image under different voxel coordinates to the same anatomical coordinates, thereby facilitating the integration of cross-sequence signal information.

[0071] 13. Use the adaptive non-local mean algorithm to perform image denoising processing on the TOF-MRA magnetic resonance images to obtain denoised images, so as to remove the possible Gaussian noise in the images.

[0072] 14. Use the signal intensity remapping method after Gaussian deconvolution to remove the magnetic resonance image artifacts caused by the inhomogeneity of the B0 field in the denoised images to obtain intermediate images, so as to remove the common artifact signals in the magnetic resonance images and improve the signal-to-noise ratio of the images.

[0073] 15. Due to the principle limitations of vascular imaging magnetic resonance sequences such as TOF-MRA, the signal enhancement amplitude of small blood vessels will be significantly lower than that of large blood vessels, resulting in difficulties in identifying small blood vessels. To solve this problem, this application uses the restricted contrast adaptive histogram equalization method to process the intermediate images, adaptively enhancing the local cerebrovascular signals to obtain the first signal-enhanced image. This step enables cerebrovascular signals of different scales to obtain similar enhancement amplitudes.

[0074] 16. Due to the existence of the partial volume effect, the vascular edge signals are significantly lower than the vascular center signals and are thus not easily recognized. To solve this problem, this application uses the method of Sobel convolution to identify the edge positions where the signal intensity changes abruptly in the first signal-enhanced image, and obtains the second signal-enhanced image based on the identified edge positions, so as to achieve the specific enhancement of vascular edge signals and improve the contrast between blood vessels and surrounding non-vascular tissues. This step enables cerebrovascular signals at different distances from the center line to obtain similar enhancements.

[0075] 17. Use the method of Gamma correction to process the second signal-enhanced image. By adjusting the transformation coefficient Gamma, non-linearly enhance the vascular tissue signals (high signals) and suppress the non-vascular tissue signals (low signals) in the magnetic resonance images to obtain preprocessed images. Among them, the transformation coefficient Gamma, as a hyperparameter for processing, can control the relative amplitudes of enhancing and suppressing signals according to the properties of the processed images to obtain better image contrast.

[0076] In another exemplary embodiment of this application, the implementation process of step 102 includes:

[0077] 21. Construct a training and testing dataset. For example, collect the paired T1-weighted imaging and TOF-MRA data of 100 healthy individuals acquired by different devices using clinical routine scanning sequences. After preprocessing the image data in the manner of the above steps 11 - 17, two neuroradiologists with rich imaging annotation experience independently annotate the cerebrovascular tissue signals in the magnetic resonance images, and a senior neuroradiologist comprehensively examines and evaluates the annotation results to complete the annotation of the cerebrovascular tissue signals in the multi-modal magnetic resonance image data, ensuring the consistency and accuracy of the data annotation. Finally, a training and testing dataset for the supervised model is constructed.

[0078] 22. Based on the 3D-UNet convolutional neural network, combine an attention gating module to construct a deep neural network model. Among them, the overall 3D-UNet convolutional neural network can adopt an adaptive design, automatically adjusting hyperparameters such as the batch size and network depth and width according to the spatial resolution and size of the input image. At the same time, an attention gating module is introduced into the skip connections of the 3D-UNet convolutional neural network, and its basic form can be expressed as:

[0079] 。

[0080] In the formula, and respectively represent the feature tensors on the encoding module and decoding module connected by the skip connection, represents the RELU activation function, represents the Sigmoid activation function, 、 and represent the linear transformation coefficients, and represent the bias coefficients, represents the processed image, represents the SAct image after resampling.

[0081] Finally, by multiplying the SAct image after resampling with the feature tensor on the decoding module, the feature information at different scales is fused, and the background noise is greatly reduced.

[0082] 23. Use the training and testing dataset to train and test the deep neural network model. When the output error of the deep neural network model after training and testing reaches the set threshold, a trained deep neural network model is obtained.

[0083] For example, the deep neural network model is trained and tested by means of five-fold cross-validation. The input of the deep neural network model is the individual-level T1-weighted imaging image and TOF-MRA magnetic resonance data, and the output is the cerebrovascular segmentation image. In order to address the differences in the specificity and sensitivity of the cerebrovascular segmentation results at different scales during the vascular segmentation process, the Tversky function can be introduced as the loss function of the deep neural network model. On the basis of the Tversky function, a weighted Tversky loss method is further proposed, that is, the reciprocal of the Euclidean distance from the voxel to the center line is used as the mismatch loss weight to strengthen the restriction on the topological structure of the segmentation result, so that the cerebrovascular segmentation results have a more consistent performance at each scale. This application defaults to selecting the model with the best result as the final output model (i.e., the trained deep neural network model), and the user can also choose to deploy the average model.

[0084] 24. Use the trained deep neural network model as the cerebrovascular tissue segmentation model, and input the preprocessed image into the cerebrovascular tissue segmentation model to obtain the cerebrovascular tissue segmentation image.

[0085] In another exemplary embodiment of this application, the implementation process of step 103 includes:

[0086] 31. Use a three-dimensional conditional random field model to penalize adjacent points with similar signal intensities on the cerebrovascular tissue segmentation image to obtain a three-dimensional space segmentation image. Adjacent points with similar signal intensities refer to pixel points whose signal intensity difference is within a set range. Among them, the three-dimensional conditional random field model takes into account the spatial continuity of vascular tissue. By establishing pairwise and single-point potential functions, adjacent points with similar signal intensities on the magnetic resonance image with different segmentation labels on the segmentation image are penalized to obtain a more continuous and reasonable segmentation image in three-dimensional space (i.e., the three-dimensional space segmentation image).

[0087] 32. Find the three-dimensional maximum connected component on the three-dimensional space segmentation image, and remove the unconnected small piece of signal noise in the three-dimensional space segmentation image based on the three-dimensional maximum connected component to obtain the finally post-processed cerebrovascular segmentation image.

[0088] In another exemplary embodiment of this application, the implementation process of step 104 includes:

[0089] 41. Centerline, normal plane estimation and three-dimensional reconstruction.

[0090] (1)Perform centerline estimation on the post-processed image. Based on the Voronoi Covariance Measure (VCM), stably extract the centerline of cerebral vessel segmentation to reduce the recognition drift of the center points at the bifurcations of cerebral vessels and the problem of voxel discontinuity.

[0091] (2)On the basis of extracting the centerline of cerebral vessels, use the λ-Maximal Segment Tangent (λ-MST) method to calculate the tangents corresponding to each voxel on the centerline, and then further calculate the normal plane of each centerline voxel. Obtain a robust estimate of the normal plane of each centerline voxel through the method of weighted average of the normal planes of adjacent voxels.

[0092] (3)For each voxel on the centerline, fit the largest circle in its normal plane and calculate the diameter size corresponding to this voxel.

[0093] (4)Combine the centerline and diameter information to achieve the 3D reconstruction of the tubular tree structure. This step can effectively prevent the local tiny errors that may be introduced by the structure segmentation algorithm from being cascaded and amplified in the vascular feature estimation, and thus can ensure the robustness of the 3D reconstruction.

[0094] 42. Extraction of basic features of cerebral vessels: The length and volume of individual tubular structures, the diameter and length of local brain regions.

[0095] 1) Extraction of the length and volume of individual cerebral vessels.

[0096] Count the number of voxels with positive labels in the 3D image of cerebral vessel segmentation (i.e., the image obtained by realizing the 3D reconstruction of the tubular tree structure) to obtain the absolute length of cerebral vessels. At the same time, considering that the size of the individual brain tissue can significantly affect the length of the cerebral vessel structure, divide the absolute length by the size of the individual brain tissue volume to obtain the relative cerebral vessel length.

[0097] Count the number of voxels obtained and multiply by the voxel volume size to obtain the absolute volume of cerebral vessels. At the same time, considering that the size of the individual brain tissue can significantly affect the volume of the cerebral vessel structure, the absolute volume of cerebral vessels can also be divided by the size of the individual brain tissue volume to obtain the relative volume, which reflects the average distribution density of cerebral vessel tissue in the whole brain tissue.

[0098] 2) Extraction of the length and diameter of cerebral vessels in local brain regions.

[0099] Using a two-step registration method, a non-linear registration is performed between the MNI standard brain template and the cerebral vascular magnetic resonance image, and the Harvard-Oxford brain parcellation map in the MNI standard brain template space is transformed into the individual-level magnetic resonance angiography image space. In each brain parcellation, the local vascular length and diameter are calculated according to the above step 1) to reflect the heterogeneity of the basic vascular characteristics in the spatial distribution of each brain region.

[0100] 43. Extraction of cerebral vascular geometric features. The cerebral vascular geometric features mainly include the curvature radius and bending degree of cerebral blood vessels.

[0101] In the cerebral vascular segmentation centerline image, according to the tangent vector flip angle and Frenet frame respectively, the curvature radius and bending degree of each voxel point on the centerline in three-dimensional space are calculated. And the average vascular curvature radius and bending degree in each brain region corresponding to the Harvard-Oxford brain parcellation map are calculated respectively to measure the geometric brain region distribution characteristics of cerebral blood vessels.

[0102] 44. Extraction of cerebral vascular topological features. The cerebral vascular topological features mainly refer to the bifurcation points of cerebral blood vessels.

[0103] Step (1): According to the cerebral vascular segmentation centerline image, the adjacency relationship of each voxel on the centerline is characterized by a tree structure, so as to realize the automatic identification of bifurcation points in the vascular tree structure, and count the number of bifurcations and their spatial distribution.

[0104] Step (2): Calculate the included angle between the tangent vectors of the adjacent voxels of the bifurcation point, which is the bifurcation angle of the bifurcation point.

[0105] Step (3): Calculate the average bifurcation angle and bifurcation density in each brain region to reflect the distribution characteristics of vascular bifurcation statistics in the brain region space.

[0106] 45. Voxel Based Morphometry (VBM).

[0107] Step 1): Iteratively register the cerebral vascular segmentation images of healthy people to obtain the cerebral vascular distribution density map at the group level of healthy people.

[0108] Step 2): Non-linearly register the individual-level cerebral vascular segmentation image with the group cerebral vascular distribution density map through non-linear registration, and obtain the Jacobian matrix of the transformation.

[0109] Step 3): Standardize the registered individual vascular image using the Jacobian matrix to obtain the distribution density of the individual cerebral blood vessels in the standard atlas space. This distribution density can quantitatively evaluate the local or overall level differences between the individual vascular distribution network density and the group.

[0110] In another exemplary embodiment of the present application, since T1-weighted imaging images and T2-FLAIR images mainly reflect the overall characteristics of various tissue structures such as white matter and gray matter in the individual brain, they can reflect the impact of vascular aging on these brain tissue structures and contain rich information on cerebrovascular aging. However, since these two types of images are not specific magnetic resonance cerebrovascular imaging sequences, the conventional two-step strategy of feature extraction after segmentation cannot comprehensively extract the information on brain vascular aging. To solve this problem, the present application utilizes the advantage of neural networks in adaptive feature encoding extraction to automatically extract the vascular aging features in T1-weighted imaging images and T2-FLAIR images. Based on this, in step 105 provided by the present application, during the process of performing convolution feature extraction on T1-weighted imaging images and T2-FLAIR images respectively using the method combining an asymmetric convolution module (Asymmetric Convolution) and a Squeeze-and-Excitation module, the asymmetric convolution module improves the image receptive field of the convolution kernel, and its specific formula is as follows:

[0111] .

[0112] In the formula, is the input feature map, is the output feature map, , , and are convolution kernels of 1×1×k, 1×k×1, k×1×1, and k×k×k respectively.

[0113] The Squeeze-and-Excitation module can introduce an attention mechanism into the neural network and adaptively adjust the weight parameters of each feature according to the correlation between different convolution features. The specific formula is as follows:

[0114] .

[0115] In the formula, is the result output by the Squeeze-and-Excitation module, F is the input feature map, σ is the sigmoid non-linear activation function, ReLU is the RectifiedLieanerUnit non-linear activation function, W 1 is the linear parameter matrix acting on the input feature map compressed by global pooling, b 1 is the bias term acting on the input feature map compressed by global pooling, W 2 is the linear parameter matrix after ReLU non-linear activation, b2 is the corresponding bias term. GlobalAvgPool represents global pooling compression, and its specific formula is:

[0116] .

[0117] This formula means averaging the width and height of the input feature map. In the formula, represents the result of the i-th global pooling compression, H represents the height of the input feature map, W represents the width of the input feature map, is the number of iterations, represents the feature map with height j and width k of the i-th global pooling input.

[0118] Based on the above description, in the actual application process, in the convolutional neural network encoder, the connection between the two asymmetric convolution modules and the Squeeze-and-Excitation module can be set to obtain a new convolutional neural network encoder, so as to realize the extraction of the convolutional features of T1-weighted imaging images and T2-FLAIR images.

[0119] Furthermore, through global pooling compression, the convolutional features of the T1-weighted imaging images and T2-FLAIR images extracted are compressed to generate unified T1-weighted imaging images and T2-FLAIR image convolutional feature groups with each feature dimension being a set value (assumed to be 1). Then, demographic indicators such as gender and cerebrovascular imaging features are directly connected to the T1-weighted imaging images and T2-FLAIR image convolutional feature groups to generate a cerebrovascular aging imaging characterization group that specifically reflects cerebrovascular features and a comprehensive feature group that reflects the overall features of cerebrovascular aging and contains two parts of information, namely, the T1-weighted imaging images and T2-FLAIR image imaging characterization groups, to obtain a comprehensive imaging feature group.

[0120] A simple fully connected neural network architecture with a single hidden layer is used to construct a biological age prediction model. The input of this constructed model is the comprehensive imaging feature group, and the comprehensive imaging feature group is used to comprehensively predict the brain aging age of patients. By comparing with the individual physical age, an accurate estimation of the brain aging index is realized.

[0121] Among them, the loss function of the biological age prediction model uses the MSE loss function, which is expressed as:

[0122] .

[0123] In the formula, represents the MSE loss function value, represents the model's estimation of the vascular age ， represents the actual physical age of the patient, represents the total sample size.

[0124] Furthermore, during the process of constructing the biological age prediction model, training and testing are also required. Based on this, the dataset is randomly divided into a training set and a testing set. In the training set, multi-model ten-fold cross-validation training is carried out, and the optimal model is selected as the final output model (i.e., the biological age prediction model). It is verified in the testing set. To prevent overfitting, the method of early stopping is used in training to fully train the initial model, greatly reducing the risk of model overfitting. The output of the final model is the biological age of the cerebrovascular, and the difference between it and the individual's physical age is the comprehensive brain aging index proposed in this application, achieving the effect of quantitatively and objectively measuring the degree of brain aging in patients' imaging using multi-modal magnetic resonance data. At the same time, the effect coefficient corresponding to each imaging feature can also be obtained in the biological age prediction model. This coefficient can quantitatively evaluate the contribution degree of the imaging feature to the overall brain aging index, indicating the imaging feature that has the greatest impact on vascular aging, and has extremely strong clinical guiding significance.

[0125] In addition to performing vascular segmentation and extraction of vascular features, this application also adds brain structure features. By fusing the vascular aging features in T1-weighted imaging images and T2-FLAIR images, combined with advanced neural network technology, quantitative evaluation of brain aging is achieved. Based on this, compared with the prior art, the main features of this application are as follows:

[0126] 1. Multi-modal image fusion.

[0127] (1-1) Joint use of T1-weighted imaging images and T2-FLAIR images: T1-weighted imaging images and T2-FLAIR images respectively reflect different characteristics of brain tissue. T1-weighted images mainly show anatomical structures, while T2-FLAIR images are more sensitive to pathological changes such as white matter lesions and edema in the brain. The combination of these two modalities can provide more abundant brain aging information, including the manifestation of abnormal brain tissue structure on cerebrovascular aging, and is more comprehensive in reflecting cerebrovascular aging.

[0128] (1-2) Adaptive feature extraction: Traditional post-segmentation feature extraction methods cannot comprehensively capture the vascular aging features in T1-weighted imaging images and T2-FLAIR images. This application utilizes the adaptive feature encoding and extraction ability of the neural network to automatically identify and extract complex features related to vascular aging, avoiding the limitations of manually designed features.

[0129] 2. Optimization of the neural network model structure.

[0130] (2-1) Asymmetric Convolution Module: To better capture local and global features in T1-weighted imaging and T2-FLAIR images, this application introduces an asymmetric convolution module that can extract features at different scales and enhance the model's sensitivity to vascular aging features.

[0131] (2-2) Squeeze-and-Excitation Mechanism: By introducing the Squeeze-and-Excitation (SE) mechanism, it can dynamically adjust the weights in the feature map, highlight important regions, and suppress irrelevant information. This helps improve the recognition accuracy of vascular aging features, especially when dealing with complex cerebrovascular structures.

[0132] (2-3) Global Pooling Compression: To effectively fuse the convolutional features of T1-weighted imaging and T2-FLAIR images, global pooling compression technology is adopted to generate a unified feature group with a feature dimension of 1 for each dimension. This process not only reduces the feature dimension but also retains key information for subsequent comprehensive analysis.

[0133] 3. Construction of Comprehensive Feature Group.

[0134] Multi-source Information Fusion: Demographic indicators such as gender and age are combined with cerebrovascular imaging features to generate an imaging characterization group that specifically reflects cerebrovascular features. At the same time, the convolutional feature groups of T1-weighted imaging and T2-FLAIR images are also incorporated to form a comprehensive feature group. This comprehensive feature group not only contains the features of the blood vessels themselves but also reflects the impact of vascular aging on brain tissue structure, providing a more comprehensive evaluation perspective.

[0135] 4. Prediction of Vascular Biological Age.

[0136] (4-1) Single Hidden Layer Fully Connected Neural Network: In the final vascular biological age prediction model, a simple single hidden layer fully connected neural network architecture is adopted to predict the cerebrovascular aging age of patients using the comprehensive imaging feature group. By comparing with the individual's physical age, an accurate estimate of the brain aging index is obtained.

[0137] (4-2) Ten-fold Cross-validation and Early Stopping Strategy: To prevent model overfitting, ten-fold cross-validation and the early stopping strategy (Early Stopping) are adopted during training. This not only ensures the generalization ability of the model but also greatly reduces the risk of overfitting, improving the stability and reliability of the model.

[0138] In summary, the method for determining the comprehensive brain aging index proposed in this application has the following significant advantages:

[0139] 1. Quantitative objective assessment: Through the fusion of multimodal imaging data and the adaptive feature extraction of neural networks, it is possible to quantitatively and objectively measure the degree of cerebrovascular aging in patients' imaging, providing a reliable assessment tool for clinicians.

[0140] 2. Effect coefficient analysis: The effect coefficient corresponding to each imaging feature output can quantitatively evaluate its contribution to the overall cerebrovascular aging index, revealing the imaging features that have the greatest impact on vascular aging. This information has strong clinical guiding significance, enabling doctors to better understand the mechanism of vascular aging and providing a basis for personalized treatment.

[0141] 3. Early screening and risk assessment: Cerebrovascular aging is an important risk factor for diseases such as stroke and cognitive impairment. Through the method of this application, potential vascular problems can be detected at an early stage, and intervention measures can be taken in a timely manner to reduce the incidence risk of patients.

[0142] In addition, in this application, T1-weighted imaging images and T2-FLAIR images respectively reflect different characteristics of brain tissue. T1-weighted images mainly show anatomical structures, while T2-FLAIR images are more sensitive to pathological changes such as white matter lesions and edema in the brain. The combination of these two modalities can not only provide more abundant information on vascular aging, but also capture structural features that have not been fully emphasized in other modalities. These features are actually another manifestation of vascular aging and can reflect the state of earlier microvascular aging.

[0143] Microvascular aging is an important precursor of cerebrovascular diseases. Especially in the early stage, changes in microvessels may be the earliest pathological features. Therefore, by fusing the features in T1-weighted imaging images and T2-FLAIR images, signs of microvascular aging can be identified earlier, enhancing the assessment of microvascular aging, and thus more comprehensively reflecting the overall aging state of cerebrovasculature.

[0144] Traditional feature extraction methods rely on manually designed algorithms, are easily affected by subjective factors, and are difficult to fully utilize the complex information in multimodal imaging. This application uses the adaptive feature encoding and extraction ability of neural networks to automatically identify and extract complex features related to vascular aging, avoiding the limitations of manually designed features. This automated feature extraction method not only improves the accuracy of the model, but also enhances its generalization ability and is applicable to different types of imaging data.

[0145] The brain aging index not only reflects the structural changes of large blood vessels, but also captures the aging characteristics of microvessels, providing a more comprehensive evaluation perspective. By comparing with the individual's physical age (i.e., the age obtained in step 107 above), the brain aging index can accurately estimate the cerebrovascular biological age of patients and reveal the potential risks of vascular aging.

[0146] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 2 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store brain aging comprehensive index determination data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for determining a brain aging comprehensive index.

[0147] Those skilled in the art can understand that Figure 2 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0148] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0149] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0150] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0151] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0152] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0153] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0154] In this article, specific examples are used to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present application; at the same time, for those of ordinary skill in the art, according to the ideas of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for determining a comprehensive index of brain aging, characterized in that: include: Acquire T1-weighted imaging images, T2-FLAIR images, and TOF-MRA magnetic resonance images; Preprocessing the T1-weighted imaging image and the TOF-MRA magnetic resonance image to obtain a preprocessed image; Obtaining a cerebral vascular tissue segmentation image based on the preprocessed image; Post-processing the cerebral vascular tissue segmentation image based on the TOF-MRA magnetic resonance image to obtain a post-processed image; Extract cerebrovascular imaging features based on post-processed images; The convolution feature extraction of T1-weighted imaging image and T2-FLAIR image was performed by combining asymmetric convolution module and Squeeze-and-Excitation module to obtain T1-weighted imaging feature and T2-FLAIR image feature. By globally pooling and compressing T1-weighted imaging features and T2-FLAIR image features, a convolution feature group of T1-weighted imaging images and T2-FLAIR images is generated, each feature dimension of which is a set value; Obtain demographic indicators; demographic indicators include gender and age; The demographic indicators and cerebrovascular imaging features were connected with the convolutional feature groups of T1-weighted imaging images and T2-FLAIR images to generate a cerebrovascular aging imaging representation group; Construct a biological age prediction model; The cerebrovascular aging image characterization group is input into the biological age prediction model to obtain the brain aging age prediction result; The comprehensive index of brain aging was obtained based on the predicted results of brain aging age and the actual age of the evaluated subjects.

2. The method for determining the comprehensive index of brain aging according to claim 1, characterized in that: The cerebral vascular tissue segmentation image is obtained based on the preprocessed image, including: Construct a training and testing data set; the training and testing data set includes paired data of T1-weighted imaging images and TOF-MRA magnetic resonance images; Based on the 3D-UNet convolutional neural network, the attention gating module is combined to build a deep neural network model; The deep neural network model is trained and tested using the training and testing data set until an output error of the deep neural network model after training and testing reaches a set threshold, thereby obtaining a trained deep neural network model; The trained deep neural network model is used as a cerebrovascular tissue segmentation model, and the preprocessed image is input into the cerebrovascular tissue segmentation model to obtain the cerebrovascular tissue segmentation image.

3. The method for determining the comprehensive index of brain aging according to claim 2, characterized in that: Based on the Tversky function, the weighted Tversky loss is used to construct the loss function of the cerebrovascular tissue segmentation model.

4. The method for determining the comprehensive index of brain aging according to claim 1, characterized in that: Preprocess the T1-weighted imaging image and the TOF-MRA magnetic resonance image to obtain a preprocessed image, including: Perform brain tissue-specific identification and segmentation on TOF-MRA magnetic resonance images to obtain brain tissue structure images; Using a linear rigid body transformation and spline interpolation method, based on the brain tissue structure image, the T1-weighted imaging image and the TOF-MRA magnetic resonance image are registered, and the registered structure is resampled to the same image space to obtain a converted image; Adaptive non-local means algorithm is used to perform image denoising on TOF-MRA magnetic resonance images to obtain denoised images. Using a signal intensity remapping method after Gaussian deconvolution, magnetic resonance image artifacts caused by B0 field inhomogeneity in the denoised image are removed to obtain an intermediate image; Processing the intermediate image using a contrast-limited adaptive histogram equalization method to obtain a first signal-enhanced image; Using the Sobel convolution method, identifying edge positions where a sudden change in signal intensity occurs in the first signal-enhanced image, and obtaining a second signal-enhanced image based on the identified edge positions; The second signal enhanced image is processed by a gamma correction method to obtain the preprocessed image.

5. The method for determining the comprehensive index of brain aging according to claim 1, characterized in that: Post-processing the cerebral vascular tissue segmentation image to obtain a post-processed image includes: A three-dimensional conditional random field model is used to perform penalty processing on adjacent points with similar signal strength on the cerebrovascular tissue segmentation image to obtain a three-dimensional spatial segmentation image; the adjacent points with similar signal strength refer to pixel points whose signal strength difference is within a set range; The three-dimensional maximum connected component is found on the three-dimensional space segmentation image, and the unconnected signal noise in the three-dimensional space segmentation image is removed based on the three-dimensional maximum connected component to obtain the post-processing image.

6. The method for determining the comprehensive index of brain aging according to claim 1, characterized in that: The convolution feature extraction is performed on the T1-weighted imaging image and the T2-FLAIR image respectively by using a method combining an asymmetric convolution module and a Squeeze-and-Excitation module to obtain T1-weighted imaging features and T2-FLAIR image features, including: In the convolutional neural network encoder, two asymmetric convolution modules are connected to the Squeeze-and-Excitation module to obtain a new convolutional neural network encoder; A new convolutional neural network encoder is used to perform convolution feature extraction on the T1 weighted imaging image and the T2-FLAIR image respectively to obtain the T1 weighted imaging features and the T2-FLAIR image features.

7. The method for determining the comprehensive index of brain aging according to claim 1, characterized in that: The biological age prediction model is constructed using a single hidden layer fully connected neural network architecture.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for determining a comprehensive brain aging index according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for determining a comprehensive brain aging index according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for determining a comprehensive brain aging index according to any one of claims 1 to 7 is implemented.

Images