Deep learning brain age testing method and system based on t1wi and qsm
By combining deep learning algorithms with magnetic resonance T1WI and QSM images, brain nuclei are segmented and features are extracted. A linear regression model is established, which solves the accuracy problem of brain age testing in existing technologies and achieves a more accurate assessment of brain aging.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN FIRST CENT HOSPITAL
- Filing Date
- 2023-08-02
- Publication Date
- 2026-06-12
AI Technical Summary
Existing brain age testing methods lack accuracy and are difficult to comprehensively assess the aging state of the brain, thus affecting the accuracy of brain health assessments.
By combining T1WI and QSM images from magnetic resonance imaging, brain nuclei are segmented using deep learning algorithms. Numerical shape features and statistical features of magnetic susceptibility are extracted, and a linear regression model is established to test brain age.
It enables more comprehensive and accurate brain age testing, improving the precision of brain aging assessment.
Smart Images
Figure CN116982935B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image analysis, and in particular to a deep learning-based brain age testing method and system based on T1WI and QSM. Background Technology
[0002] As we age, our brains also age, which can manifest as a decline in information processing abilities, memory, and calculation skills. Brain age is an important indicator of brain health. In healthy individuals, brain age is close to or even younger than chronological age. However, when the body is diseased, the brain ages rapidly, resulting in a brain age that exceeds chronological age. Therefore, brain age testing is crucial for assessing brain health. Summary of the Invention
[0003] T1-weighted images (T1WI) from magnetic resonance imaging (MRI) highlight longitudinal relaxation differences in tissues, minimizing the influence of other tissue characteristics, such as lateral relaxation, on the image. Quantitative Susceptibility Mapping (QSM) images fully utilize phase information and can quantify tissue magnetic susceptibility, highlighting the outline and internal details of iron-containing tissue structures. Combining the features of these two types of images to build a brain age testing model allows for more comprehensive and accurate brain age assessment. Therefore, this invention provides a deep learning-based brain age testing method and system based on T1WI and QSM.
[0004] To achieve the above objectives, the present invention provides the following solution:
[0005] In a first aspect, the present invention provides a deep learning-based brain age testing method based on T1WI and QSM, comprising:
[0006] Acquire T1WI and QSM image data from the test subjects;
[0007] Based on deep learning algorithms and the test subject's magnetic resonance T1WI data and QSM image data, a three-dimensional segmentation map of the test subject's brain nuclei was determined;
[0008] Based on the three-dimensional segmentation map of the test subject's brain nuclei, the statistical characteristics of the corresponding numerical shape features and magnetic susceptibility values of the test subject are calculated.
[0009] The brain age of the tester is calculated based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the tester. The brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.
[0010] Optionally, before performing the step of determining the three-dimensional segmentation map of the test subject's brain nuclei based on the deep learning algorithm and the test subject's magnetic resonance T1WI data and QSM image data, the method further includes:
[0011] The magnetic resonance T1WI data and QSM image data of the test subject were registered and preprocessed to obtain the corresponding processed magnetic resonance T1WI data and QSM image data of the test subject.
[0012] Optionally, based on deep learning algorithms and the test subject's magnetic resonance T1WI and QSM image data, a three-dimensional segmentation map of the test subject's brain nuclei is determined, specifically including:
[0013] The processed magnetic resonance T1WI data and QSM image data of the test subject are input into the image segmentation model to obtain a three-dimensional segmentation map of the test subject's brain nuclei;
[0014] The image segmentation model was obtained by training the EFAU-Net neural network based on the corresponding magnetic resonance T1WI data, QSM image data, and three-dimensional segmentation map of brain nuclei of healthy individuals.
[0015] Optionally, the three-dimensional segmentation map of the brain nuclei includes multiple segmentation regions, namely the caudate nucleus region, the globus pallidus region, the putamen region, the thalamus region, and the background region.
[0016] Optionally, the training process of the image segmentation model is as follows:
[0017] Construct a sample dataset; the sample dataset includes multiple sample data; the sample data includes input data and corresponding label data; the input data is processed magnetic resonance T1WI data and QSM image data corresponding to healthy individuals, and the label data is a three-dimensional segmentation map of brain nuclei corresponding to healthy individuals;
[0018] Construct the EFAU-Net neural network;
[0019] The EFAU-Net neural network is trained using a sample dataset to make the loss value of the EFAU-Net neural network less than a set value, thus obtaining a trained EFAU-Net neural network; the trained EFAU-Net neural network is an image segmentation model.
[0020] Optionally, the calculation process for numerical shape features is as follows:
[0021] Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei:
[0022] Based on the segmented regions, calculate the mesh volume, voxel volume, surface area, surface area to mesh volume ratio, sphericity, maximum 3D diameter, maximum 2D diameter in the axial plane, maximum 2D diameter in the coronal plane, maximum 2D diameter in the sagittal plane, first principal axis length, second principal axis length, minimum principal axis length, elongation, and flatness.
[0023] Optionally, the calculation process for the statistical characteristics of the magnetic susceptibility value is as follows:
[0024] Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei:
[0025] Calculate the average value of the magnetic susceptibility based on the segmented regions;
[0026] Calculate the standard deviation of the magnetic susceptibility value based on the segmented regions.
[0027] Secondly, the present invention provides a deep learning-based brain age testing system based on T1WI and QSM, comprising:
[0028] The data acquisition module is used to acquire the test subject's magnetic resonance T1WI data and QSM image data;
[0029] The brain nucleus 3D segmentation map determination module is used to determine the 3D segmentation map of the brain nucleus of the test subject based on the deep learning algorithm and the test subject's magnetic resonance T1WI data and QSM image data.
[0030] The feature calculation module is used to calculate the statistical characteristics of the numerical shape features and magnetic susceptibility values of the test subject based on the three-dimensional segmentation map of the test subject's brain nuclei.
[0031] The brain age calculation module is used to calculate the brain age of the test subject based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the test subject. The brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.
[0032] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0033] This invention segments and tests brain age using magnetic resonance T1WI data and QSM image data. It achieves automated segmentation through deep learning algorithms and extracts multiple features from the two types of images to establish a regression model, thereby achieving a more comprehensive and accurate brain age test. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0035] Figure 1 A flowchart illustrating the deep learning-based brain age testing method based on T1WI and QSM provided in this embodiment of the invention;
[0036] Figure 2 This is a schematic diagram of the EFAU-Net neural network structure provided in an embodiment of the present invention;
[0037] Figure 3 This is a schematic diagram of the structure of the EFA module provided in an embodiment of the present invention. Detailed Implementation
[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0040] Example 1
[0041] This embodiment provides a deep learning-based brain age testing method based on T1WI and QSM. This embodiment utilizes deep learning-based medical image segmentation technology to automate the segmentation of magnetic resonance T1WI and QSM images, establishes a linear regression model between several feature values and age, and finally performs brain age testing based on the linear regression model.
[0042] like Figure 1 As shown, the deep learning-based brain age testing method based on T1WI and QSM provided in this embodiment includes:
[0043] Step 101: Acquire the test subject's magnetic resonance T1WI data and QSM image data.
[0044] Step 102: Based on the deep learning algorithm and the test subject's MRI T1WI data and QSM image data, determine the three-dimensional segmentation map of the test subject's brain nuclei. The three-dimensional segmentation map of the brain nuclei includes multiple segmented regions, namely the caudate nucleus region, the globus pallidus region, the putamen region, the thalamus region, and the background region.
[0045] Step 103: Based on the three-dimensional segmentation map of the test subject's brain nuclei, calculate the statistical characteristics of the corresponding numerical shape features and magnetic susceptibility values.
[0046] Step 104: Calculate the brain age of the test subject based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the test subject; wherein, the brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.
[0047] In a preferred embodiment, before performing step 102, the method further includes:
[0048] The magnetic resonance T1WI data and QSM image data of the test subject were registered and preprocessed to obtain the corresponding processed magnetic resonance T1WI data and QSM image data of the test subject.
[0049] In this embodiment, step 102 specifically includes:
[0050] The processed T1WI and QSM images of the test subjects were input into the image segmentation model to obtain a three-dimensional segmentation map of the brain nuclei of the test subjects. The image segmentation model was obtained by training the EFAU-Net neural network based on the T1WI, QSM images, and three-dimensional segmentation map of the brain nuclei of healthy individuals.
[0051] Furthermore, the training process of the image segmentation model is as follows:
[0052] (1) Construct a sample dataset; the sample dataset includes multiple sample data; the sample data includes input data and corresponding label data; the input data is the processed magnetic resonance T1WI data and QSM image data corresponding to healthy individuals, and the label data is the three-dimensional segmentation map of brain nuclei corresponding to healthy individuals.
[0053] (2) Construct the EFAU-Net neural network.
[0054] like Figure 2 As shown, the encoder and decoder in the EFAU-Net neural network are arranged symmetrically in a 'U' shape. The number of channels in the convolutional layers of the encoder are 32, 64, 128, 256, and 256, respectively. The decoder performs feature fusion and then restores the size through upsampling. Specifically:
[0055] The encoder performs four downsampling operations using max pooling, and the decoder performs four upsampling operations to restore the size. At each stage, skip connections are used to concatenate the corresponding encoding layer features with the upsampled decoding layer features. Furthermore, an efficient frequency attention (EFA) module is introduced into the skip connections to adaptively adjust the importance of different frequency feature components, allowing the network to focus on the most important information in the feature map. The EFA module is specifically described as follows: Figure 3 As shown.
[0056] (3) The EFAU-Net neural network is trained using a sample dataset so that the loss value of the EFAU-Net neural network is less than a set value, and the trained EFAU-Net neural network is obtained; the trained EFAU-Net neural network is an image segmentation model.
[0057] Furthermore, the sample data construction process is as follows:
[0058] Step S1: Register the magnetic resonance T1WI data and QSM image data of N healthy individuals, where N is a large value. The registration method is a rigid affine transformation based on mutual information.
[0059] Step S2: Preprocess the registered magnetic resonance T1WI data and QSM image data, and perform data augmentation on the preprocessed magnetic resonance T1WI data and QSM image data; the preprocessing includes resampling, z-score normalization, etc.; the data augmentation includes random scaling, random rotation, random flipping and random Gaussian smoothing.
[0060] Step S3: Match the magnetic resonance T1WI data and QSM image data obtained after step S2 with the corresponding three-dimensional segmentation map of brain nuclei of healthy individuals to obtain sample data.
[0061] Furthermore, the training process of the EFAU-Net neural network is as follows:
[0062] Step A: Transform the input features to the frequency domain using a three-dimensional DCT transform.
[0063] The conversion formula is:
[0064]
[0065] Among them, E i X represents the feature of the i-th layer encoder. i E represents i The frequency domain feature map after DCT transformation, H is E i The height, W is E iThe width, D is E i The lengths of E, h, w, and d are... i The spatial coordinates are t, v, and u, which are all generalized frequency variables. The constant normalization term is omitted here.
[0066] Step B: Perform channel-based global average pooling.
[0067] The pooling formula is: X i = agg_channel(X) i );
[0068] Where agg_channel represents a one-dimensional average pooling with a kernel size of c, where c is the number of channels, i.e., the total number of frequency features in the feature map.
[0069] Step C: Perform a 3D convolution operation with a kernel size of k. The specific formula is as follows:
[0070] A i =σ(w i X′ i );
[0071] Among them, A i The weights between the different frequencies are the final values, where w1 represents the weight parameters of the 3D convolution, and σ is the activation function Sigmoid.
[0072] Step D: Multiply the weights by the frequency domain feature map, using the following formula:
[0073] Y i =X i ⊙A i ;
[0074] Among them, A i As the weight, Y i This represents the frequency domain feature map after frequency feature calibration, and ⊙ represents the dot product operation.
[0075] Step E: After three-dimensional IDCT transformation, the frequency domain feature map is restored to the spatial domain.
[0076] The IDCT formula is:
[0077]
[0078] Among them, E i efa It is E i The feature map is output after efficient frequency domain attention.
[0079] The loss function of the EFAU-Net neural network consists of the Dice loss function and the CE loss function. The formula for the loss function is:
[0080]
[0081] Among them, L Dice L is the Dice loss function for the true label and the predicted label. CE is the cross-entropy loss function between the true label and the predicted label; N is the number of samples; p is the true value; q is the predicted value; C is the number of detection categories. In this embodiment, the detection categories C = 5, which are: caudate nucleus, globus pallidus, putamen, thalamus, and background.
[0082] During training, the optimizer chosen is SGD, with an initial learning rate of 0.01. The learning rate decreases continuously with each training epoch, and is dynamically adjusted according to the following formula:
[0083]
[0084] Where lr represents the learning rate adjusted in each round of training, lr init This indicates the initial learning rate. `epoch` represents the current training iteration. max This indicates the maximum number of training iterations, and `power` is used to control the shape of the learning rate adjustment curve.
[0085] Preferably, lr init =0.01, epoch max =1000, power=0.9.
[0086] Step S4: Use the neural network trained in step S3 to segment the T1WI and QSM processed in step S2 to generate a three-dimensional segmentation map of brain nuclei.
[0087] Image segmentation is performed using a trained EFAU-Net neural network. Specifically, the encoder extracts features using convolutional layers and then performs downsampling using max pooling to expand the receptive field. The decoder performs feature fusion and then performs upsampling to restore the image size. Skip connections fuse information from shallow and deep layers. An EFA module is introduced into the skip connections to adaptively adjust the importance of different frequency features, allowing the network to focus more on the more important frequency feature components.
[0088] In this embodiment, the calculation process for the numerical shape features is as follows:
[0089] Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei:
[0090] S11. Calculate the mesh volume based on the segmented region ROI; the mesh volume is calculated from the triangular mesh of the segmented region ROI, and the specific calculation formula is as follows:
[0091]
[0092]
[0093] For each face i, a triangle is defined, with the three vertices a. i b i c i The origin of the image is O, and V is V. i N is the mesh volume corresponding to surface i. f Indicates the number of faces.
[0094] S12. Calculate the voxel volume based on the segmented region of interest (ROI). The voxel volume is obtained by multiplying the number of voxels in the segmented ROI by the volume of a single voxel. The specific calculation formula is as follows:
[0095]
[0096] Among them, V k N represents the calculated voxel volume of surface k. v Indicates the number of faces.
[0097] S13. Calculate the surface area based on the segmented Region of Interest (ROI). The surface area is calculated from the triangular mesh of the segmented ROI, and the specific calculation formula is as follows:
[0098]
[0099]
[0100] For each face i, a triangle is defined, with the three vertices a. i b i c i The origin of the image is O, A i N is the mesh volume corresponding to surface i. f Indicates the number of faces.
[0101] S14. Calculate the ratio of surface area to mesh volume based on the segmented ROI. The specific calculation formula is as follows:
[0102]
[0103] Where A is the surface area of the segmented region ROI, and V is the mesh volume of the segmented region ROI.
[0104] S15. Calculate sphericity based on the segmented Region of Interest (ROI). Sphericity is a dimensionless measure of the roundness of the segmented ROI relative to a sphere, independent of direction and scale, and ranges from [0, 1]. A value closer to 1 represents a closer approximation to a perfect sphere. The calculation formula is:
[0105]
[0106] Where A is the surface area of the segmented region ROI, and V is the mesh volume of the segmented region ROI.
[0107] S16. Obtain the maximum three-dimensional diameter based on the segmented region ROI; the maximum three-dimensional diameter represents the maximum pairwise Euclidean distance obtained at the vertices of the mesh on the surface of the segmented region ROI.
[0108] S17. Obtain the maximum two-dimensional diameter of the axial plane based on the segmented region ROI; the maximum two-dimensional diameter of the axial plane represents the maximum paired Euclidean distance between the vertices of the mesh on the surface of the segmented region ROI in the axial plane.
[0109] S18. Obtain the maximum two-dimensional diameter of the coronal plane based on the segmented region ROI; the maximum two-dimensional diameter of the coronal plane represents the maximum pairwise Euclidean distance between the vertices of the mesh on the surface of the segmented region ROI in the coronal plane.
[0110] S19. Obtain the maximum two-dimensional diameter of the sagittal plane based on the segmented region ROI; the maximum two-dimensional diameter of the sagittal plane represents the maximum pairwise Euclidean distance between the vertices of the mesh on the surface of the segmented region ROI in the sagittal plane.
[0111] S10. Obtain the first principal axis length based on the segmented region ROI; the first principal axis length is the maximum axis length of the ellipsoid enclosing the segmented region ROI, and the specific calculation formula is as follows:
[0112]
[0113] Where, λ major This represents the length of the maximum principal component axis obtained by the ellipsoid enclosing the segmented region ROI.
[0114] S21. Obtain the second principal axis length based on the segmented region ROI; the second principal axis length is the length of the second axis of the ellipsoid enclosing the segmented region ROI, and the specific calculation formula is as follows:
[0115]
[0116] Where, λ minor This represents the length of the second principal component axis obtained by the ellipsoid enclosing the segmented region ROI.
[0117] S22. Obtain the minimum principal axis length based on the segmented region ROI; the minimum axis length is the minimum axis length of the ellipsoid enclosing the segmented region ROI, and the specific calculation formula is as follows:
[0118]
[0119] Where, λ leastThis represents the length of the minimum principal component axis obtained by the ellipsoid enclosing the segmented region ROI.
[0120] S23. Calculate the elongation rate based on the segmented ROI; the elongation rate represents the relationship between the two largest principal components in the shape of the segmented ROI, and the specific calculation formula is as follows:
[0121]
[0122] Where, λ minor λ represents the length of the second principal component axis obtained by the ellipsoid enclosing the region of interest (ROI). major This represents the length of the maximum principal component axis obtained by the ellipsoid enclosing the segmented region ROI.
[0123] S24. Calculate the flatness based on the segmented ROI; flatness represents the relationship between the largest and smallest principal components in the shape of the segmented ROI, and the specific calculation formula is as follows:
[0124]
[0125] Where, λ least λ represents the length of the minimum principal component axis obtained by the ellipsoid enclosing the region of interest (ROI). major This represents the length of the maximum principal component axis obtained by the ellipsoid enclosing the segmented region ROI.
[0126] In this embodiment, the calculation process of the statistical characteristics of the magnetic susceptibility value is as follows:
[0127] Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei.
[0128] (1) Calculate the average value of the magnetic susceptibility based on the segmented region ROI. The calculation formula is as follows:
[0129]
[0130] Where i represents a pixel in the segmented region ROI, X(i) is the grayscale intensity value corresponding to pixel i, and there are a total of N pixels. p indivual.
[0131] (2) Calculate the standard deviation of the magnetic susceptibility value based on the segmented region ROI. The calculation formula is as follows:
[0132]
[0133] Where i is a pixel in the segmented region ROI, and X(i) is the gray value corresponding to pixel i. The average grayscale value of the segmented region ROI is N, and there are a total of N pixels. p indivual.
[0134] Example 2
[0135] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a deep learning brain age testing system based on T1WI and QSM is provided below.
[0136] This embodiment provides a deep learning-based brain age testing system based on T1WI and QSM, including:
[0137] The data acquisition module is used to acquire the test subject's magnetic resonance T1WI data and QSM image data.
[0138] The brain nucleus 3D segmentation map determination module is used to determine the 3D segmentation map of the test subject's brain nucleus based on deep learning algorithms and the test subject's magnetic resonance T1WI data and QSM image data.
[0139] The feature calculation module is used to calculate the statistical characteristics of the numerical shape features and magnetic susceptibility values of the test subject based on the three-dimensional segmentation map of the test subject's brain nuclei.
[0140] The brain age calculation module is used to calculate the brain age of the test subject based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the test subject. The brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.
[0141] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.
[0142] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A deep learning brain age testing method based on T1WI and QSM, characterized in that, include: Acquire T1WI and QSM image data from the test subjects; Registration and preprocessing were performed on the T1WI and QSM image data of the test subject to obtain the corresponding processed T1WI and QSM image data of the test subject. Based on deep learning algorithms and the test subject's T1WI and QSM image data, a three-dimensional segmentation map of the test subject's brain nuclei was determined, specifically including: The processed magnetic resonance T1WI data and QSM image data of the test subject are input into the image segmentation model to obtain a three-dimensional segmentation map of the test subject's brain nuclei; Among them, the image segmentation model was obtained by training the EFAU-Net neural network based on the magnetic resonance T1WI data, QSM image data and three-dimensional segmentation map of brain nuclei of healthy individuals; Based on the three-dimensional segmentation map of the test subject's brain nuclei, the statistical characteristics of the corresponding numerical shape features and magnetic susceptibility values of the test subject are calculated. The brain age of the tester is calculated based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the tester. The brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.
2. The deep learning-based brain age testing method based on T1WI and QSM according to claim 1, characterized in that, The three-dimensional segmentation map of the brain nuclei includes multiple segmented regions, namely the caudate nucleus region, the globus pallidus region, the putamen region, the thalamus region, and the background region.
3. The deep learning-based brain age testing method based on T1WI and QSM according to claim 1, characterized in that, The training process of the image segmentation model is as follows: Construct a sample dataset; the sample dataset includes multiple sample data; the sample data includes input data and corresponding label data; the input data is processed magnetic resonance T1WI data and QSM image data corresponding to healthy individuals, and the label data is a three-dimensional segmentation map of brain nuclei corresponding to healthy individuals; Construct the EFAU-Net neural network; The EFAU-Net neural network is trained using a sample dataset to make its loss value less than a set value, thus obtaining a trained EFAU-Net neural network; the trained EFAU-Net neural network is an image segmentation model.
4. The deep learning-based brain age testing method based on T1WI and QSM according to claim 2, characterized in that, The calculation process for numerical shape features is as follows: Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei: Based on the segmented regions, calculate the mesh volume, voxel volume, surface area, surface area to mesh volume ratio, sphericity, maximum 3D diameter, maximum 2D diameter in the axial plane, maximum 2D diameter in the coronal plane, maximum 2D diameter in the sagittal plane, first principal axis length, second principal axis length, minimum principal axis length, elongation, and flatness.
5. The deep learning-based brain age testing method based on T1WI and QSM according to claim 2, characterized in that, The calculation process for the statistical characteristics of magnetic susceptibility values is as follows: Perform the following operations on any segmented region in the three-dimensional segmentation map of the brain nuclei: Calculate the average value of the magnetic susceptibility based on the segmented regions; Calculate the standard deviation of the magnetic susceptibility value based on the segmented regions.
6. A deep learning-based brain age testing system based on T1WI and QSM, characterized in that, include: The data acquisition module is used to acquire the test subject's magnetic resonance T1WI data and QSM image data; The brain nucleus 3D segmentation map determination module is used to perform registration and preprocessing operations on the test subject's magnetic resonance T1WI data and QSM image data to obtain the corresponding processed magnetic resonance T1WI data and QSM image data of the test subject. It is also used to determine the three-dimensional segmentation map of the brain nuclei of the test subject based on deep learning algorithms and the test subject's magnetic resonance T1WI data and QSM image data, specifically including: The processed magnetic resonance T1WI data and QSM image data of the test subject are input into the image segmentation model to obtain a three-dimensional segmentation map of the test subject's brain nuclei; Among them, the image segmentation model was obtained by training the EFAU-Net neural network based on the magnetic resonance T1WI data, QSM image data and three-dimensional segmentation map of brain nuclei of healthy individuals; The feature calculation module is used to calculate the statistical characteristics of the numerical shape features and magnetic susceptibility values of the test subject based on the three-dimensional segmentation map of the test subject's brain nuclei. The brain age calculation module is used to calculate the brain age of the test subject based on the brain age test model and the statistical characteristics of the numerical shape characteristics and magnetic susceptibility values of the test subject. The brain age test model is obtained by linear regression of the statistical characteristics of the age, numerical shape characteristics and magnetic susceptibility values of healthy individuals.