A disease-specific human brain aging assessment method and system

Abstract

The present invention is suitable for the field of comprehensive assessment technology improvement, and provides a disease-specific human brain aging assessment method, including: S1, extracting age-related brain tissue volume features from the T1-weighted magnetic resonance brain images of the subject; S2, using the trained XGBoost regression model to estimate the subject's brain age according to the input brain tissue volume features; S3, performing SHAP feature attribution on the subject's brain age prediction results, quantifying the contribution of each brain tissue to brain aging, and expanding the single value of brain age into a feature contribution vector reflecting regional aging; S4, using the XGBoost classifier to evaluate the risk probability of the subject suffering from neurodegenerative diseases according to the input brain age vector. By performing feature attribution on brain age prediction, the contribution of each brain tissue to human brain aging is quantified, the regional aging of the brain is refined, and the spatial specificity for different brain diseases is improved.

Description

Technical Field

[0001] The present invention belongs to the field of comprehensive assessment technology improvement, and in particular relates to a disease-specific human brain aging assessment method. Background Art

[0002] Similar to age, brain age is a comprehensive indicator for assessing the degree of brain aging. It is normal for the brain to age to a certain extent as people age, and it is generally believed that the brain age of a healthy person is their actual age. However, due to the influence of neurodegenerative diseases, some brain tissues will shrink or expand at an accelerated rate (such as the hippocampus of patients with Alzheimer's disease will shrink more severely than normal aging), deviating from the normal aging trajectory of the human brain, resulting in a certain difference between brain age and actual age. The academic community has proven that neuroimaging (especially T1-weighted magnetic resonance imaging, which reflects the structure of the human brain) combined with machine learning algorithms can accurately estimate the age of healthy people, that is, brain age prediction. This method has also been widely used in brain aging research in patients with brain diseases, and accelerated brain aging (brain age greater than actual age) has been observed in a variety of neurological diseases (Alzheimer's disease, mild cognitive impairment, schizophrenia, etc.). Some studies have further suggested that brain age can be used as a risk marker for some brain diseases. However, the single value of brain age ignores the regional differences in human brain aging and is not disease-specific (for example, abnormal aging of the medial temporal lobe is most severe in patients with Alzheimer's disease, while Parkinson's disease mainly affects the patient's striatum). The accuracy rate for identifying patients with neurodegenerative diseases is low.

[0003] Machine learning simulates human learning behavior through algorithms. Through continuous training, it discovers patterns and correlations from large data sets, and then forms a model based on the results of data analysis to make the best decisions and predictions for new input data. A major difficulty in machine learning is that the decisions it makes are based entirely on training experience, and the relationship between the prediction results and the input features is not clear. Therefore, the predicted brain age based on machine learning is essentially a "black box model." The model interpretation method is used to explain the reasons why the machine learning model makes decisions. One type of method explains the prediction results of each sample by quantifying the contribution of the feature to the prediction. In brain age prediction, the model interpretation method can quantify the contribution of each neuroimaging feature (brain tissue volume) to the prediction of brain age, attribute the aging of the brain to each brain tissue, and reflect the regional aging of each brain tissue. Summary of the invention

[0004] The purpose of the present invention is to provide a disease-specific method for assessing human brain aging, aiming to solve the above-mentioned technical problems.

[0005] The present invention is achieved by providing a disease-specific human brain aging assessment method, the disease-specific human brain aging assessment method comprising the following steps:

[0006] S1. Extract age-related brain tissue volume features from the subjects' T1-weighted magnetic resonance brain images; T1 is a modality of magnetic resonance imaging that can better image brain tissue structure;

[0007] S2, using the trained XGBoost regression model to estimate the subject’s brain age based on the input brain tissue volume features;

[0008] S3. Perform SHAP feature attribution on the brain age prediction results of the subjects, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging;

[0009] S4. The risk probability of the subject suffering from neurodegenerative diseases (Alzheimer's disease / Parkinson's disease) is evaluated according to the input brain age vector through the XGBoost classifier.

[0010] A further technical solution of the present invention is: the step S1 further includes the following steps:

[0011] S21 used the analysis tools FastSurfer and SPM12 to perform brain tissue segmentation on the brain images uploaded by the subjects;

[0012] S22, obtaining the absolute volume of each brain tissue by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel;

[0013] S23. The absolute volume of each brain tissue is normalized using the intracranial volume to obtain the relative volume of each brain tissue as the input feature of the brain age prediction model.

[0014] A further technical solution of the present invention is: in step S21, the analysis tool FastSurfer segments the human brain magnetic resonance image into 95 brain regions of the whole brain based on the DKT atlas, and the analysis tool SPM12 segments the human brain magnetic resonance image into gray matter, white matter and cerebrospinal fluid.

[0015] A further technical solution of the present invention is: in the normalization of step S23, the ICV estimation algorithm in SPM12 is used to achieve accurate intracranial volume estimation based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance image, and the estimated intracranial volume is used to normalize the absolute volume of each brain region to obtain the relative volume of each brain region as the input feature of the brain age prediction model.

[0016] A further technical solution of the present invention is: in step S3, the brain age of the subject is predicted based on the relative volume and gender information of 98 brain regions of the subject, using the XGBoost regression algorithm, based on the relative volume and gender information of 95 brain regions and gray matter, white matter and cerebrospinal fluid of the DKT atlas, and the actual age of the subject is the prediction target.

[0017] A further technical solution of the present invention is: in step S4, the contribution of each brain tissue to the brain aging of the subject is calculated using the SHAP feature attribution algorithm, and the brain age prediction of each subject is explained based on the additive feature attribution model: Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline (here the average age of the training set samples), N is the number of input features of the brain age prediction model (99), and Φ ij is feature j for subject x i Contribution to brain age prediction.

[0018] Another object of the present invention is to provide a disease-specific human brain aging assessment system, the disease-specific human brain aging assessment system comprising

[0019] A human brain magnetic resonance image feature extraction module, used to extract age-related brain tissue volume features from the subject's T1-weighted magnetic resonance brain images;

[0020] The brain age prediction module is used to estimate the subject's brain age based on the input brain tissue volume features using the trained XGBoost regression model;

[0021] The regional aging attribution module is used to perform SHAP feature attribution on the brain age prediction results of the subjects, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging;

[0022] The neurodegenerative disease risk screening module is used to evaluate the risk probability of a subject suffering from a neurodegenerative disease (Alzheimer's disease / Parkinson's disease) according to the input brain age vector through the XGBoost classifier.

[0023] A further technical solution of the present invention is: the human brain magnetic resonance image feature extraction module also includes

[0024] The segmentation unit is used to perform brain tissue segmentation on the brain images uploaded by the subjects using the analysis tools FastSurfer and SPM12 respectively;

[0025] A volume calculation unit, used for obtaining the absolute volume of each brain tissue by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel;

[0026] The model feature unit is used to normalize the absolute volume of each brain tissue using the intracranial volume to obtain the relative volume of each brain tissue as the input feature of the brain age prediction model.

[0027] A further technical solution of the present invention is: the analysis tool FastSurfer in the segmentation unit segments the human brain magnetic resonance image into 95 brain regions of the whole brain based on the DKT atlas, and the analysis tool SPM12 segments the human brain magnetic resonance image into gray matter, white matter and cerebrospinal fluid; the model feature unit adopts the ICV estimation algorithm in SPM12 during normalization, and based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance image, accurate intracranial volume estimation is achieved, and the estimated intracranial volume is used to normalize the absolute volume of each brain region to obtain the relative volume of each brain region as the input feature of the brain age prediction model.

[0028] A further technical solution of the present invention is: the regional aging attribution module predicts the brain age of the subject based on the relative volume and gender information of the 98 brain regions of the subject, using the XGBoost regression algorithm, based on the relative volume and gender information of the 95 brain regions and gray matter, white matter and cerebrospinal fluid of the DKT atlas, and the actual age of the subject is the prediction target;

[0029] The neurodegenerative disease risk screening module uses the SHAP feature attribution algorithm to calculate the contribution of each brain tissue to the subject's brain aging, and interprets the brain age prediction of each subject based on the additive feature attribution model: Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline (here the average age of the training set samples), N is the number of input features of the brain age prediction model (99), and Φ ij is feature j for subject x i Contribution to brain age prediction.

[0030] The beneficial effects of the present invention are: accurate brain age estimation is achieved by combining human brain magnetic resonance imaging feature extraction and machine learning algorithm (XGBoost). By attributing features to brain age prediction, the contribution of each brain tissue to human brain aging is quantified, regional brain aging is refined, and spatial specificity for different brain diseases is improved. The knowledge of brain aging assessment (brain age vector) is transferred to the screening of neurological diseases (Alzheimer's disease, Parkinson's disease), which can accurately identify the risk of disease in subjects. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a flow chart of a disease-specific human brain aging assessment method provided by an embodiment of the present invention.

[0032] Figure 2 It is a structural block diagram of a disease-specific human brain aging assessment system provided by an embodiment of the present invention.

[0033] Figure 3 It is a schematic diagram of a human brain magnetic resonance image feature extraction module provided by an embodiment of the present invention.

[0034] Figure 4 Schematic diagram of a brain age prediction module provided by an embodiment of the present invention.

[0035] Figure 5 It is a schematic diagram of a regional aging attribution module provided in an embodiment of the present invention.

[0036] Figure 6 It is a schematic diagram of the SHAP feature attribution model provided by an embodiment of the present invention.

[0037] Figure 7 Schematic diagram of a neurodegenerative disease risk screening module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] like Figure 1-7 As shown, the disease-specific human brain aging assessment method provided by the present invention is described in detail as follows:

[0039] During the entire algorithm process, the subject only needs to upload a T1-weighted MRI brain image according to the prompts and enter age and gender information to obtain the brain age, brain age vector and risk probability of Alzheimer's disease or Parkinson's disease corresponding to the image. At the beginning of the algorithm, the program will prompt the subject to upload a T1-weighted MRI brain image and enter age and gender (female input "0", male input "1") information.

[0040] Step S1, extracting age-related brain tissue volume features from the T1-weighted MRI brain images of the subject; the human brain MRI image feature extraction module extracts brain tissue volume features from the T1-weighted MRI brain images of the subject; the algorithm uses the brain MRI image analysis tools FastSurfer and SPM12 to perform whole-brain 95 brain region segmentation and gray matter, white matter, and cerebrospinal fluid segmentation on the brain images uploaded by the subject based on the DKT atlas; the absolute volume of each brain tissue is obtained by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel; the volumes of gray matter, white matter, and cerebrospinal fluid are added together to obtain the intracranial volume of the subject; the absolute volume of each brain tissue is normalized using the intracranial volume (the volume of each brain tissue is divided by the intracranial volume) to obtain the relative volume of 98 brain regions (95 brain regions based on the DKT atlas and gray matter, white matter, and cerebrospinal fluid) as the input features of the brain age prediction model.

[0041] Step S2, using the trained XGBoost (eXtreme Gradient Boosting) regression model to estimate the subject's brain age based on the input brain tissue volume features; Figure 2 As shown, the pre-trained XGBoost brain age prediction model is used to estimate the subject's brain age based on the input brain tissue volume features; Figure 2 As shown in the figure, the brain age prediction model used by the algorithm is based on the XGBoost regression algorithm, which is pre-trained on a multi-center public data set (3344 healthy human samples), with the relative volume of 98 brain regions extracted by the "human brain magnetic resonance imaging feature extraction module" plus gender information (0 female, 1 male) as input features, and age as the prediction target. And accurate prediction results (average absolute error 3.64 years) were obtained on the multi-center test set (830 healthy human samples). When the algorithm is running, the trained brain age prediction model is directly called to estimate the brain age of the subject based on the relative volume of 98 brain regions of the subject extracted by the "human brain magnetic resonance imaging feature extraction module" and the gender information (0 female, 1 male) input by the subject.

[0042] Step S3, perform SHAP (SHapley Additive exPlanations) feature attribution on the brain age prediction results of the subject, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging; Figure 3As shown in the figure, the feature attribution algorithm is used to quantify the contribution of each brain tissue (feature) to the brain aging (brain age prediction) of the subject, and the single value of brain age is expanded into a feature contribution vector reflecting regional brain aging; the algorithm uses the SHAP model interpretation algorithm to construct the connection between the input features and the prediction target in the brain age prediction model in advance, forming a feature attribution model for brain age prediction. When the algorithm is running, the constructed SHAP feature attribution model is directly called to calculate the contribution of each feature to the brain age prediction (brain aging) of the subject based on the relative volume of the 98 brain regions of the subject extracted by the "Human Brain Magnetic Resonance Image Feature Extraction Module" and the gender information (female 0 male 1) input by the subject. These feature contributions reflect the local aging of the corresponding brain region and constitute a brain age vector.

[0043] SHAP feature attribution model, such as Figure 5 As shown, the sum of all feature contributions and the model baseline (60) equals the subject's predicted brain age, and the specific calculation of feature contributions is based on game theory.

[0044] Step S4, using the XGBoost classifier to evaluate the risk probability of the subject suffering from a neurodegenerative disease (Alzheimer's disease / Parkinson's disease) according to the input brain age vector. Figure 6 As shown in the figure, the algorithm uses the brain age vector (excluding gender contribution) plus the real age and gender information of the sample as features, and based on the XGBoost classification algorithm, pre-trains the XGBoost classifier on the ADNI dataset and the PPMI dataset to distinguish between normal people and patients in the prodromal stage of Alzheimer's disease / Parkinson's disease. When the algorithm is running, it directly calls the trained classifier, and classifies the subjects according to the brain age vector (excluding gender contribution) obtained by the "regional aging attribution module" and the age and gender information (female 0 male 1) input by the subjects, and outputs the risk probability of the subjects suffering from Alzheimer's disease or Parkinson's disease respectively.

[0045] The present invention realizes accurate brain age prediction based on human brain magnetic resonance imaging feature extraction and machine learning regression algorithm (XGBoost); at the same time, the SHAP model interpretation algorithm is used to quantify the contribution of each feature to the prediction of the subject's brain age, and the overall aging of the brain is specifically attributed to each brain tissue. In this way, the single value of brain age reflecting the comprehensive aging of the brain is refined into a brain age vector reflecting the local aging of each brain tissue. Finally, combined with the machine learning classification algorithm (XGBoost), the risk of the subject suffering from neurodegenerative diseases (Alzheimer's disease, Parkinson's disease) is evaluated according to the subject's brain age vector.

[0046] The brain tissue segmentation unit uses the deep learning brain imaging tool FastSurfer to segment the human brain magnetic resonance images into 95 brain regions based on the DKT atlas (Desikan-Killiany-Tourville atlas); and uses SPM12 to segment the human brain magnetic resonance images into gray matter, white matter, and cerebrospinal fluid.

[0047] The brain tissue volume calculation unit calculates the absolute volume of each brain region according to the number of voxels and the volume of voxels in each brain region obtained by segmentation:

[0048] Absolute volume of brain region = number of voxels in the brain region × voxel volume

[0049] The ICV normalization unit uses the ICV estimation algorithm in SPM12 to achieve accurate intracranial volume estimation based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance imaging, and then uses the estimated intracranial volume to normalize the absolute volume of each brain region to obtain the relative volume of each brain region:

[0050] Relative volume of brain region = absolute volume of brain region ÷ intracranial volume

[0051] The brain age prediction module predicts the brain age of the subject based on the relative volume and gender information of the subject's 98 brain regions. The brain age prediction model uses the XGBoost regression algorithm, with the relative volume and gender information of 95 brain regions and gray matter, white matter and cerebrospinal fluid based on the DKT atlas as the features, and the actual age of the subject as the prediction target. It is trained and verified in advance on multi-center healthy human data (4174 people), and can accurately predict normal brain aging (mean absolute error of 3.64 years).

[0052] The regional aging attribution module uses the SHAP (Shapley Additive Explanations) feature attribution algorithm to calculate the contribution of each brain tissue (feature) to the subject's brain aging (brain age prediction). The SHAP algorithm explains the brain age prediction of each subject based on the additive feature attribution model:

[0053]

[0054] Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline (here the average age of the training set samples is 60 (years old)), N is the number of input features of the brain age prediction model (99), and Φ ij is feature j for subject x i Contribution of brain age prediction (Shapley value, specific calculation based on game theory), 99 feature contributions (Φ ij,j=1,2,...,99) reflects the corresponding 98 brain regions and genders for the subjects x i The degree of the effects of brain aging constitutes the brain age vector.

[0055] The neurodegenerative disease risk screening module includes an Alzheimer's disease screening classifier and a Parkinson's disease screening classifier, which are used to assess the risk of a subject suffering from Alzheimer's disease and Parkinson's disease, respectively.

[0056] The Alzheimer's disease screening classifier is based on the XGBoost classification algorithm, with brain age vector (excluding gender contribution) and the actual age and gender information of the subject as features, and is trained on the ADNI data set to detect Alzheimer's disease prodromal patients at risk of disease. When the subject's brain age vector is input into the trained classifier, the classifier will give the risk probability of the subject suffering from Alzheimer's disease.

[0057] The Parkinson's disease screening classifier is based on the XGBoost classification algorithm, features the brain age vector and the age and gender information of the subject, is trained on the PPMI data set, and detects patients in the prodromal stage of Parkinson's disease who are at risk of developing the disease. When the subject's brain age vector is input into the trained classifier, the classifier will give the risk probability of the subject suffering from Parkinson's disease.

[0058] Another object of the present invention is to provide a disease-specific human brain aging assessment system, the disease-specific human brain aging assessment system comprising

[0059] A human brain magnetic resonance image feature extraction module, used to extract age-related brain tissue volume features from the subject's T1-weighted magnetic resonance brain images;

[0060] The brain age prediction module is used to estimate the subject's brain age based on the input brain tissue volume features using the trained XGBoost regression model;

[0061] The regional aging attribution module is used to perform SHAP feature attribution on the brain age prediction results of the subjects, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging;

[0062] The neurodegenerative disease risk screening module is used to evaluate the risk probability of a subject suffering from a neurodegenerative disease (Alzheimer's disease / Parkinson's disease) according to the input brain age vector through the XGBoost classifier.

[0063] The human brain magnetic resonance image feature extraction module also includes

[0064] The segmentation unit is used to perform brain tissue segmentation on the brain images uploaded by the subjects using the analysis tools FastSurfer and SPM12 respectively;

[0065] A volume calculation unit, used for obtaining the absolute volume of each brain tissue by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel;

[0066] The model feature unit is used to normalize the absolute volume of each brain tissue using the intracranial volume to obtain the relative volume of each brain tissue as the input feature of the brain age prediction model.

[0067] The analysis tool FastSurfer in the segmentation unit segments the human brain magnetic resonance image into 95 brain regions of the whole brain based on the DKT atlas, and the analysis tool SPM12 segments the human brain magnetic resonance image into gray matter, white matter and cerebrospinal fluid; the model feature unit adopts the ICV estimation algorithm in SPM12 during normalization, and achieves accurate intracranial volume estimation based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance image, and uses the estimated intracranial volume to normalize the absolute volume of each brain region to obtain the relative volume of each brain region as the input feature of the brain age prediction model.

[0068] The regional aging attribution module predicts the brain age of the subject based on the relative volume and gender information of the subject's 98 brain regions, using the XGBoost regression algorithm, based on the relative volume and gender information of the 95 brain regions and gray matter, white matter and cerebrospinal fluid of the DKT atlas, and the actual age of the subject as the prediction target;

[0069] The neurodegenerative disease risk screening module uses the SHAP feature attribution algorithm to calculate the contribution of each brain tissue to the subject's brain aging, and interprets the brain age prediction of each subject based on the additive feature attribution model: Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline (here the average age of the training set samples), N is the number of input features of the brain age prediction model (99), and Φ ij is feature j for subject x i Contribution to brain age prediction.

[0070] Combine human brain magnetic resonance imaging feature extraction and machine learning algorithm (XGBoost) to achieve accurate brain age estimation. Through SHAP feature attribution of brain age prediction, the contribution of each brain tissue to human brain aging is quantified, the regional aging of the brain is refined, and the spatial specificity for different brain diseases is improved. Transferring the knowledge of brain aging assessment (brain age vector) to the screening of neurological diseases (Alzheimer's disease, Parkinson's disease) can accurately identify the risk of disease in subjects.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A disease-specific method for assessing human brain aging, characterized in that: The disease-specific human brain aging assessment method comprises the following steps: S1. Extract age-related brain tissue volume features from the subjects' T1-weighted magnetic resonance brain images; S2, using the trained XGBoost regression model to estimate the subject’s brain age based on the input brain tissue volume features; S3. Perform SHAP feature attribution on the brain age prediction results of the subjects, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging; S4. The risk probability of the subject suffering from neurodegenerative diseases is evaluated according to the input brain age vector through the XGBoost classifier.

2. The disease-specific human brain aging assessment method according to claim 1, characterized in that: The step S1 also includes the following steps: S21 used the analysis tools FastSurfer and SPM12 to perform brain tissue segmentation on the brain images uploaded by the subjects; S22, obtaining the absolute volume of each brain tissue by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel; S23. The absolute volume of each brain tissue is normalized using the intracranial volume to obtain the relative volume of each brain tissue as the input feature of the brain age prediction model.

3. The disease-specific human brain aging assessment method according to claim 2, characterized in that: In step S21, the analysis tool FastSurfer segments the human brain magnetic resonance image into 95 brain regions based on the DKT atlas, and the analysis tool SPM12 segments the human brain magnetic resonance image into gray matter, white matter and cerebrospinal fluid.

4. The disease-specific human brain aging assessment method according to claim 3, characterized in that: In step S23, the ICV estimation algorithm in SPM12 is used for normalization, and accurate intracranial volume estimation is achieved based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance image. The estimated intracranial volume is used to normalize the absolute volume of each brain region to obtain the relative volume of each brain region as the input feature of the brain age prediction model.

5. The disease-specific human brain aging assessment method according to claim 4, characterized in that: In step S3, the brain age of the subject is predicted based on the relative volumes of the 98 brain regions and the gender information of the subject, using the XGBoost regression algorithm, based on the 95 brain regions of the DKT atlas and the relative volumes of gray matter, white matter, and cerebrospinal fluid and the gender information, and the actual age of the subject is the prediction target.

6. The disease-specific human brain aging assessment method according to claim 5, characterized in that: In step S4, the contribution of each brain tissue to the brain aging of the subject is calculated using the SHAP feature attribution algorithm, and the brain age prediction of each subject is explained based on the additive feature attribution model: Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline, N is the number of input features of the brain age prediction model, and Φ ij is the characteristic j for subject x i Contribution to brain age prediction.

7. A disease-specific human brain aging assessment system, characterized in that: The disease-specific human brain aging assessment system comprises A human brain magnetic resonance image feature extraction module, used to extract age-related brain tissue volume features from the subject's T1-weighted magnetic resonance brain images; The brain age prediction module is used to estimate the subject's brain age based on the input brain tissue volume features using the trained XGBoost regression model; The regional aging attribution module is used to perform SHAP feature attribution on the brain age prediction results of the subjects, quantify the contribution of each brain tissue to brain aging, and expand the single value of brain age into a feature contribution vector reflecting regional aging; The neurodegenerative disease risk screening module is used to evaluate the risk probability of a subject suffering from a neurodegenerative disease based on the input brain age vector through the XGBoost classifier.

8. The disease-specific human brain aging assessment system according to claim 7, characterized in that: The human brain magnetic resonance image feature extraction module also includes The segmentation unit is used to perform brain tissue segmentation on the brain images uploaded by the subjects using the analysis tools FastSurfer and SPM12 respectively; A volume calculation unit, used for obtaining the absolute volume of each brain tissue by multiplying the number of voxels in each brain region obtained by segmentation by the volume of a single voxel; The model feature unit is used to normalize the absolute volume of each brain tissue using the intracranial volume to obtain the relative volume of each brain tissue as the input feature of the brain age prediction model.

9. The disease-specific human brain aging assessment system according to claim 8, characterized in that: The analysis tool FastSurfer in the segmentation unit segments the human brain magnetic resonance image into 95 brain regions of the whole brain based on the DKT atlas, and the analysis tool SPM12 segments the human brain magnetic resonance image into gray matter, white matter and cerebrospinal fluid; the model feature unit adopts the ICV estimation algorithm in SPM12 during normalization, and achieves accurate intracranial volume estimation based on the gray matter, white matter, and cerebrospinal fluid segmentation results of the subject's brain magnetic resonance image, and uses the estimated intracranial volume to normalize the absolute volume of each brain region to obtain the relative volume of each brain region as the input feature of the brain age prediction model.

10. The disease-specific human brain aging assessment system according to claim 9, characterized in that: The regional aging attribution module predicts the brain age of the subject based on the relative volume and gender information of the subject's 98 brain regions, using the XGBoost regression algorithm, based on the relative volume and gender information of the 95 brain regions and gray matter, white matter and cerebrospinal fluid of the DKT atlas, and the actual age of the subject as the prediction target; The neurodegenerative disease risk screening module uses the SHAP feature attribution algorithm to calculate the contribution of each brain tissue to the subject's brain aging, and interprets the brain age prediction of each subject based on the additive feature attribution model: Among them, f(x i ) is the subject x i The predicted brain age, E tr [f(x)] is a fixed baseline, N is the number of input features of the brain age prediction model, and Φ ij is the characteristic j for subject x i Contribution to brain age prediction.

Images