A brain feature extraction method, system, device and medium for MRI images

Abstract

The present invention discloses a method, system, device and medium for extracting brain features from MRI images, and relates to the technical field of intelligent biomedical signal processing. When extracting imaging brain features related to the function of the cerebral lymphatic system and the brain structure and morphology in T1WI sequence images, T2WI sequence images and DTI sequence images, the present invention fully exploits the deep features of multiple MRI sequence images. The extracted imaging features related to the function of the cerebral lymphatic system and the brain structure and morphology include the diffusion degree values ​​of the bilateral brain tissue in the patient's brain when converted into fiber bundles, and the fiber bundle reference values ​​that cause the fiber bundles to deviate when the patient's brain discharges. The fiber bundle reference values ​​can reflect which parameters are diffusion parameters when the bilateral brain tissue is abnormal. The diffusion degree values ​​and diffusion parameters reflected by these imaging features are closely related to the patient's epileptic language disorder function and can be used as standards for predicting language disorders in epileptic patients.

Description

Technical Field

[0001] The present invention relates to the field of intelligent biomedical signal processing technology, and in particular to a method, system, device and medium for extracting brain features from MRI images. Background Art

[0002] Epilepsy is one of the most common central nervous system disorders. Its occurrence not only reduces patients' quality of life but also places tremendous pressure and a heavy burden on families and society. Among childhood epilepsy, Rolandic epilepsy, characterized by spikes in the Rolandic area on the EEG, is the most common epilepsy syndrome in childhood, accounting for approximately 15.7% of epilepsy cases. The age of onset of Rolandic epilepsy is generally between 3 and 13 years old, with a peak at 7 to 8 years old. Recent studies have shown that up to 12-65% of children with Rolandic epilepsy experience cognitive impairment, particularly language impairment. Even after epileptic seizures cease and cognitive function improves significantly, language impairment persists. Rolandic epilepsy is more common during school age, and abnormal electrical activity occurs in key language pathways, affecting almost all aspects of language. Furthermore, school age is a critical period for brain development. Therefore, Rolandic epilepsy has a serious and far-reaching impact on the lifelong development of children with Rolandic epilepsy.

[0003] At present, the diagnosis of language disorders in children with Rolandic epilepsy is often made 18-32 months after the onset of epilepsy, which is a delayed diagnosis and greatly affects the prognosis of children. At the same time, there is a lack of effective biomarkers for predicting language disorders in Rolandic epilepsy. The age of 5-6 is the "critical point" of neural plasticity in brain development, with strong plasticity, which is the golden period for intervention. If the risk of language disorders in children with Rolandic epilepsy can be predicted in the early stages of the disease, early intervention will be of great significance and value in reducing the burden on families and society and improving prognosis. Currently, neuropsychological assessment is mostly used to evaluate the language function of patients with Rolandic epilepsy, but it has the characteristics of strong subjectivity of manual assessment and lacks objective evaluation methods. At the same time, the brain changes closely related to Rolandic epilepsy patients with language dysfunction are not yet fully understood in current research. In addition, current research lacks the prediction of language disorders in patients with Rolandic epilepsy in the early stages of the disease.

[0004] To address the problems in the above research, researchers extracted data from Magnetic Resonance Imaging (MRI) to predict language disorders in epileptic patients. However, current methods usually focus on extracting shallow features such as shape changes within a single MRI sequence in MRI images, making it difficult to fully explore the deep features of multiple MRI sequences within MRI images, and difficult to apply to the prediction of language disorders in epileptic patients. Summary of the Invention

[0005] Embodiments of the present invention provide a method, system, device, and medium for extracting brain features from MRI images, which can solve the problem in the prior art that current methods generally focus on extracting shallow features such as shape changes within a single MRI sequence in an MRI image, and are unable to fully exploit deep features of multiple MRI sequences in the MRI image.

[0006] An embodiment of the present invention provides a method for extracting brain features from MRI images, comprising the following steps:

[0007] Obtaining original MRI images of the patient's brain and MRI images after abnormal brain discharge, wherein the MRI images include T1WI sequence images including the cerebral cortex, T2WI sequence images including cerebral blood vessels, and DTI sequence images;

[0008] The original DTI sequence images and the DTI sequence images after abnormal brain discharge were converted into fractional anisotropy FA maps, and the brain tissue parts in the perivascular spaces on both sides of the brain in the fractional anisotropy FA maps were projected into fiber bundles.

[0009] Obtain the diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images after abnormal brain discharge;

[0010] The fiber bundles on both sides of the brain in the original DTI sequence images and the DTI sequence images after abnormal brain discharge were quantified, and the range of changes in the fiber bundle quantization values ​​before and after the abnormal brain discharge was obtained. Each change in the fiber bundle quantization value range corresponds to a diffusion parameter of the brain white matter microstructure;

[0011] The number and volume of spaces around cerebral blood vessels were extracted from T2WI sequence images after abnormal brain discharge, and used together with the diffusion rate of fiber bundles on both sides of the brain as the structural characteristics of cerebral blood vessels; the size, volume and morphology of the cerebral cortex structure were extracted from T1WI sequence images after abnormal brain discharge, and used together with the diffusion parameters of the white matter microstructure as the structural characteristics of the cerebral cortex, and the structural characteristics of the cerebral cortex and the structural characteristics of cerebral blood vessels were used as brain characteristics.

[0012] Preferably, the converting of the original DTI sequence images and the DTI sequence images after abnormal brain discharge into fractional anisotropy FA maps, and projecting the brain tissue portions in the perivascular spaces on both sides of the brain in the fractional anisotropy FA maps into fiber bundles, respectively, comprises:

[0013] The motion artifact correction method is used to correct the motion artifacts in the original DTI sequence images and the DTI sequence images after abnormal brain discharge. Based on the corrected original DTI sequence images and the DTI sequence images after abnormal brain discharge, the color-coded anisotropy fractional FA maps corresponding to the original DTI sequence images and the DTI sequence images after abnormal brain discharge are generated respectively.

[0014] The lateral ventricle body level of the pseudo-color image of the color-coded fractional anisotropy FA map corresponding to the original DTI sequence image and the DTI sequence image after abnormal brain discharge, and the fiber bundles were projected in the bilateral cerebral hemispheres.

[0015] Preferably, the step of obtaining the diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images after abnormal brain discharge includes:

[0016] After projecting the brain tissue parts in the perivascular spaces on both sides of the brain in the fractional anisotropy FA map corresponding to the DTI sequence images after abnormal brain discharge into fiber bundles, the areas corresponding to the fiber bundles were taken as the joint fiber areas;

[0017] Regions of interest (ROIs) of 4 × 3 pixels were drawn within the combined fiber region, and the diffusivities of each fiber in the X, Y, and Z directions were obtained based on the ROIs.

[0018] Preferably, the diffusion parameters of the brain white matter microstructure include:

[0019] Fractional anisotropy FA, ​​mean diffusivity MD, axial diffusivity AD and radial diffusivity RD of white matter microstructural regions.

[0020] Preferably, the extraction of the size, volume and morphology of the cerebral cortex structure includes:

[0021] Segment the T1WI sequence images after abnormal brain discharge into gray matter area, white matter area and cerebrospinal fluid area, and remove the skull area in the T1WI sequence images;

[0022] The cortical volume, cortical thickness, gyrification index, sulcus depth, type dimension and folding curvature index of each tissue in T1WI sequence images were identified and used as the size, volume and morphology of the cerebral cortical structure.

[0023] Preferably, the extraction of the number and volume of perivascular spaces includes:

[0024] The brain shell area is removed from the T2WI sequence images after abnormal brain discharge, and the remaining brain tissue area after removing the brain shell area is segmented to obtain the white matter area, gray matter area and cerebrospinal fluid area;

[0025] Identify the number and volume of white matter areas, gray matter areas, cerebrospinal fluid areas, and perivascular spaces.

[0026] Preferably, the extraction of brain features includes:

[0027] The structural characteristics of cerebral blood vessels are used as imaging features related to the function of the cerebral glymphatic system, and the structural characteristics of the cerebral cortex are used as imaging features related to the brain structure and morphology.

[0028] Among the imaging features related to the function of the glymphatic system and the morphology of brain structure, the t-test was used to preliminarily screen out features with significant correlation ≤ 0.05. Based on the features screened by the t-test, Lasso regression analysis was performed to obtain features with strong correlation with brain neural changes.

[0029] Features that are strongly correlated with brain neural changes are set as brain features.

[0030] An embodiment of the present invention further provides a brain feature extraction system for MRI images, comprising:

[0031] An imaging module is used to obtain original MRI images of the patient's brain and MRI images after abnormal brain discharges, wherein the MRI images include T1WI sequence images including the cerebral cortex, T2WI sequence images including cerebral blood vessels, and DTI sequence images;

[0032] A parameter extraction module is used to convert the original DTI sequence images and the DTI sequence images after abnormal brain discharge into anisotropy fractional FA maps, and project the brain tissue parts in the perivascular spaces on both sides of the brain in the anisotropy fractional FA maps into fiber bundles;

[0033] Obtain the diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images after abnormal brain discharge;

[0034] The fiber bundles on both sides of the brain in the original DTI sequence images and the DTI sequence images after abnormal brain discharge were quantified, and the range of changes in the fiber bundle quantization values ​​before and after the abnormal brain discharge was obtained. Each change in the fiber bundle quantization value range corresponds to a diffusion parameter of the brain white matter microstructure;

[0035] The feature construction module is used to extract the number and volume of spaces around cerebral blood vessels from T2WI sequence images after abnormal brain discharge, and use them together with the diffusion rate of fiber bundles on both sides of the brain as the structural characteristics of cerebral blood vessels; extract the size, volume and morphology of cerebral cortical structure from T1WI sequence images after abnormal brain discharge, and use them together with the diffusion parameters of white matter microstructure as the structural characteristics of cerebral cortex, and use the structural characteristics of cerebral cortex and cerebral blood vessels as brain characteristics.

[0036] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0037] The memory is used to store computer programs;

[0038] The processor is used to implement the steps of the above-mentioned method for extracting brain features from MRI images when executing the computer program stored in the memory.

[0039] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, wherein when the computer program is executed by a processor, the computer program implements the steps of the above-mentioned method for extracting brain features from MRI images.

[0040] The embodiments of the present invention provide a method, system, device, and medium for extracting brain features from MRI images. Compared with the prior art, the methods and systems have the following advantages:

[0041] The present invention first obtains T1WI sequence images, T2WI sequence images and DTI sequence images of the patient's brain, and extracts the number and volume of the perivascular spaces and the diffusion rate of the bilateral fiber bundles in the brain from each sequence image to obtain imaging features related to the function of the cerebral lymphatic system; extracts the structural features of the cerebral cortex and the diffusion parameters of the white matter microstructure from each sequence image to obtain imaging features related to the brain structure morphology; the present invention extracts the imaging features related to the function of the cerebral lymphatic system and the brain structure morphology from the T1WI sequence images, T2WI sequence images and DTI sequence images. When imaging brain features, the deep features of multiple MRI sequence images are fully explored. The extracted imaging features related to the function of the brain lymphatic system and the brain structure morphology include the diffusion degree values ​​of the bilateral brain tissue in the patient's brain when converted into fiber bundles, and the fiber bundle reference values ​​that cause the fiber bundles to deviate when the patient's brain discharges. The fiber bundle reference values ​​can reflect which parameters are the diffusion parameters when the bilateral brain tissue is abnormal. The diffusion degree values ​​and diffusion parameters reflected by these imaging features are closely related to the patient's epileptic language disorder function and can be used as a standard for predicting language disorders in epileptic patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1A schematic diagram of the overall process of a method for extracting brain features from MRI images provided by an embodiment of the present invention;

[0043] Figure 2 A simplified flowchart of a method for extracting brain features from MRI images provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of a process for PVS quantitative analysis of a brain feature extraction method for MRI images provided by an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of a DTI-ALPS index processing flow for a method for extracting brain features from MRI images provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of feature screening of a method for extracting brain features from MRI images provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0048] See also Figure 1 , an embodiment of the present invention provides a method for extracting brain features from MRI images, comprising:

[0049] 1. Obtain clinical data, brain MRI image datasets, and neuropsychological scores of newly diagnosed Rolandic epilepsy patients. Specifically:

[0050] 1. Recruit subjects with Rolandic epilepsy and set inclusion and exclusion criteria.

[0051] Inclusion criteria included:

[0052] ①Children aged 6-9 whose mother tongue is Chinese.

[0053] ②The clinical diagnosis conforms to the Rolandic diagnostic criteria established by the International League Against Epilepsy.

[0054] ③No previous use of anti-epileptic drugs.

[0055] ④No other neuropsychiatric disorders.

[0056] ⑤There are no contraindications for MRI examination.

[0057] Exclusion criteria include:

[0058] ① The MRI image is incomplete or cannot be interpreted due to motion artifacts.

[0059] ② Neuropsychological assessment data are incomplete.

[0060] ③ Conventional MRI of the head shows abnormalities.

[0061] 2. Collection of clinical data, including age, gender, onset time of epileptic seizures, duration of seizures and frequency of seizures.

[0062] 3. Brain MRI images of Rolandic epilepsy patients were acquired using a 3.0T MRI high-resolution magnetic resonance scanner (SignaHDxt 3.0T MRI, GE Healthcare). The scanning sequence included: 3D-T1WI data, T2WI data, and DTI data. The specific scanning parameters are shown in Table 1.

[0063] Table 1 Scanning parameters

[0064]

[0065] 4. Neuropsychological evaluation: Specifically, neuropsychological assessment of children is conducted at the early stage of the disease and 18 months after the onset of the disease.

[0066] The specific assessment scale is: Wechsler Intelligence Scale for Children (Chinese Fourth Edition): It includes tests such as decoding, vocabulary, letter-number sorting, matrix reasoning, comprehension, symbol retrieval, building blocks, similarity, number memorization, and drawing. It can obtain the subject's verbal IQ and verbal comprehension factor, thereby evaluating children's verbal comprehension ability.

[0067] 2. Process the images in the dataset to obtain multiple imaging features and perform feature screening; specifically,

[0068] 1. T1WI images can be processed to obtain cortical volume, thickness, gyration index, and folding curvature indices.

[0069] The SBM method was used to analyze 3D-T1WI data. The CAT12 toolkit was used for processing based on the MATLAB platform. The main steps included image format conversion, spatial normalization, brain tissue segmentation, image modulation, and smoothing. The specific process is as follows:

[0070] ① Image format conversion: convert all image formats from DICOM to NIfTI format.

[0071] ② Spatial standardization: Standardization is to align the brains of subjects of different volumes and shapes to the standard brain template space MNI to unify the coordinate space of all images.

[0072] ③ Brain tissue segmentation: Remove the skull and automatically segment each subject's 3D-T1WI image into gray matter, white matter, and cerebrospinal fluid.

[0073] ④ Image modulation: multiply each spatially normalized image by the relative volume before and after normalization to maintain the total amount of each tissue.

[0074] ⑤ Smoothing: The smoothing kernel size for cortical thickness and sulcus depth feature image processing is 15 mm, and the smoothing kernel size for fractal dimension and folding feature image processing is 25 mm.

[0075] Use Get TIV in the Statistical Analysis module in the CAT12 panel to extract the specific values ​​of cortical volume; use Wrriting options in the Segment module to extract the specific values ​​of cortical thickness; and use the Extract Additional surface Parametters module in the Surface Tools module to extract the specific values ​​of sulcus depth, type dimension, and gyratory index.

[0076] 2. Processing T2WI images can obtain the number and volume of spaces around cerebral blood vessels. The main steps include preprocessing and quantitative index extraction. The specific process is as follows: Figure 3 Specifically including:

[0077] ① Use ITK software to remove the braincase from the head MRI image.

[0078] ② Using the Segment under the fMRI module in the spm 12 toolkit based on the MATLAB platform, the brain tissue was segmented to obtain white matter, gray matter and cerebrospinal fluid.

[0079] ③ Based on the MATLAB platform and the 2D Frangi filter algorithm, the axial T2WI images were segmented to obtain the perivascular space (PVS).

[0080] ④Based on the FSL platform, the number and volume of segmented PVS were quantitatively analyzed according to the following instructions.

[0081] fslmaths ${f} -mul ${f}-mask ${f}_roi

[0082] cluster -i ${f}_roi -t 0.5 --mm>>${f}_clusters.msf

[0083] 3. Based on the DTI images, the diffusion tensor image analysis along the perivascular space index (DTI-ALPS index) is calculated; Figure 4 As shown, specifically including:

[0084] ① Use motion artifact correction method to correct motion artifacts.

[0085] ② Use FSL software to generate color-coded fractional anisotropy (FA) maps.

[0086] ③ At the level of the lateral ventricle body in pseudo-color FA images, 4 × 3 pixel regions of interest were drawn in the projection fiber and association fiber areas of both cerebral hemispheres to obtain the diffusivity of each fiber in the X-, Y-, and Z-axis directions.

[0087] ④Calculate according to the formula DTI-ALPS, which is expressed as:

[0088] index = .

[0089] Where: Dxxproj and Dxxassoci represent the diffusivities of the projected fiber and the associated fiber in the X-axis direction, respectively; Dyyproj represents the diffusivity of the projected fiber in the Y-axis direction; and Dzzassoci represents the diffusivity of the associated fiber in the Z-axis direction.

[0090] 4. Parameter calculation based on diffusion MRI. Specifically including:

[0091] ① Perform brain region extraction, eddy current correction, head motion correction and artifact removal on DTI.

[0092] ②Acquire brain white matter microstructural parameters (FA, MD, AD, RD) based on DTI sequence images.

[0093] 5. Further filter the brain T1WI, T2WI and DTI sequence images to obtain filtered images.

[0094] 6. Based on the filtered images, imaging omics features of brain T1WI, T2WI and DTI sequences were extracted to obtain omics features.

[0095] 7. Combine the t-test and Lasso method to preliminarily screen the most representative features, further combine them with neuropsychological scores, eliminate features with low correlation with prognosis, and thus obtain a small number of important features that have a strong correlation with the prediction results of Rolandic epilepsy language disorder; Figure 5 shown.

[0096] Third, the important features are input into the machine learning model, and the disease is classified according to the classification label to obtain the classification results of Rolandic epilepsy with and without language disorders.

[0097] 4. Calculate the area under the curve, sensitivity, specificity, accuracy and other indicators of the training set and validation set of each model, evaluate the prediction effect, and select the optimal model.

[0098] The specific steps include:

[0099] 1. Data collection.

[0100] 1.1 Data Collection: Based on the previously described inclusion criteria, cranial MRI imaging data were collected from subjects with Rolandic epilepsy. Information on age, gender, seizure onset, duration, and frequency was also collected. Neuropsychological assessments were performed at the onset of epilepsy and 18 months after onset.

[0101] 2. Build a model to predict language impairment in patients with Rolandic epilepsy.

[0102] 2.1 Based on the results of neuropsychological assessment 18 months after onset, subjects with Rolandic epilepsy were divided into Rolandic epilepsy patients with language impairment and Rolandic patients without language impairment;

[0103] 2.2 The subjects with Rolandic epilepsy and those without language impairment were further randomly divided into a training set and a test set in a ratio of 8:2. The sample selection and data set distribution of the two groups are shown in Table 2.

[0104] Table 2 Prediction of language impairment in the Rolandic epilepsy dataset

[0105]

[0106] 2.3 The cranial MRI imaging data were analyzed and processed, and univariate and multivariate logistic regression analyses were performed to determine that the cortical thickness, number and volume of perivascular spaces (PVS) in the left supramarginal gyrus and superior temporal gyrus were highly correlated with patients with Rolandic epilepsy accompanied by language disorders.

[0107] 2.4 The DICOM-formatted cranial MRI image data were imported into 3D Slicer software, and a senior physician manually outlined the region of interest (the left superior temporal gyrus on the T1WI sequence and the perivascular space (VRS) at the centrum semiovale level on the T2WI sequence). This was then reviewed by another senior physician. If there were any objections, the physicians communicated and negotiated with each other to reach a final conclusion.

[0108] 2.5 Radiomics features were extracted from the region of interest using the Python PyRadiomics toolkit. The extracted features were summarized, and features with significant correlations ≤ 0.05 were initially screened using a t-test. Lasso regression analysis was then performed on the features screened by the t-test to identify a smaller number of features with strong correlations with the prediction of Rolandic epilepsy language disorder. In this step, omics features were extracted from the imaging images and progressively screened using t-tests and lasso regression analysis to identify a small number of omics features with strong correlations with Rolandic epilepsy language disorder as key features. Ultimately, five radiomics features were selected as the optimal feature subset.

[0109] 2.6 Training a machine learning model for Rolandic epilepsy language disorder classification.

[0110] The extracted omics features were input into the machine learning algorithm for training, and model training was performed through logistic regression analysis (LRA), K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), Light Gradient Boosting machine (LightGBM), and Extreme Gradient Boosting (XGBoost).

[0111] Sensitivity, specificity, accuracy, area under the receiver operating characteristic curve (AUC), and F1-score were used to evaluate the diagnostic performance of the model. Table 3 shows a comparison of the prediction results of each algorithm.

[0112] Table 3 Prediction results of each algorithm

[0113]

[0114] This study extracts radiomic features from multiple MRI sequences (T1WI, T2WI, and DTI) and performs three feature screenings in combination with the Lasso method and neuropsychological assessment results, thereby constructing an optimal clinical-imaging-neuropsychological multimodal visualization model for Rolandic epilepsy with language disorders. This model can achieve individualized assessment of language function in children with Rolandic epilepsy and objectivity in clinical assessment. This model can be used to conduct a preliminary assessment of language function in children with Rolandic epilepsy in the early stages of the disease.

[0115] This study explored the relationship between fine brain structural changes and language function, finding that the node efficiency of the left superior temporal gyrus, the node efficiency of the left inferior frontal gyrus, the bilateral superior longitudinal fasciculi, and the left uncinate fasciculus were closely related to language function. Furthermore, imaging genomics features were extracted based on multiple MRI sequences (T1WI, T2WI, and DTI), and two feature screenings were performed using the Lasso method and neuropsychological assessment results. This led to the construction of an optimal clinical-imaging-neuropsychological multimodal visualization model for Rolandic epilepsy with language disorders. This model can achieve individualized assessment of the language function of children with Rolandic epilepsy and objectivity in clinical assessment. This model can be used to conduct a preliminary assessment of the language function of children with Rolandic epilepsy in the early stages of the disease.

[0116] The above-described embodiments merely illustrate several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for extracting brain features from MRI images, characterized in that: The following steps are involved: Obtaining original MRI images of the patient's brain and MRI images after abnormal brain discharge, wherein the MRI images include T1WI sequence images including the cerebral cortex, T2WI sequence images including cerebral blood vessels, and DTI sequence images; The original DTI sequence images and the DTI sequence images after abnormal brain discharge were converted into fractional anisotropy FA maps, and the brain tissue parts in the perivascular spaces on both sides of the brain in the fractional anisotropy FA maps were projected into fiber bundles. Obtain the diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images after abnormal brain discharge; The fiber bundles on both sides of the brain in the original DTI sequence images and the DTI sequence images after abnormal brain discharge were quantified, and the range of changes in the fiber bundle quantization values ​​before and after the abnormal brain discharge was obtained. Each change in the fiber bundle quantization value range corresponds to a diffusion parameter of the brain white matter microstructure; The number and volume of spaces around cerebral blood vessels were extracted from T2WI sequence images after abnormal brain discharge, and used together with the diffusion rate of fiber bundles on both sides of the brain as the structural characteristics of cerebral blood vessels; the size, volume and morphology of the cerebral cortex structure were extracted from T1WI sequence images after abnormal brain discharge, and used together with the diffusion parameters of the white matter microstructure as the structural characteristics of the cerebral cortex, and the structural characteristics of the cerebral cortex and the structural characteristics of cerebral blood vessels were used as brain characteristics.

2. The method for extracting brain features from MRI images according to claim 1, wherein: The method converts the original DTI sequence images and the DTI sequence images after abnormal brain discharge into anisotropy fractional FA images, and projects the brain tissue portions in the spaces around cerebral blood vessels on both sides of the brain in the anisotropy fractional FA images into fiber bundles, including: The motion artifact correction method is used to correct the motion artifacts in the original DTI sequence images and the DTI sequence images after abnormal brain discharge. Based on the corrected original DTI sequence images and the DTI sequence images after abnormal brain discharge, the color-coded anisotropy fractional FA maps corresponding to the original DTI sequence images and the DTI sequence images after abnormal brain discharge are generated respectively. The lateral ventricle body level of the pseudo-color image of the color-coded fractional anisotropy FA map corresponding to the original DTI sequence image and the DTI sequence image after abnormal brain discharge, and the fiber bundles were projected in the bilateral cerebral hemispheres.

3. The method for extracting brain features from MRI images according to claim 2, wherein: The diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images obtained after abnormal brain discharges includes: After projecting the brain tissue parts in the perivascular spaces on both sides of the brain in the fractional anisotropy FA map corresponding to the DTI sequence images after abnormal brain discharge into fiber bundles, the areas corresponding to the fiber bundles were taken as the joint fiber areas; Regions of interest (ROIs) of 4 × 3 pixels were drawn within the combined fiber region, and the diffusivities of each fiber in the X, Y, and Z directions were obtained based on the ROIs.

4. The method for extracting brain features from MRI images according to claim 1, wherein: The diffusion parameters of the brain white matter microstructure include: Fractional anisotropy FA, ​​mean diffusivity MD, axial diffusivity AD and radial diffusivity RD of white matter microstructural regions.

5. The method for extracting brain features from MRI images according to claim 1, wherein: The extraction of the size, volume and morphology of the cerebral cortex structure includes: Segment the T1WI sequence images after abnormal brain discharge into gray matter area, white matter area and cerebrospinal fluid area, and remove the skull area in the T1WI sequence images; The cortical volume, cortical thickness, gyrification index, sulcus depth, type dimension and folding curvature index of each tissue in T1WI sequence images were identified and used as the size, volume and morphology of the cerebral cortical structure.

6. The method for extracting brain features from MRI images according to claim 1, wherein: The extraction of the number and volume of the perivascular spaces includes: The brain shell area is removed from the T2WI sequence images after abnormal brain discharge, and the remaining brain tissue area after removing the brain shell area is segmented to obtain the white matter area, gray matter area and cerebrospinal fluid area; Identify the number and volume of white matter areas, gray matter areas, cerebrospinal fluid areas, and perivascular spaces.

7. The method for extracting brain features from MRI images according to claim 1, wherein: The extraction of brain features includes: The structural characteristics of cerebral blood vessels are used as imaging features related to the function of the cerebral glymphatic system, and the structural characteristics of the cerebral cortex are used as imaging features related to the brain structure and morphology. Among the imaging features related to the function of the glymphatic system and the morphology of brain structure, the t-test was used to preliminarily screen out features with significant correlation ≤ 0.

05. Based on the features screened by the t-test, Lasso regression analysis was performed to obtain features with strong correlation with brain neural changes. Features that are strongly correlated with brain neural changes are set as brain features.

8. A brain feature extraction system for MRI images, characterized in that: include: An imaging module is used to obtain original MRI images of the patient's brain and MRI images after abnormal brain discharges, wherein the MRI images include T1WI sequence images including the cerebral cortex, T2WI sequence images including cerebral blood vessels, and DTI sequence images; A parameter extraction module is used to convert the original DTI sequence images and the DTI sequence images after abnormal brain discharge into anisotropy fractional FA maps, and project the brain tissue parts in the perivascular spaces on both sides of the brain in the anisotropy fractional FA maps into fiber bundles; Obtain the diffusivity of the fiber bundles on both sides of the brain in the DTI sequence images after abnormal brain discharge; The fiber bundles on both sides of the brain in the original DTI sequence images and the DTI sequence images after abnormal brain discharge were quantified, and the range of changes in the fiber bundle quantization values ​​before and after the abnormal brain discharge was obtained. Each change in the fiber bundle quantization value range corresponds to a diffusion parameter of the brain white matter microstructure; The feature construction module is used to extract the number and volume of spaces around cerebral blood vessels from T2WI sequence images after abnormal brain discharge, and use them together with the diffusion rate of fiber bundles on both sides of the brain as the structural characteristics of cerebral blood vessels; extract the size, volume and morphology of cerebral cortical structure from T1WI sequence images after abnormal brain discharge, and use them together with the diffusion parameters of white matter microstructure as the structural characteristics of cerebral cortex, and use the structural characteristics of cerebral cortex and cerebral blood vessels as brain characteristics.

9. An electronic device, characterized in that: include: memory and processor; The memory is used to store computer programs; The processor is configured to implement the steps of the method for extracting brain features from MRI images as described in any one of claims 1 to 7 when executing the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that Used to store a computer program, which, when executed by a processor, implements the steps of the method for extracting brain features from MRI images as described in any one of claims 1 to 7.

Images