Classification Method and System for Autism Spectrum Disorder Combining Multi-Parameter MRI Features

Abstract

The present invention relates to a classification method and system for autism spectrum disorder by combining multi-parameter MRI features, including the steps of: extracting cerebrospinal fluid volume features from T1 images; extracting perivascular space diffusion tensor features, fractional anisotropy features, and DTI brain network features from DTI images; inputting the cerebrospinal fluid volume features, perivascular space diffusion tensor features, fractional anisotropy features, and DTI brain network features into a classification model for classification, and calculating classification performance evaluation indicators. The present invention combines multiple features of cerebrospinal fluid volume, DTI-ALPS, FA, and brain network node efficiency to more comprehensively describe the changes in glymphatic system activity and white matter structure in autistic patients. By using the AutoGluon classification framework, adopting a multi-model simultaneous training mode, and using a multi-layer stack integration form, and using a predefined model and hyperparameter search space, the accuracy of the classification method is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly to a classification method and system for autism spectrum disorder by combining multi-parameter MRI features. Background Art

[0002] Autism spectrum disorder (ASD) is a neurodevelopmental disorder, and its core symptoms include social communication and interaction disorders, and restricted repetitive stereotyped behaviors. ASD is prone to occur in preschool children aged 3 - 6 years. This period is a critical period for the rapid growth of the brain and the development of various abilities. If it cannot be accurately diagnosed and intervened in time, it will seriously affect their learning and social interaction, bringing heavy economic and mental burdens to families and society. Therefore, accurate diagnosis and condition assessment of preschool ASD children are helpful to guide clinical scientific and effective intervention and improve the quality of life of children. At present, clinical scales can be used to evaluate the condition of ASD, but mainly rely on the subjective judgment of doctors and lack objective basis. Therefore, there is an urgent need for objective, reliable and highly effective biomarkers.

[0003] Previous studies have confirmed that ASD is related to neurodevelopmental variations in brain structure and functional connectivity. Diffusion tensor imaging (DTI) is an MRI technique for non-invasively evaluating the direction and connectivity of white matter fiber tracts in the brain, which can reveal the microstructural damage of white matter fiber tracts in the brain. For example, fractional anisotropy (FA) can be extracted from DTI images, which describes the anisotropy degree of the diffusion process; white matter network analysis can also be constructed from DTI images. Combining network analysis theory, network metrics such as node efficiency can be calculated; in addition, diffusion tensor imaging analysis along the perivascular space based on diffusion technology (DTI-ALPS) can be used to evaluate the glymphatic system activity of autistic patients. There is a glymphatic system in the brain. Cerebrospinal fluid in the subarachnoid space and interstitial fluid transport soluble Aβ protein, tau protein and energy metabolite lactate out of the brain tissue through the perivascular space. The structural and functional damage caused by ASD will lead to the destruction of the brain's autoregulatory function, neurovascular cerebrospinal fluid decoupling and the accumulation of metabolites, thus hindering the glymphatic circulation.

[0004] Cerebrospinal fluid flows into the interior of the brain through a pipeline system controlled by astrocytes. In addition to DTI-ALPS, the volume value of cerebrospinal fluid can also be calculated from T1-weighted structural magnetic resonance, which can reflect the activity of the glymphatic system. The amount of cerebrospinal fluid volume is related to the hypoplasia of arachnoid granulations and the disorder of the meningeal lymphatic system in ASD patients in the early stage, resulting in reduced absorption.

[0005] The existing technologies all assist in the diagnosis of autism spectrum disorder based on magnetic resonance imaging. For example: The existing solution 1 is a Chinese invention patent with the name of an automatic discrimination system, storage medium and device for autism spectrum disorder, and the application number is 202111050081.2; The existing solution 2 is a Chinese invention patent with the name of a method for predicting autism based on nuclear magnetic resonance images and finding biomarkers, and the application number is 202211000237.0. The existing solution 1 extracts functional connection and white matter connection features from two modal images of resting-state functional magnetic resonance (BOLD) and DTI. The existing solution 2 extracts functional connection features from functional magnetic resonance images. The above two existing solutions mainly involve BOLD functional magnetic resonance and DTI, and only extract brain network features, without extracting cerebrospinal fluid volume features and fractional anisotropy of white matter fiber bundles that can reflect the activity of the glymphatic system.

[0006] Therefore, there is an urgent need for a classification method for autism spectrum disorder that combines multi-parameter MRI features, combines white matter fiber damage, abnormal brain connections, and glymphatic system circulation disorder features, and improves the accuracy of multi-modal MRI in the auxiliary diagnosis of ASD patients. Summary of the Invention

[0007] To achieve the above objects and other advantages of the present invention, the first object of the present invention is to provide a classification method for autism spectrum disorder that combines multi-parameter MRI features, including the following steps:

[0008] Extract cerebrospinal fluid volume features from T1 images;

[0009] Extract perivascular space diffusion tensor features along the DTI images;

[0010] Extract fractional anisotropy features from DTI images;

[0011] Extract DTI brain network features from DTI images:

[0012] Input the cerebrospinal fluid volume features, the perivascular space diffusion tensor features along the, the fractional anisotropy features, and the DTI brain network features into a trained classification model for classification, and calculate the classification performance evaluation index.

[0013] Further, the extracting cerebrospinal fluid volume features from T1 images includes the following steps:

[0014] Convert the format of the original T1 images;

[0015] Redirect and resample each image into a standard format;

[0016] Perform N3 bias field correction on each image and segment the brain region;

[0017] Classify the data into gray matter, white matter, and total cerebrospinal fluid, calculate the volumes of extra-axial cerebrospinal fluid, intra-axial cerebrospinal fluid, and total cerebrospinal fluid, and extract the cerebrospinal fluid volume features.

[0018] Further, the format conversion of the original T1 image is to convert the original T1 Dicom data into NII format data.

[0019] Further, the standard format includes a standard voxel resolution and a standard volume size. The standard voxel resolution is 256×256×256 mm 3 , and the standard volume size is 1×1×1 mm 3 .

[0020] Further, the classification of the data into gray matter, white matter, and total cerebrospinal fluid is performed using a level set-based tissue segmentation method, and the data is classified into gray matter, white matter, and total cerebrospinal fluid by using cortical thickness constraints and longitudinal consistency constraints.

[0021] Further, the classification of the data into gray matter, white matter, and total cerebrospinal fluid further includes the following steps:

[0022] On the image including only gray matter and white matter, fill the intra-axial cerebrospinal fluid region and separate the extra-axial cerebrospinal fluid.

[0023] Further, the extraction of the perivascular space diffusion tensor features from the DTI image includes the following steps:

[0024] Perform format conversion on the original DTI image;

[0025] Extract the files of the DTI image to obtain the color-coded fractional anisotropy map and the diffusivity map, and based on the SWI-derived venous image, determine the axial plane where the vein is perpendicular to the lateral ventricle;

[0026] According to the color-coded principal diffusion direction map, draw the regions of interest in the left hemisphere of the brain and extract the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest;

[0027] Calculate the perivascular space diffusion tensor features through the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest.

[0028] Further, the format conversion of the original DTI image is to convert the original DTI Dicom data into NII format data.

[0029] Further, the extraction of the files of the DTI image is to extract the nii.gz image file, bvals, and bvecs files of the format-converted DTI data.

[0030] Further, the calculation formula for extracting the perivascular space diffusion tensor features is:

[0031] DTI - ALPS index = mean(Dxproj, Dxassoc) / mean(Dyproj, Dzassoc)

[0032] Among them, Dxproj represents the x - axis diffusivity in the projection fiber region, Dxassoc represents the x - axis diffusivity in the association fiber region, Dyproj represents the y - axis diffusivity in the projection fiber region, and Dzassoc represents the z - axis diffusivity in the association fiber region.

[0033] Furthermore, the steps for extracting the anisotropy fraction features from the DTI image include the following:

[0034] Perform format conversion on the original DTI image;

[0035] Pre - process the DTI image after format conversion to generate a diffusion tensor metric matrix;

[0036] Affinely register the white matter fiber bundle anatomical template to the anisotropy fraction image of the experimental data for anatomical localization;

[0037] Calculate the anisotropy fraction values in the corresponding regions during anatomical localization to obtain the anisotropy fraction features.

[0038] Furthermore, the format conversion of the original DTI image is to convert the original DTI Dicom data into NII - format data.

[0039] Furthermore, the calculation formula for the anisotropy fraction features is:

[0040]

[0041] Among them, λ1, λ2, and λ3 represent the three principal axis diameters of the ellipsoid, representing the longest axis diameter, the shortest axis diameter, and the intermediate axis diameter.

[0042] Furthermore, the steps for extracting DTI brain network features from the DTI image include the following:

[0043] Perform format conversion on the original T1 image and DTI image;

[0044] Pre - process the converted T1 image and DTI image to generate a diffusion tensor metric matrix;

[0045] By defining nodes, determining edges, and constructing a DTI brain network, obtain the DTI brain network features of each subject;

[0046] Obtain brain network parameters through the DTI brain network features and FN parameters of the ASD patient group and the normal group, and calculate the node efficiency and centrality indices.

[0047] Further, the format conversion of the original TI image and DTI image is to convert the original TI Dicom data and DTI Dicom data into NII format data.

[0048] Further, the classification model uses the AutoGluon automated machine learning framework.

[0049] Further, the training of the AutoGluon automated machine learning framework includes the following steps:

[0050] Randomly divide the training data set and the test data set according to a preset ratio;

[0051] Perform the same normalization processing on the features of the training set and the test set;

[0052] Put all the training set data into the AutoGluon automated machine learning framework for cross-validation, extract the optimal model in the AutoGluon automated machine learning framework and save the model parameters;

[0053] Input the test set data into the AutoGluon optimal parameter model for classification, and calculate the classification performance evaluation index.

[0054] Further, the classification performance evaluation indexes include accuracy, specificity, sensitivity and AUC value.

[0055] The second object of the present invention is to provide an autism spectrum disorder classification system combining multi-parameter MRI features to implement the above method, including a cerebrospinal fluid volume feature extraction module, a perivascular space diffusion tensor feature extraction module, a fractional anisotropy feature extraction module, a DTI brain network feature extraction module, and a classification performance evaluation index calculation module; wherein,

[0056] The cerebrospinal fluid volume feature extraction module is used to extract cerebrospinal fluid volume features from the T1 image;

[0057] The perivascular space diffusion tensor feature extraction module is used to extract perivascular space diffusion tensor features from the DTI image;

[0058] The fractional anisotropy feature extraction module is used to extract fractional anisotropy features from the DTI image;

[0059] The DTI brain network feature extraction module is used to extract DTI brain network features from the DTI image:

[0060] The classification performance evaluation index calculation module is used to input the cerebrospinal fluid volume feature, the perivascular space diffusion tensor feature along blood vessels, the fractional anisotropy feature, and the DTI brain network feature into a trained classification model for classification and calculate the classification performance evaluation index.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] The present invention provides a classification method and system for autism spectrum disorder combining multi-parameter MRI features to achieve high-accuracy auxiliary diagnosis of autism. The present invention combines multiple features of cerebrospinal fluid volume, DTI-ALPS, FA, and brain network node efficiency, which can more comprehensively describe the changes in glymphatic system activity and white matter structure in autistic patients. By using the AutoGluon classification framework, a multi-model simultaneous training mode is adopted, and in the form of multi-layer stack integration, using a predefined model and hyperparameter search space, data processing, feature selection, model selection, hyperparameter optimization, model integration, etc. are completed; using the AutoGluon classification framework can improve the accuracy of the classification method.

[0063] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly and implement it in accordance with the content of the specification, the following takes the preferred embodiments of the present invention and combines the accompanying drawings to describe in detail as follows. The specific implementation manners of the present invention are given in detail by the following embodiments and their accompanying drawings. Description of the Drawings

[0064] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0065] Figure 1 It is a flowchart of the classification method for autism spectrum disorder in Embodiment 1;

[0066] Figure 2 It is a flowchart of extracting cerebrospinal fluid volume features from T1 images in Embodiment 1;

[0067] Figure 3 It is a flowchart of extracting perivascular space diffusion tensor features along blood vessels from DTI images in Embodiment 1;

[0068] Figure 4 It is a flowchart of extracting fractional anisotropy features from DTI images in Embodiment 1;

[0069] Figure 5 It is a flowchart of extracting DTI brain network features from DTI images in Embodiment 1;

[0070] Figure 6Training process of the AutoGluon automated machine learning framework in Example 1 Figure 1 ;

[0071] Figure 7 Flowcharts of DTI image feature extraction process and T1 image feature extraction process in Example 1;

[0072] Figure 8 Training process of the AutoGluon automated machine learning framework in Example 1 Figure 2 ;

[0073] Figure 9 Schematic diagram of the autism spectrum disorder classification system in Example 2. Detailed implementation manners

[0074] Next, in combination with the accompanying drawings and specific implementation manners, the present invention will be further described. It should be noted that, on the premise of no conflict, the following-described embodiments or technical features can be combined arbitrarily to form new embodiments.

[0075] Example 1

[0076] An autism spectrum disorder classification method combining multi-parameter MRI features, as shown in Figure 1 、 Figure 7 , includes the following steps:

[0077] S1. Extract cerebrospinal fluid volume features from T1 images; specifically, as shown in Figure 2 , includes the following steps:

[0078] S11. Convert the format of the original T1 images; specifically, convert the original T1 Dicom data into NII format data.

[0079] Since the data has intensity inhomogeneity and the image size does not match the standard template, and the directions (axial, coronal, and sagittal views), voxel resolutions, and volume sizes of the original input images may be different. Therefore, S12. Reorient and resample each image into a standard format for further data analysis. The standard format includes a standard voxel resolution and a standard volume size. In this embodiment, the standard voxel resolution and the standard volume size are set to 256×256×256mm 3 and 1×1×1mm 3 .

[0080] After all the input images are reoriented and resampled, S13. Perform N3 bias field correction on each image to eliminate possible intensity inhomogeneity. Use brain extraction methods such as BSE and BET to segment the brain region.

[0081] S14. Classify the data into gray matter, white matter, and total cerebrospinal fluid (CSF), calculate the volumes of extra-axial CSF, intra-axial CSF, and total CSF, and extract the CSF volume features. Specifically, use a level set-based tissue segmentation method to classify the data into gray matter, white matter, and total CSF by using cortical thickness constraints and longitudinal consistency constraints.

[0082] On the image including only gray matter and white matter, fill the intra-axial CSF region. For example, the intra-axial CSF region can be filled using a manual annotation tool to separate the extra-axial CSF.

[0083] S2. Extract the perivascular space diffusion tensor (DTI-ALPS) features from the DTI image; specifically, as Figure 3 shown, it includes the following steps:

[0084] S21. Convert the format of the original DTI image; specifically, convert the original DTI Dicom data into NII format data.

[0085] S22. Extract the files of the DTI image, that is, extract the nii.gz image file, bvals, and bvecs files of the DTI data after format conversion to obtain the color-coded fractional anisotropy map and diffusivity map, and based on the SWI-derived venous image, determine the axial plane where the vein is perpendicular to the lateral ventricle;

[0086] S23. Draw the region of interest (ROI) in the left hemisphere of the brain according to the color-coded principal diffusion direction map, and extract the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest;

[0087] S24. Calculate the perivascular space diffusion tensor features through the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest. The calculation formula is:

[0088] DTI-ALPS index = mean(Dxproj, Dxassoc) / mean(Dyproj, Dzassoc)

[0089] where Dxproj represents the diffusivity of the x-axis in the projection fiber region, Dxassoc represents the diffusivity of the x-axis in the association fiber region, Dyproj represents the diffusivity of the y-axis in the projection fiber region, Dzassoc represents the diffusivity of the z-axis in the association fiber region; DTI-ALPS index represents the DTI-ALPS index, and mean represents the mean operation.

[0090] S3. Extract the fractional anisotropy (FA) features from the DTI image: specifically, as Figure 4 shown, it includes the following steps:

[0091] S31. Convert the format of the original DTI image; specifically, convert the original DTI Dicom data into NII format data.

[0092] S32. Preprocess the DTI image after format conversion. For example, import the converted T1 image and DTI image into the PANDA toolbox, and generate a diffusion tensor metric matrix after preprocessing such as skull stripping and eddy current effect correction.

[0093] S33. Affinely register the white matter fiber bundle anatomical template to the FA image of the experimental data for anatomical localization; in this embodiment, the JHUaltlas is affinely registered to the FA image of the experimental data for anatomical localization. JHUaltlas is a commonly used white matter fiber bundle anatomical template, which is based on statistical data and describes the position, orientation and size of common fiber bundles in the human brain.

[0094] S34. Calculate the fractional anisotropy value of the corresponding region in the anatomical localization, and calculate the fractional anisotropy feature. The calculation formula is:

[0095]

[0096] Among them, λ1, λ2, and λ3 represent the three radial lines in the ellipsoidal direction, representing the longest axis radial line, the shortest axis radial line, and the intermediate axis radial line.

[0097] S4. Extract DTI brain network features from the DTI image: specifically, as Figure 5 shown, including the following steps:

[0098] S41. Convert the format of the original TI image and DTI image; specifically, convert the original TI Dicom data and DTI Dicom data into NII format data.

[0099] S42. Preprocess the T1 image and DTI image after format conversion. For example, import the converted T1 image and DTI image into the PANDA toolbox, and generate a diffusion tensor metric matrix after preprocessing such as skull stripping and eddy current effect correction.

[0100] S43. Obtain the DTI brain network features of each subject by defining nodes, determining edges and constructing the DTI brain network.

[0101] S44. Obtain brain network parameters through the DTI brain network features of the ASD patient group and the normal group and the FN parameter. For example, input the DTI brain network features of the ASD patient group and the normal group, as well as the FN parameter, into the GRETNA software to obtain brain network parameters and calculate node efficiency and centrality metrics.

[0102] S5. Input the cerebrospinal fluid volume characteristics, diffusion tensor characteristics along the perivascular space, fractional anisotropy characteristics, and DTI brain network characteristics into the trained classification model for classification, and calculate the classification performance evaluation indicators.

[0103] In this embodiment, the classification model uses the AutoGluon automated machine learning framework. As Figure 6 , Figure 8 shown, the training of the AutoGluon automated machine learning framework includes the following steps:

[0104] S51. Randomly divide the training dataset and the test dataset according to a preset ratio (for example: 8 to 2);

[0105] S52. Perform the same normalization process on the features of the training set and the test set to maintain data consistency.

[0106] S53. Put all the training set data into the AutoGluon automated machine learning framework, perform 5-fold cross-validation, extract the optimal model in the AutoGluon automated machine learning framework, and save the model parameters;

[0107] S54. Input the test set data into the AutoGluon optimal parameter model for classification, and calculate the classification performance evaluation indicators. Among them, the classification performance evaluation indicators include accuracy, specificity, sensitivity, and AUC value.

[0108] In this embodiment, autism spectrum disorder classification experiments are respectively carried out based on cerebrospinal fluid characteristics, DTI-ALPS characteristics, FA characteristics, DTI brain network characteristics, and multi-parameter MRI characteristics combining cerebrospinal fluid, DTI-ALPS, FA characteristics, and brain network. The experimental results are shown in Table 1. The results show that the accuracy, sensitivity, specificity, and AUC value of classifying autism spectrum disorder by combining multi-parameter MRI characteristics of cerebrospinal fluid, DTI-ALPS, FA characteristics, and brain network have been significantly improved.

[0109] Table 1 Experimental Results

[0110]

[0111] The prior art mainly involves BOLD functional magnetic resonance and DTI. The magnetic resonance images used are different from those of the present invention. The present invention mainly uses two modalities, T1 and DTI, and not only extracts brain network characteristics, but also extracts cerebrospinal fluid volume characteristics and fractional anisotropy of white matter fiber bundles that can reflect the activity of the glymphatic system.

[0112] The present invention provides a classification method, system and device for autism spectrum disorder by combining the lymphatic system and nerve fiber connections, which combines white matter fiber damage, abnormal brain connections and lymphatic system circulation disorder characteristics, and improves the accuracy of multi-modal MRI in the auxiliary diagnosis of ASD patients.

[0113] Through multi-modal magnetic resonance technology, the present invention describes the abnormal changes of the lymphatic circulation and white matter network in ASD patients. Specifically, the extracted cerebrospinal fluid characteristics can be used to describe the association between the severity of ASD disease and the cerebrospinal fluid volume, and the fractional anisotropy of the extracted white matter fiber bundles and the white matter brain network characteristics can be used to describe the association between the severity of ASD disease and brain dysfunction. The above characteristics can help improve the accuracy of the auxiliary diagnosis of autism spectrum disorder.

[0114] The machine learning model in the present invention adopts the AutoGluon framework. AutoGluon is an advanced AutoML framework, including extremely randomized trees, k-nearest neighbors, gradient boosting machines, random forests and tabular neural networks, adopting a multi-model simultaneous training mode and using a multi-layer stack integration form. Its framework has a rich set of predefined models and hyperparameter search spaces, and completes data processing, feature selection, model selection, hyperparameter optimization, model integration and finally result classification. Using the AutoGluon framework can obtain a model with better classification performance.

[0115] Embodiment 2

[0116] An autism spectrum disorder classification system combining multi-parameter MRI features realizes the above method, as Figure 9 shown. The system includes a cerebrospinal fluid volume feature extraction module, a perivascular space diffusion tensor feature extraction module along the perivascular space, a fractional anisotropy feature module, a DTI brain network feature extraction module, and a classification performance evaluation index calculation module; among them,

[0117] The cerebrospinal fluid volume feature extraction module is used to extract cerebrospinal fluid volume features from T1 images;

[0118] The perivascular space diffusion tensor feature extraction module along the perivascular space is used to extract perivascular space diffusion tensor features from DTI images;

[0119] The fractional anisotropy feature module is used to extract fractional anisotropy features from DTI images;

[0120] The DTI brain network feature extraction module is used to extract DTI brain network features from DTI images:

[0121] The classification performance evaluation index calculation module is used to input the cerebrospinal fluid volume feature, the diffusion tensor feature along the perivascular space, the fractional anisotropy feature, and the DTI brain network feature into the trained classification model for classification, and calculate the classification performance evaluation index.

[0122] For a detailed description of the system, reference may be made to the corresponding description in the above method embodiment, which will not be elaborated here.

[0123] It should also be noted that the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, commodity or device including the element.

[0124] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments may be referred to each other, and the differences between each embodiment and other embodiments are emphasized.

[0125] The above is only for the embodiments of this specification and is not used to limit one or more embodiments of this specification. For those skilled in the art, one or more embodiments of this specification can have various changes and transformations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included within the scope of the claims of one or more embodiments of this specification.

Claims

1. A classification method for autism spectrum disorder that combines multi-parameter MRI features, characterized in that, It includes the following steps: Extract cerebrospinal fluid volume features from T1 images; Extract perivascular space diffusion tensor features along the vessels from DTI images; Extract fractional anisotropy features from DTI images; Extract DTI brain network features from DTI images: Input the cerebrospinal fluid volume features, the perivascular space diffusion tensor features along the vessels, the fractional anisotropy features, and the DTI brain network features into a trained classification model for classification, and calculate classification performance evaluation indicators; The step of extracting perivascular space diffusion tensor features along the vessels from DTI images includes the following steps: Convert the format of the original DTI images; Extract the files of the DTI images to obtain color-coded fractional anisotropy maps and diffusivity maps, and based on the SWI-derived venous images, determine the axial plane where the veins are perpendicular to the lateral ventricles; According to the color-coded principal diffusion direction maps, draw regions of interest in the left hemisphere of the brain, and extract the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest; Calculate the perivascular space diffusion tensor features through the diffusivity values of the x-axis, y-axis, and z-axis of each region of interest; The step of extracting fractional anisotropy features from DTI images includes the following steps: Convert the format of the original DTI images; Preprocess the DTI images after format conversion to generate a diffusion tensor metric matrix; Affinely register the white matter fiber bundle anatomical template to the fractional anisotropy images of the experimental data for anatomical localization; Calculate the fractional anisotropy values of the corresponding regions in the anatomical localization to obtain fractional anisotropy features; The calculation formula for the fractional anisotropy features is: where λ1, λ2, and λ3 represent the three principal direction diameters of the ellipsoid, representing the longest axis diameter, the shortest axis diameter, and the intermediate axis diameter.

2. The autism spectrum disorder classification method combining multi-parameter MRI features according to claim 1, characterized in that The step of extracting cerebrospinal fluid volume features from T1 images includes the following steps: Convert the format of the original T1 images; Redirect and resample each image into a standard format; Perform N3 bias field correction on each image and segment the brain regions; Classify the data into gray matter, white matter, and total cerebrospinal fluid, calculate the extra-axial cerebrospinal fluid, intra-axial cerebrospinal fluid, and total cerebrospinal fluid volumes, and extract cerebrospinal fluid volume features.

3. The autism spectrum disorder classification method combining multi-parameter MRI features according to claim 2, characterized in that: The conversion of the format of the original T1 images is to convert the original T1 Dicom data into NII format data.

4. The classification method for autism spectrum disorder combining multi-parameter MRI features according to claim 2, wherein: The standard format includes a standard voxel resolution and a standard volume size. The standard voxel resolution is 256×256×256 mm 3 , and the standard volume size is 1×1×1 mm 3 .

5. The autism spectrum disorder classification method combining multi-parameter MRI features according to claim 2, wherein: The classification of the data into gray matter, white matter, and total cerebrospinal fluid is to use a level set-based tissue segmentation method to classify the data into gray matter, white matter, and total cerebrospinal fluid by using cortical thickness constraints and longitudinal consistency constraints.

6. The classification method of autism spectrum disorder combining multi-parameter MRI features according to claim 5, characterized in that: The classification of the data into gray matter, white matter, and total cerebrospinal fluid further includes the following steps: On the images including only gray matter and white matter, fill the intra-axial cerebrospinal fluid region and separate the extra-axial cerebrospinal fluid.

7. The autism spectrum disorder classification method combining multi-parameter MRI features according to claim 1, characterized in that: The conversion of the format of the original DTI images is to convert the original DTI Dicom data into NI I format data.

8. The classification method for autism spectrum disorder combining multi-parameter MRI features according to claim 7, characterized in that: The extraction of the files of the DTI images is to extract the nii.gz image file, bvals, and bvecs files of the DTI data after format conversion.

9. The autism spectrum disorder classification method combining multi-parameter MRI features according to claim 1, wherein: The calculation formula for extracting perivascular space diffusion tensor features is: DTI-ALPS index = mean(Dxproj, Dxassoc) / mean(Dyproj, Dzassoc) Among them, Dxproj represents the x-axis diffusivity in the projection fiber region, Dxassoc represents the x-axis diffusivity in the association fiber region, Dyproj represents the y-axis diffusivity in the projection fiber region, and Dzassoc represents the z-axis diffusivity in the association fiber region.

10. A classification method for autism spectrum disorder combining multi-parameter MRI features according to claim 1, characterized in that: The format conversion of the original DTI image is to convert the original DTI Dicom data into NI I format data.

11. A classification method for autism spectrum disorder combining multi-parameter MRI features according to claim 1, characterized in that: The extraction of DTI brain network features from the DTI image includes the following steps: Perform format conversion on the original T1 image and DTI image; Preprocess the converted T1 image and DTI image to generate a diffusion tensor metric matrix; Obtain the DTI brain network features of each subject by defining nodes, determining edges, and constructing a DTI brain network; Obtain brain network parameters through the DTI brain network features and FN parameters of the ASD patient group and the normal group, and calculate node efficiency and centrality indexes.

12. A classification method for autism spectrum disorder combining multi-parameter MRI features according to claim 11, characterized in that: The format conversion of the original T1 image and DTI image is to convert the original T1 Dicom data and DTI Dicom data into NI I format data.

13. A classification method for autism spectrum disorder combining multi-parameter MRI features as claimed in claim 1, characterized in that: The classification model uses the AutoGluon automated machine learning framework.

14. A classification method for autism spectrum disorder combining multi-parameter MRI features as claimed in claim 13, wherein: The training of the AutoGluon automated machine learning framework includes the following steps: Randomly divide the training data set and the test data set according to a preset ratio; Perform the same normalization processing on the features of the training set and the test set; Put all the training set data into the AutoGluon automated machine learning framework for cross-validation, extract the optimal model in the AutoGluon automated machine learning framework, and save the model parameters; Input the test set data into the AutoGluon optimal parameter model for classification, and calculate the classification performance evaluation indexes.

15. A classification method for autism spectrum disorder combining multi-parameter MRI features as claimed in claim 1 or 14, characterized in that: The classification performance evaluation indexes include accuracy, specificity, sensitivity, and AUC value.

16. A classification system for autism spectrum disorder that combines multi-parameter MRI features, implementing the method according to any one of claims 1 to 15, characterized in that: It includes a cerebrospinal fluid volume feature extraction module, a perivascular space diffusion tensor feature extraction module, a fractional anisotropy feature extraction module, a DTI brain network feature extraction module, and a classification performance evaluation index calculation module; among them, The cerebrospinal fluid volume feature extraction module is used to extract cerebrospinal fluid volume features from the T1 image; The perivascular space diffusion tensor feature extraction module is used to extract perivascular space diffusion tensor features from the DTI image; The fractional anisotropy feature extraction module is used to extract fractional anisotropy features from the DTI image; The DTI brain network feature extraction module is used to extract DTI brain network features from the DTI image: The classification performance evaluation index calculation module is used to input the cerebrospinal fluid volume features, the perivascular space diffusion tensor features, the fractional anisotropy features, and the DTI brain network features into the trained classification model for classification, and calculate the classification performance evaluation indexes.

Images