Data processing apparatus and data processing method

Abstract

The invention provides a data processing device and method. The device comprises the steps that a data acquisition module acquires structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data; the data processing module respectively preprocesses the three kinds of data to obtain a grey matter volume value, a function connection value and a white matter microstructure value; the analysis and statistics module is used for carrying out statistical analysis on the three types of preprocessed data to obtain a difference grey matter volume value, a difference function connection value and a difference white matter microstructure value; a first determination module determines a most representative difference grey matter volume value, a representative difference function connection value and a representative difference white matter microstructure value according to a preset regression model and the three values after statistical analysis; and the second determination module determines a data evaluation result and an evaluation result grade according to the three most representative values and a machine learning model. According to the invention, the accuracy of ASD diagnosis and evaluation is improved, the diagnosis and evaluation time is shortened, and the efficiency of clinical diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a data processing device and a data processing method. Background Art

[0002] The diagnosis and severity assessment of ASD (Autism Spectrum Disorder) is extremely important for the treatment and care of ASD patients. In related technologies, behavioral assessment and structured interview tools are usually used to diagnose and evaluate ASD, such as the Childhood Autism Rating Scale and the Autism Diagnostic Observation Scale.

[0003] However, the diagnosis and assessment of ASD using the above scheme has problems such as high subjectivity, large errors, long time consumption, lack of support from objective biomarkers such as standardized medical images or biological molecules, and small coverage of severity and early childhood subjects. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0005] To this end, the first purpose of the present invention is to propose a data processing device, which realizes the diagnosis and evaluation of ASD: improving accuracy, adding support for objective biomarkers such as standardized medical images or biological molecules, strengthening coverage of subjects of different severities and early childhood, reducing subjective influences, and reducing the duration of diagnosis and evaluation.

[0006] To this end, a second object of the present invention is to provide a data processing method.

[0007] In order to achieve the above-mentioned purpose, an embodiment of the first aspect of the present invention proposes a data processing device, which includes: a data acquisition module, which is used to acquire scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data; a data processing module, which is used to preprocess the structural magnetic resonance data, the functional magnetic resonance data and the diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values; an analysis and statistics module, which is used to perform statistical analysis on the gray matter volume values, functional connectivity values ​​and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values ​​and differential white matter microstructure values; a first determination module, which is used to determine the most representative differential gray matter volume values, representative differential functional connectivity values ​​and representative differential white matter microstructure values ​​according to a preset regression model, the differential gray matter volume values, the differential functional connectivity values ​​and the differential white matter microstructure values; a second determination module, which is used to determine the data evaluation result and the evaluation result level according to the most representative differential gray matter volume values, the representative differential functional connectivity values ​​and the representative differential white matter microstructure values ​​and a machine learning model.

[0008] According to the data processing device of the embodiment of the present invention, structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data are acquired through a data acquisition module, the data processing module preprocesses the above three types of data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, the analysis and statistics module performs statistical analysis on the three types of preprocessed data to obtain differential gray matter volume values, differential functional connectivity values ​​and differential white matter microstructure values, the first determination module determines the most representative differential gray matter volume values, representative differential functional connectivity values ​​and representative differential white matter microstructure values ​​according to a preset regression model and the three statistically analyzed values, the second determination module determines the data evaluation results and the evaluation result level according to the three most representative values ​​and the machine learning model, so as to realize the diagnosis and evaluation of ASD, improve accuracy, increase the support of objective biomarkers such as standardized medical images or biological molecules, strengthen the coverage of objects of different severity and early childhood, reduce subjective influence, and reduce the diagnosis and evaluation time.

[0009] In some embodiments, the data acquisition module is specifically used to: obtain a first parameter setting for structural magnetic resonance imaging, a second parameter setting for functional magnetic resonance imaging, and a third parameter setting for diffusion tensor imaging; and obtain the structural magnetic resonance data, the functional magnetic resonance data, and the diffusion tensor imaging data according to the first parameter setting, the second parameter setting, and the third parameter setting.

[0010] In some embodiments, the data processing module is specifically used to: remove artifact data and head motion data in the structural magnetic resonance data to obtain the structural magnetic resonance data to be processed; determine the tissue probability map of a set age group based on the structural magnetic resonance data to be processed; perform tissue segmentation on the structural magnetic resonance data to be processed based on the tissue probability map to determine a tissue segmentation image; normalize the tissue segmentation image to the standard coordinate space to obtain an initial gray matter volume value; and smooth the initial gray matter volume value based on a Gaussian smoothing kernel to obtain the gray matter volume value.

[0011] In some embodiments, the data processing module is specifically used to: remove preset preceding frames of the functional MRI data to obtain functional MRI data to be processed; perform slice time correction and head motion calibration on the functional MRI data to be processed to obtain corrected functional MRI data; detect and remove abnormal points of the corrected functional MRI data to obtain target functional MRI data; normalize the target functional MRI data to a standard coordinate space, and perform Gaussian smoothing on the target functional MRI data according to a Gaussian kernel to obtain smoothed functional MRI data; perform delinear trending, covariate removal and bandpass filtering on the smoothed functional MRI data to obtain target preprocessed functional MRI data; extract the time series of the preprocessed functional MRI data according to a preset atlas, and calculate the Pearson correlation coefficient between each region in the preset atlas; determine the functional connectivity value according to the Pearson correlation coefficient.

[0012] In some embodiments, the data processing module is specifically used to: perform head motion and eddy current distortion correction on the diffusion tensor imaging data to obtain corrected diffusion tensor imaging data; align and transform all the corrected diffusion tensor imaging data to the standard space to obtain standardized tensor imaging data; create an average tensor map based on the standardized tensor imaging data, and create an average white matter skeleton; perform permutation-based nonparametric statistics according to the randomization tool to obtain statistically significant white matter microstructure differences, and extract white matter microstructure values. In some embodiments, the analysis and statistics module is specifically used to: determine the level grouping; perform the level grouping corresponding to the gray matter volume value, functional connectivity value and white matter microstructure value to obtain the difference gray matter volume value, difference functional connectivity value and difference white matter microstructure value.

[0013] In some embodiments, the first determination module is specifically used to: input the different gray matter volume values, the differential functional connectivity values ​​and the differential white matter microstructure values ​​into the preset regression model to correspondingly extract the representative differential gray matter volume values, the representative differential functional connectivity values ​​and the representative differential white matter microstructure values.

[0014] In some embodiments, the data processing device also includes a data output module, which is specifically used to: obtain feature values ​​input by a user, and output the data evaluation results based on the feature values, wherein the data evaluation results include data evaluation classifications and evaluation level prediction values ​​corresponding to the data evaluation classifications.

[0015] In some embodiments, the data processing device further comprises: a user interaction interface, wherein the user interaction interface is used to receive magnetic resonance data input by a user and output the data evaluation result and the evaluation result grade.

[0016] In order to achieve the above-mentioned purpose, an embodiment of the second aspect of the present invention proposes a data processing method, which includes: acquiring scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data; preprocessing the structural magnetic resonance data, the functional magnetic resonance data and the diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values; performing statistical analysis on the gray matter volume values, functional connectivity values ​​and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values ​​and differential white matter microstructure values; determining the most representative differential gray matter volume values, representative differential functional connectivity values ​​and representative differential white matter microstructure values ​​according to a preset regression model, the differential gray matter volume values, the differential functional connectivity values ​​and the differential white matter microstructure values; determining data evaluation results and evaluation result levels according to the most representative differential gray matter volume values, the representative differential functional connectivity values ​​and the representative differential white matter microstructure values ​​and a machine learning model.

[0017] According to the data processing method of the embodiment of the present invention, by acquiring structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data, the above three types of data are preprocessed respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, and statistical analysis is performed on the three preprocessed data to obtain difference gray matter volume values, difference functional connectivity values ​​and difference white matter microstructure values. The most representative difference gray matter volume values, representative difference functional connectivity values ​​and representative difference white matter microstructure values ​​are determined according to a preset regression model and the three statistically analyzed values. The data evaluation results and the evaluation result levels are determined according to the three most representative values ​​and the machine learning model to achieve the diagnosis and evaluation of ASD.

[0018] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which: Figure 1 is a block diagram of a data processing device according to an embodiment of the present invention; Figure 2 is a block diagram of a data processing device according to another embodiment of the present invention; Figure 3 is a comparison diagram of gray matter volume differences between children with ASD of different severity and children with typical development according to one embodiment of the present invention; Figure 4 is a schematic diagram of three groups of functional connectivity differences according to one embodiment of the present invention; Figure 5 is a schematic diagram of the differences in white matter microstructures among three groups according to one embodiment of the present invention; Figure 6 is a comparison diagram of AUC (area under the curve) curves of a machine learning model according to an embodiment of the present invention; Figure 7 is a block diagram of a visualization interface of a prediction model according to an embodiment of the present invention; Figure 8 is a flow chart of a data processing method according to an embodiment of the present invention; Fig. 9 It is a detailed flow chart of a data processing method according to an embodiment of the present invention.

[0020] Reference numerals: Data processing device 100; Data collection module 1; data processing module 2; analysis and statistics module 3; first determination module 4; second determination module 5; Subject inclusion 10; Clinical scale assessment 11; MRI data acquisition 12; MRI data preprocessing 13; Feature extraction 14; Machine learning model construction 15; Model evaluation 16; Optimal model extraction and visualization 17; Model application 18; The NB model scored 30; the LR model scored 31. DETAILED DESCRIPTION

[0021] Embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. Embodiments of the present invention are described in detail below.

[0022] ASD is a neurodevelopmental disorder whose core symptoms include deficits in social communication and interaction, as well as narrow interests and stereotyped behaviors. The disease has a high incidence rate, develops in early childhood and persists throughout life, placing a huge economic and medical burden on the patient's family and society. Studies have shown that early diagnosis and intervention can significantly improve the prognosis of ASD individuals and improve their quality of life. Especially for ASD patients with severe symptoms, timely diagnosis and accurate severity assessment are crucial to developing personalized treatment and long-term care plans.

[0023] In the related art, the diagnosis and severity assessment of ASD, for example, usually rely on behavioral assessment and structured interview tools, such as the Childhood Autism Rating Scale and the Autism Diagnostic Observation Scale. However, this method is not only time-consuming, but also has certain errors in the accuracy and consistency of the assessment results, which may lead to delayed diagnosis and miss the best time for early intervention. In addition, due to the complexity of ASD symptoms, the symptoms vary greatly with the individual's developmental stage, age of onset, and whether there are intellectual disabilities and other conditions, which further increases the difficulty of diagnosis and assessment.

[0024] For example, MRI (Magnetic Resonance Imaging) data is combined with machine learning methods to classify ASD and TD (typically developing children) based on single-modality MRI data to complete the diagnosis of ASD. However, this method lacks subgroup classification for the severity of ASD. In addition, the subjects covered by previous diagnostic models are mostly concentrated in ASD in late children, adolescents and adults, and there is a lack of relevant MRI data objects in early children, resulting in limited understanding of brain structure and function of this important population. However, early diagnosis of ASD is crucial to improving prognosis. Early and accurate assessment can provide a key basis for the development of personalized intervention and long-term care plans. Further exploration of the structural and functional changes in the brain of early children will help to improve the early diagnosis rate of ASD, optimize the timing of therapeutic intervention, and improve the quality of life of ASD individuals.

[0025] Therefore, the data processing device of the embodiment of the present invention is used to obtain structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data through the data acquisition module, the data processing module preprocesses the above three types of data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, the analysis and statistics module performs statistical analysis on the three preprocessed data to obtain difference gray matter volume values, difference functional connectivity values ​​and difference white matter microstructure values, the first determination module determines the most representative difference gray matter volume values, representative difference functional connectivity values ​​and representative difference white matter microstructure values ​​according to the preset regression model and the three statistically analyzed values, the second determination module determines the data evaluation results and the evaluation result level according to the three most representative values ​​and the machine learning model, so as to realize the diagnosis and evaluation of ASD, improve the accuracy of diagnosis and evaluation, increase the support of objective biomarkers such as standardized medical images or biological molecules, strengthen the coverage of objects of different severity and early childhood, reduce the influence of subjectivity, and reduce the diagnosis and evaluation time.

[0026] Combine the following Figure 1-Figure 8 The data processing method according to the embodiment of the present invention is described.

[0027] like Figure 1 The data processing device 100 of the present invention comprises: a data acquisition module 1, a data processing module 2, an analysis and statistics module 3, a first determination module 4 and a second determination module 5.

[0028] The data processing device 100 according to the embodiment of the present invention is explained below.

[0029] The data acquisition module is used to acquire the scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data.

[0030] In an embodiment, the structural magnetic resonance data is obtained from a first parameter setting of structural magnetic resonance imaging, the functional magnetic resonance data is obtained from a second parameter setting of functional magnetic resonance imaging, and the diffusion tensor imaging data is obtained from a third parameter setting of diffusion tensor imaging. The first parameter setting includes repetition time, echo time, flip angle, layer thickness, and voxel size of sMRI (Structural Magnetic Resonance Imaging) scanning; the second parameter setting includes repetition time, echo time, flip angle, layer thickness, number of layers, voxel size, and number of times of fMRI (Functional Magnetic Resonance Imaging) scanning; the third parameter setting includes DTI (Diffusion Tensor Imaging, By integrating sMRI, fMRI and DTI imaging data, it is possible to evaluate abnormalities in brain structure and function from multiple angles, significantly improving the comprehensiveness of information and the credibility of diagnostic results, thereby improving the accuracy of ASD diagnosis and assessment, increasing the support of objective biomarkers such as standardized medical images or biomolecules for ASD diagnosis and assessment, and reducing subjective influences.

[0031] The data processing module is used to preprocess the structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values.

[0032] In an embodiment, the gray matter volume value is the volume size or number of gray matter areas in the brain; the functional connectivity value is the strength or degree of functional connection or interaction between different brain regions; and the white matter microstructure value is a quantitative indicator of the microstructural characteristics of white matter.

[0033] Structural MRI data, functional MRI data, and diffusion tensor imaging data were preprocessed to obtain gray matter volume values, functional connectivity values, and white matter microstructure values. For example, sMRI data processing VBM (voxel-based morphometry) involves voxel-level comparison of local gray matter density. Structural image processing is performed using corresponding tools, such as MATLAB (Matrix Laboratory) and SPM (Statistical Parametric Mapping) software. Preliminary quality checks are performed on images to remove obvious artifacts and head motion data. Tissue probability maps for specific age groups are generated using template-related toolboxes or tool sets. Group-level templates are generated based on all participating subjects using functions in SPM software. The images are then tissue segmented and segmented into white matter, gray matter, and cerebrospinal fluid. The tissue segmentation images are spatially normalized and normalized to the standard space. An initial GMV (gray matter volume) is generated for each participating subject. Volume), smooth the initial gray matter volume value with an appropriate Gaussian smoothing kernel to obtain the gray matter volume value; use the functional MRI data analysis tool to process the functional MRI data in the SPM software, remove the image frames with the preset frame number of the first five frames, perform slice time correction on the images after removing the preset frame number, and perform head motion calibration to correct the functional MRI data to obtain corrected functional MRI data; use the functional MRI data analysis tool to detect the corrected functional MRI data, detect abnormal points, and obtain target functional MRI data; normalize the target functional MRI data to the standard coordinate space, and perform spatial smoothing with an appropriate Gaussian kernel to obtain smoothed functional MRI data; use functional MRI imaging data According to the preprocessing technology, physiological noise and other artifacts such as white matter, cerebrospinal fluid and head movement are reduced. Global signal regression can be selected according to needs. Bandpass filtering is performed before calculating functional connectivity. The Pearson correlation coefficient of the time series between each brain region in the functional magnetic resonance data after target preprocessing is obtained according to the time series to obtain the functional connectivity value between each brain region; the diffusion tensor imaging data is corrected for head motion and eddy current distortion to obtain corrected diffusion tensor imaging data; a diffusion tensor model is fitted according to the corrected diffusion tensor imaging data; the corrected diffusion tensor imaging data is weighted to obtain a registered diffusion tensor imaging image; the white matter microstructure value is obtained according to the preset diffusion tensor model and the registered diffusion tensor imaging image. Data preparation is made to determine the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value.

[0034] The analysis and statistics module is used to perform statistical analysis on the gray matter volume value, the functional connectivity value and the white matter microstructure value to obtain the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value.

[0035] In an embodiment, when performing statistical analysis on gray matter volume values, functional connectivity values, and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values, for example, the participants are grouped according to the severity of ASD into three groups, namely, a mild-to-moderate ASD group referred to as ML (Mild-to-Moderate level), a severe ASD group referred to as SL (Severe level), and a typically developing child TD (Typically Developing) group referred to as TD. For sMRI data, variance analysis of ML, SL, and TD groups is performed using the smoothed gray matter volume to obtain differential gray matter volume values ​​of the three groups; for fMRI data, network-level inference uses network-based statistical analysis through non-parametric statistical derivation, multiple residual randomization iterations, etc., to obtain differential functional connectivity values; for DTI data, permutation-based non-parametric statistics are performed according to the randomization tool to compare the differences in anisotropy fraction, mean diffusivity, radial diffusivity, and axial diffusivity of the ROI region between the three groups to obtain differential white matter microstructure values. In order to determine the difference in gray matter volume values, difference in functional connectivity values ​​and difference in white matter microstructure values, it can effectively distinguish between ASD and TD and between mild, moderate and severe manifestations of ASD, improve the accuracy of diagnosis and severity classification, and enhance clinical understanding and response to different symptoms of ASD.

[0036] The first determination module is used to determine the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value according to the preset regression model, the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value.

[0037] In an embodiment, the preset regression model is a model that can extract representative differential gray matter volume values, representative differential functional connectivity values, and representative differential white matter microstructure values ​​according to demand training. When the most representative differential gray matter volume values, representative differential functional connectivity values, and representative differential white matter microstructure values ​​are determined according to the preset regression model, differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values, for example, in order to evaluate the diagnostic prediction and severity assessment of ASD, a minimum absolute shrinkage and selection operator regression model with five-fold cross validation is used to identify the most representative disease classification and severity assessment features from the differential features of the above three groups, and then the relevant algorithm is used to identify the most representative disease classification and severity assessment features. The software applies machine learning algorithms and other operations to extract the optimal model parameters for ASD diagnosis and severity prediction, and inputs the differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values ​​into the preset regression model to determine the most representative differential gray matter volume values, representative differential functional connectivity values, and representative differential white matter microstructure values. Compared with traditional behavioral assessments, the use of multimodal MRI data combined with machine learning algorithms can more objectively and accurately identify ASD children and improve diagnostic accuracy. Moreover, through standardized data processing and model prediction, the diagnosis time can be shortened, allowing clinicians to obtain objective diagnostic results in a shorter period of time and formulate intervention strategies accordingly.

[0038] The second determination module is used to determine the data evaluation results and the evaluation result level based on the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value and the machine learning model.

[0039] In an embodiment, the most representative difference gray matter volume values, representative difference functional connectivity values, and representative difference white matter microstructure values ​​are input into a machine learning model to determine data evaluation results and evaluation result levels, and to perform diagnosis and severity prediction of ASD. When the prediction results of the model are consistent with the clinical diagnosis or evaluation, the results are considered reliable; otherwise, re-evaluation of the patient should be considered, which enhances the objectivity of clinical diagnosis and evaluation, reduces the possibility of misdiagnosis, and ensures the accuracy of diagnosis and evaluation results.

[0040] According to the data processing device of the embodiment of the present invention, structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data are acquired through the data acquisition module, the data processing module preprocesses the above three types of data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, the analysis and statistics module performs statistical analysis on the three types of preprocessed data to obtain difference gray matter volume values, difference functional connectivity values ​​and difference white matter microstructure values, the first determination module determines the most representative difference gray matter volume values, representative difference functional connectivity values ​​and representative difference white matter microstructure values ​​according to a preset regression model and the three statistically analyzed values, the second determination module determines the data evaluation results and the evaluation result level according to the three most representative values ​​and the machine learning model, so as to realize the diagnosis and severity assessment of ASD, improve accuracy, increase the support of objective biomarkers such as standardized medical images or biological molecules, strengthen the coverage of objects of different severity and early childhood, reduce subjective influence, and reduce the diagnosis and evaluation time.

[0041] In some embodiments, the data acquisition module is specifically used to: obtain a first parameter setting for structural magnetic resonance imaging, a second parameter setting for functional magnetic resonance imaging, and a third parameter setting for diffusion tensor imaging; and obtain structural magnetic resonance data, functional magnetic resonance data, and diffusion tensor imaging data according to the first parameter setting, the second parameter setting, and the third parameter setting.

[0042] In an embodiment, the first parameter setting includes the repetition time, echo time, flip angle, layer thickness and voxel size of an sMRI (structural Magnetic Resonance Imaging) scan; the second parameter setting includes the repetition time, echo time, flip angle, layer thickness, number of layers, voxel size and number of times of an fMRI (functional Magnetic Resonance Imaging) scan; the third parameter setting includes the voxel size, repetition time, echo time, flip angle, layer thickness, data matrix and gradient direction of a DTI (diffusion tensor imaging) scan.

[0043] like Figure 2As shown, it is a block diagram of a data processing device according to another embodiment of the present invention. When obtaining the scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data, first, subjects are included 10, and the subjects included 10 are subjects participating in ASD diagnosis and evaluation. For example, 150 children are selected as subjects included 10; secondly, the subjects are evaluated by clinical scale 11, and the clinical scale evaluation 11 is to evaluate the severity of ASD of the subjects through standardized behavioral evaluation tools, such as CARS (Childhood Autism Rating Scale) and ADOS (Autism Diagnostic Observation Schedule) and other standardized behavioral evaluation tools. After the evaluation is completed, the subjects are grouped, for example, into a mild to moderate ASD group, a severe ASD group and a typically developing child TD (typically developing group, Developing) group; finally, performing MRI data acquisition 12 on the participants, obtaining the first parameter setting of structural MRI, the second parameter setting of functional MRI and the third parameter setting of diffusion tensor imaging, so as to achieve the acquisition of structural MRI data, functional MRI data and diffusion tensor imaging data.

[0044] Structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data are correspondingly acquired according to the first parameter setting, the second parameter setting and the third parameter setting to provide a data basis for the diagnosis and evaluation of ASD.

[0045] For example, a multimodal MRI scan is performed on each participant according to the conditions and needs of the participants using a corresponding magnetic resonance imaging device to collect data including structural MRI, functional MRI and diffusion tensor imaging. Specifically, images of all participants are collected using a suitable magnetic resonance scanner. Children or young children who are unable to cooperate in completing the entire MRI scanning process in a conscious state are sedated by oral sedatives or other methods 30 minutes before the MRI scan or before a time set according to the conditions and needs of the participants. Then, all participants are scanned with sMRI, and the obtained structural magnetic resonance data are, for example: repetition time = 8.1 ms, echo time = 3.7 ms, flip angle = 8°, layer thickness = 0.5 mm, and voxel size = 1.0 × 1.0 × 0.5 mm³; then, all participants are scanned with fMRI, and the obtained functional magnetic resonance data are, for example: repetition time = 2000 ms, echo time = 30 ms, flip angle = 90°, layer thickness = 3 mm, number of layers = 50, and voxel size = 2.75 × 2.75 × 3.6 mm³, 200 scans were collected; DTI scans were then performed on all participating subjects, and the diffusion tensor imaging data obtained were, for example: voxel size = 2×2×2 mm³, repetition time = 10000 ms, TE = 96.5 ms, flip angle = 90°, layer thickness = 2 mm, data matrix = 128×128, gradient directions = 33, and the DTI sequence included a baseline image of the DTI scan and images with a diffusion sensitivity factor of 1000 s / mm² in 32 directions.

[0046] In some embodiments, the data processing module is specifically used to: remove artifact data and head motion data in the structural magnetic resonance data to obtain the structural magnetic resonance data to be processed; determine the tissue probability map of the set age group based on the structural magnetic resonance data to be processed; perform tissue segmentation on the structural magnetic resonance data to be processed based on the tissue probability map to determine the tissue segmentation image; normalize the tissue segmentation image to the standard coordinate space to obtain the initial gray matter volume value; smooth the initial gray matter volume value based on the Gaussian smoothing kernel to obtain the gray matter volume value.

[0047] In the embodiments, artifact data is an image of a non-real tissue structure that appears in magnetic resonance imaging, data that does not conform to the actual anatomical structure or tissue; head motion data is data that produces artifacts in magnetic resonance imaging and affects image quality; the tissue probability map for a set age group is a tissue probability map in an image of a participant according to a set age group; tissue segmentation is the segmentation of tissue in an image; the standard coordinate space is a standard brain three-dimensional space coordinate system obtained based on MRI scans of a large number of normal subjects; Gaussian smoothing is an image processing technique that weights pixels using a Gaussian function to achieve a blurring effect on the image.

[0048] like Figure 2 As shown, after the nuclear magnetic resonance data acquisition 12 is completed, the nuclear magnetic resonance data needs to be preprocessed 13. When the structural magnetic resonance data is preprocessed to obtain the gray matter volume value, the artifact data and head motion data in the structural magnetic resonance data are removed to obtain the structural magnetic resonance data to be processed. For example, sMRI data processing VBM (voxel-based morphometry) involves voxel-level comparison of local gray matter density, and structural image processing is performed using corresponding tools, such as MATLAB (Matrix Laboratory) and SPM (Statistical Parametric Mapping) software, etc., to perform a preliminary quality check on the image to remove obvious artifacts and head motion data, so as to achieve interference removal and more accurate data.

[0049] The tissue probability map of a set age group is determined according to the magnetic resonance data of the structure to be processed. For example, a template-dependent toolbox or tool set is used to generate a tissue probability map of a specific age group, so as to achieve preliminary classification of images corresponding to the participating objects.

[0050] Perform tissue segmentation on the tissue probability map to determine a tissue segmentation image. For example, use a function in the SPM software to generate a group-level template according to all participating objects, and perform tissue segmentation on the image to segment the image into white matter, gray matter, and cerebrospinal fluid to achieve further processing of the image.

[0051] The tissue segmentation image is normalized to the standard coordinate space to obtain the initial gray matter volume value. For example, the tissue segmentation image is spatially normalized, the segmented image is normalized to the standard space, and an initial GMV (gray matter volume) is generated for each participating object to obtain a more accurate initial gray matter volume value.

[0052] The initial gray matter volume value is smoothed based on a Gaussian smoothing kernel to obtain a gray matter volume value. For example, the initial gray matter volume value is smoothed using a suitable Gaussian smoothing kernel to obtain a gray matter volume value, so as to obtain a more accurate gray matter volume value.

[0053] In some embodiments, the data processing module is specifically used to: remove preset preceding frames of functional MRI data to obtain functional MRI data to be processed; perform slice time correction and head motion calibration on the functional MRI data to be processed to obtain corrected functional MRI data; detect and remove abnormal points of the corrected functional MRI data to obtain target functional MRI data; normalize the target functional MRI data to a standard coordinate space, and perform Gaussian smoothing on the target functional MRI data according to a Gaussian kernel to obtain smoothed functional MRI data; remove linear trends, remove covariates, and perform bandpass filtering on the smoothed functional MRI data to obtain target preprocessed functional MRI data; extract the time series of the preprocessed functional MRI data according to a preset atlas, and calculate the Pearson correlation coefficient between each region in the preset atlas; determine the functional connectivity value according to the Pearson correlation coefficient.

[0054] In an embodiment, preset pre-order frame: the preset number of frames at the beginning of scanning; slice time correction: in functional magnetic resonance imaging, due to the difference in acquisition time between scanning layers, time offsets in time series data may occur. Slice time correction aims to adjust these time offsets so that the time series data of all voxels are aligned in time, thereby eliminating artifacts caused by different slice orders; head motion calibration: correcting artifacts caused by head movement during scanning to ensure data accuracy; outliers: in the process of head motion calibration, outliers refer to data that are significantly different from other data points, which may represent extreme cases of head movement or errors in data acquisition.

[0055] When preprocessing the functional magnetic resonance data to obtain the functional connectivity value, the preset preceding frames of the functional magnetic resonance data are removed to obtain the functional magnetic resonance data to be processed. For example, the functional magnetic resonance data are processed in the SPM software using a functional magnetic resonance data analysis tool, and the image frames with a preset number of frames being the first five frames are removed to ensure the accuracy of the functional magnetic resonance data to be processed.

[0056] The fMRI data to be processed are subjected to slice time correction and head motion calibration to obtain corrected fMRI data. For example, slice time correction is performed on the image with a preset number of frames removed, and head motion calibration is performed to correct the fMRI data to obtain corrected fMRI data, so as to ensure the accuracy of the corrected fMRI data.

[0057] Detect abnormal points of the corrected functional MRI data to obtain target functional MRI data. For example, use a functional MRI data analysis tool to detect the corrected functional MRI data, detect abnormal points, and obtain target functional MRI data to ensure the accuracy of the target functional MRI data.

[0058] The target functional magnetic resonance data is normalized to the standard coordinate space, and the target functional magnetic resonance data is smoothed according to the Gaussian kernel to obtain the smoothed functional magnetic resonance data. For example, the target functional magnetic resonance data is normalized to the standard coordinate space, and the spatial smoothing is performed with a suitable Gaussian kernel to obtain the smoothed functional magnetic resonance data to reduce noise and improve the identifiability of the signal. This process ensures the accuracy and reliability of the data.

[0059] The smoothed functional MRI data are subjected to de-linear trending, covariate removal, and band-pass filtering to obtain preprocessed functional MRI data. For example, the functional MRI data preprocessing technology is used to reduce physiological noise and other artifacts, such as white matter, cerebrospinal fluid, and head movement. Global signal regression can be chosen not to be performed according to needs, and band-pass filtering can be performed before calculating functional connectivity to ensure the accuracy of the target functional MRI data.

[0060] The displacement between frames in the preprocessed functional magnetic resonance data is obtained, and the abnormal functional magnetic resonance data is removed according to the frame displacement. For example, the preset displacement is the critical value for judging abnormal frames, and the preset displacement is set to 0.5 mm. The preset global signal standard score change value is the critical value for judging abnormal frames, and the preset global signal standard score change value is set to 3. Frames with frame displacement exceeding the preset displacement of 0.5 m or global signal standard score change exceeding 3 are marked as abnormal points and removed. If the average frame displacement exceeds 0.5 mm, the entire time series will be excluded.

[0061] The time series of preprocessed fMRI data was extracted according to the preset atlas, and the Pearson correlation coefficients between the various regions in the preset atlas were calculated. For example, the preset atlas included 132 brain regions: cortex, subcortical region, and cerebellum region from the automatic anatomical annotation atlas; the Pearson correlation coefficient is a statistical indicator used to measure the degree of linear correlation between two variables X and Y. Functional connectivity analysis was performed in the mild-to-moderate ASD group, the severe ASD group, and the TD group of typically developing children. The analysis included the definition of brain regions and multiple regions of interest in the labeling set, including the cortex, subcortical region, and cerebellum region, and the Pearson correlation coefficients between the various brain regions were calculated to prepare for determining the functional connectivity value. Functional connectivity values ​​were determined based on the Pearson correlation coefficient. For example, in order to enhance the normality of statistical analysis, the correlation coefficient was transformed by Fisher's z-transformation. The variance analysis method was used to analyze and extract the differential functional connectivity values ​​of the mild-to-moderate ASD group, the severe ASD group, and the TD group of typically developing children. For example, ASD children had abnormal functional connectivity in the temporal auditory cortex, the occipital visual cortex, and the temporoparietal junction. The severe ASD group showed coexistence of hyperconnectivity and hypoconnectivity, and the mild ASD group showed a hypoconnectivity pattern, so as to achieve the determination of functional connectivity values ​​and prepare for differential functional connectivity values.

[0062] In some embodiments, the data processing module is specifically used to: perform head motion and eddy current distortion correction on diffusion tensor imaging data to obtain corrected diffusion tensor imaging data; align and transform all corrected diffusion tensor imaging data to a standard space to obtain standardized tensor imaging data; create an average tensor map based on the standardized tensor imaging data, and create an average white matter skeleton; perform permutation-based nonparametric statistics based on a randomization tool to obtain statistically significant white matter microstructural differences, and extract white matter microstructural values.

[0063] In the embodiments, eddy current distortion correction: a method for processing image distortion caused by eddy current effect in MRI technology. During MRI scanning, eddy current effect can cause image distortion, affecting the accuracy and reliability of the image; diffusion tensor model: a mathematical tool that describes the molecular diffusion characteristics from a three-dimensional perspective, which is a 3x3 symmetric, positive definite matrix containing three eigenvalues ​​(λ1, λ2, λ3) and corresponding eigenvectors. These eigenvectors and eigenvalues ​​respectively reflect the diffusion direction and diffusion degree of water molecules.

[0064] The diffusion tensor imaging data are preprocessed respectively. When the white matter microstructure value is obtained, the diffusion tensor imaging data is corrected for head motion and eddy current distortion to obtain corrected diffusion tensor imaging data. For example, head motion correction and eddy current distortion correction are performed through a comprehensive analysis tool of diffusion tensor magnetic resonance imaging to obtain corrected diffusion tensor imaging data. In this process, abnormal eddy current points are image abnormalities caused by eddy current phenomena generated in conductive materials (such as metal parts of MRI scanners and patients' bodies) due to rapid changes in gradient magnetic fields in magnetic resonance imaging. When correcting the eddy current distortion of the diffusion tensor imaging data, the abnormal eddy current points in the diffusion tensor imaging data are obtained to determine the abnormal image. For example, head motion correction and eddy current distortion correction are performed through a comprehensive analysis tool of brain functional magnetic resonance imaging, eddy current abnormal points in the image are detected, determined to be abnormal images, and Gaussian processes are used to predict and replace the abnormal images to obtain normal images to ensure the accuracy of the corrected diffusion tensor imaging data. After obtaining the corrected diffusion tensor imaging data, the comprehensive analysis tool is used to align and transform the corrected diffusion tensor imaging data into a standard space to obtain standardized tensor imaging data. In other words, the comprehensive analysis tool is used to align the diffusion weighted image to the b0 image, which is an image obtained when the diffusion sensitivity parameter, i.e., the b value, is set to 0 during DTI scanning, to obtain standardized tensor imaging data. That is, the diffusion tensor model is obtained by fitting the standardized tensor imaging data. By aligning and transforming the corrected diffusion tensor imaging data into a standard space, the distortion caused by eddy currents and subject motion can be reduced.

[0065] An average tensor map is created based on the standardized tensor imaging data, and an average white matter skeleton is created. The average tensor map is a map obtained by calculating the average value of the relative RMS (Root Mean Square Displacement) displacement of the standardized tensor imaging data. The average white matter skeleton is obtained based on the average tensor map; for example, since significant head movement may have a significant impact on the results of diffusion tensor imaging data, the root mean square displacement is used to quantify the head movement, and the relative RMS displacement average between each image and the previous image is calculated to create an average tensor map, where each tensor can be calculated through its three main eigenvectors (λ1, λ2, λ3), thereby obtaining the values ​​of the anisotropy fraction, the average diffusivity, the radial diffusivity, and the axial diffusivity; all images are aligned to the standard space through affine transformation to obtain the average white matter skeleton.

[0066] Permutation-based nonparametric statistics are performed according to the randomization tool to obtain statistically significant white matter microstructural differences and extract white matter microstructural values. For example, white matter labels are used to segment white matter to obtain multiple ROI (Region of Interest) areas, and then the average anisotropy fraction, average diffusivity, radial diffusivity, and axial diffusivity values ​​of each fiber bundle ROI area in the definition and labeling set of the brain region are calculated, that is, the white matter microstructural values, in preparation for obtaining differential white matter microstructural values.

[0067] In some embodiments, the analysis and statistics module is specifically used to: determine the level grouping; and level group the gray matter volume values, functional connectivity values, and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values.

[0068] In an embodiment, when the gray matter volume values, functional connectivity values, and white matter microstructure values ​​are statistically analyzed to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values, the level grouping is determined, for example, Figure 2 As shown, in the corresponding feature extraction 14, the participants were grouped according to the severity of ASD and divided into three groups, namely, the mild-to-moderate ASD group referred to as ML (Mild-to-Moderate level), the severe ASD group referred to as SL (Severe level) and the typically developing children TD (Typically Developing) group referred to as TD. For the classification of ASD and TD, a single feature set, i.e., only sMRI features, only fMRI features or only DTI features, and a combined feature set, i.e., sMRI, fMRI and DTI features, were used as input feature sets to classify the severity of ASD, and then the data were randomly divided into training and test sets. Each algorithm used a grid search with five-fold cross-validation to find the best parameters in the training samples, and then used these best parameters to retrain the entire training set, and used indicators such as accuracy on the test set, area under the receiver operating characteristic curve, sensitivity and specificity to evaluate the performance of the classifier in order to achieve more accurate result comparison and processing.

[0069] Gray matter volume values, functional connectivity values, and white matter microstructure values ​​were grouped to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values. For example, for sMRI data, the smoothed gray matter volume was used to perform variance analysis of the ML, SL, and TD groups to obtain the difference values ​​of the three groups, differential gray matter volume. Age, gender, and total intracranial volume were used as irrelevant covariates in the model. In order to exclude edge effects, an absolute threshold mask was applied to the gray matter volume. For example, the results obtained were obtained using an uncorrected whole-brain voxel threshold of p<0.001, and a cluster-level family-wise error rate-corrected multiple comparison with a significance level set at p<0.05. For brain regions showing significant statistical differences, post hoc tests were used to ensure the accuracy of the differential gray matter volume. For fMRI data, network-level inference was performed using network-based statistical analysis through non-parametric The results were evaluated by false discovery rate control at p<0.05 at the connection level and by FWE control at p<0.05 at the network level. ANOVA was performed between groups to identify any significant functional connectivity differences. For brain regions showing significant differences, post hoc tests were performed, and age, gender, and head motion were regressed as covariates to ensure the accuracy of the differential functional connectivity values. For DTI data, ANOVA was used to compare the differences in the values ​​of fractional anisotropy, mean diffusivity, radial diffusivity, and axial diffusivity of the ROI regions between the three groups. The covariates were age and gender, and post hoc tests were performed. The statistical significance threshold was set at p<0.05, and multiple comparison corrections were performed using statistical methods that controlled the false discovery rate to ensure the accuracy of the differential white matter microstructure values.

[0070] like Figure 3 As shown, a comparison diagram of gray matter volume differences between children with ASD of different severity and children with typical development in one embodiment of the present invention. ML is the mild-moderate group; SL is the severe group; TD is the normal development group; ** indicates the whole brain voxel threshold p<0.01, *** indicates the whole brain voxel threshold p<0.001; Figure A shows the areas with increased gray matter volume in the three groups, mainly involving the left superior temporal gyrus, and Figure B shows the differences in gray matter volume in STG among the three groups (SL>ML>TD).

[0071] like Figure 4The figure shows the difference in functional connectivity between three groups according to one embodiment of the present invention. ASD children have abnormal functional connectivity in the temporal auditory cortex, occipital visual cortex and temporoparietal junction; the severe ASD group showed coexistence of hyperconnectivity and hypoconnectivity, and the mild ASD group showed a hypoconnectivity pattern. Figures A and B show the difference between the ML group and the SL group; Figure C shows the difference between the ML group and the TD group; Figures D and E show the difference between the SL group and the TD group; the lines in Figures A, C and D indicate weakened functional connectivity; the lines in Figures B and E indicate enhanced functional connectivity.

[0072] like Figure 5 , a schematic diagram of the differences in white matter microstructures among three groups according to an embodiment of the present invention. MD is mean diffusivity; RD is radial diffusivity; AD is axial diffusivity; R is right; L is left; ACR is anterior corona radiata; SCR is superior corona radiata; RIC is posterior limb of internal capsule; EC is external capsule; SLF is superior longitudinal fasciculus; UF is uncinate fasciculus; ML is mild-moderate group; SL is severe group; TD is normal development group; it shows the increase of MD, RD, and AD values ​​in white matter fiber bundles of ASD children with different severities. The leftmost column represents the ML group, the middle column represents the SL group, and the rightmost column represents the TD group.

[0073] In some embodiments, the first determination module is specifically used to: input the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value into a preset regression model to correspondingly extract the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value.

[0074] In an embodiment, the preset regression model is a model that is trained according to demand and can extract the most representative difference gray matter volume value, the representative difference functional connectivity value, and the representative difference white matter microstructure value. When the most representative difference gray matter volume value, the representative difference functional connectivity value, and the representative difference white matter microstructure value are determined according to the preset regression model, the difference gray matter volume value, the difference functional connectivity value, and the difference white matter microstructure value, for example, Figure 2As shown, in the corresponding machine learning model construction 15 and model evaluation 16, in order to evaluate the diagnostic prediction and severity assessment of ASD, the minimum absolute shrinkage and selection operator regression model with five-fold cross-validation is used to identify the most representative disease classification and severity assessment features from the difference characteristics of the above three groups. In the corresponding optimal model extraction and visualization, five commonly used machine learning algorithms are applied through relevant algorithm software, including: support vector machine, random forest, K nearest neighbor algorithm, logistic regression and naive Bayes. In order to improve the reliability of the model, the experiment can be repeated many times, and the accuracy, area under the curve, sensitivity and specificity of the test set are calculated. The non-parametric permutation test is used for multiple iterations to evaluate the statistical significance of the observed classification accuracy, and the extraction of the optimal model for ASD diagnosis and The optimal model parameters for severity prediction are obtained, and the corresponding platform visualization model is used to input the differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values ​​into the preset regression model to correspondingly extract the representative differential gray matter volume values, representative differential functional connectivity values, and representative differential white matter microstructure values, so as to predict whether the participants are ASD or TD, and classify individuals into ML and SL subgroups when applicable. By fusing sMRI, fMRI, and DTI imaging data, it is possible to evaluate abnormalities in brain structure and function from multiple angles, significantly improve the comprehensiveness of information and the credibility of diagnostic results, and thus realize the diagnosis and evaluation of ASD, improve its accuracy, increase the support of objective biomarkers such as standardized medical images or biological molecules, reduce subjective influences, and reduce the duration of diagnosis and evaluation.

[0075] like Figure 6 As shown, the AUC (Area Under Curve) curve comparison of the machine learning model of one embodiment of the present invention. It shows the performance differences of different models, namely single-modality and multi-modality features in classifying ASD and assessing severity; Figure A shows the average ROC (Receiver Operating Characteristic Curve) curves of various models used to distinguish between children with autism spectrum disorder and normal development, among which the naive Bayes model combining sMRI, fMRI and DTI features performs best in distinguishing ASD from TD, with an AUC score of 0.9335, i.e., a NB (Naive Bayes Model) model score of 30. Figure B shows the average ROC curves of different models used to distinguish mild, moderate and severe ASD, and the logistic regression model combining sMRI and fMRI features performs best, with an AUC score of 0.9218, i.e., a LR (Logistic Regression Model) model score of 31.

[0076] In some embodiments, the data processing device also includes a data output module, which is specifically used to: obtain feature values ​​input by the user, and output data evaluation results based on the feature values, wherein the data evaluation results include data evaluation classification and evaluation level prediction values.

[0077] In an embodiment, data evaluation classification is the classification of user input data, for example, classification according to characteristics related to ASD such as age; the evaluation level prediction value is the evaluation of the severity level of ASD; there is a data output module in the data processing device, in which the characteristic values ​​input by the user can be obtained, the data output module evaluates the characteristic values ​​input by the user, and outputs an evaluation result, the evaluation result includes the classification of data evaluation and the evaluation level prediction value, so as to improve the efficiency of diagnosis and evaluation and improve the user experience.

[0078] In some embodiments, the data processing device further comprises: a user interaction interface, the user interaction interface being used to receive magnetic resonance data characteristic values ​​input by a user, and output data evaluation results and evaluation result grades.

[0079] In the embodiment, the data evaluation result is the valid result corresponding to each data input by the user; the evaluation result level is the level in the evaluation level prediction value; Figure 7 FIG. 1 is a block diagram of a visualization interface of a prediction model of an embodiment of the present invention. After determining the data diagnosis result based on the representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value and the machine learning model, for example, combined with Figure 2 In the corresponding model application 18, an interactive application can be developed based on the optimal model parameters, allowing the user to input MRI feature values ​​into the parameters of the ASD diagnosis prediction model and the parameters of the ASD severity prediction model in the interactive interface of the application. After the parameters are entered, the system corresponding to the application will automatically diagnose and predict the severity of ASD. The interface of the application is intuitive and easy to operate, which greatly improves the efficiency of clinical diagnosis.

[0080] It is understandable that, based on the interactive interface of the above application, in clinical applications, the application can integrate the analysis and statistics module, the first determination module, the second determination module and other information processing modules related to diagnosis and severity prediction level assessment, so as to achieve data acquisition to obtain structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data, pre-process the above three data respectively, obtain gray matter volume value, functional connectivity value and white matter microstructure value, input gray matter volume value, functional connectivity value and white matter microstructure value in the interactive interface of the application, and the application will automatically extract the most representative difference gray matter volume value, representative difference functional connectivity value and representative difference white matter microstructure value, and combine the machine learning model in the application to perform diagnosis and severity prediction level assessment. It is worth emphasizing that the above-mentioned model consumes less time in actual use. From data acquisition to inputting the interactive interface after pre-processing, it only takes about 4 hours to obtain the diagnosis and severity prediction level assessment results. Its pre-processing method is simple, and general technicians can master it after one week of training, which significantly improves the practicality and promotion value of the patent.

[0081] According to the data processing device of the embodiment of the present invention, structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data are acquired through the data acquisition module, the data processing module preprocesses the above three types of data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, the analysis and statistics module performs statistical analysis on the three types of preprocessed data to obtain difference gray matter volume values, difference functional connectivity values ​​and difference white matter microstructure values, the first determination module determines the most representative difference gray matter volume values, representative difference functional connectivity values ​​and representative difference white matter microstructure values ​​according to a preset regression model and the three statistically analyzed values, the second determination module determines the data evaluation results and the evaluation result level according to the three most representative values ​​and the machine learning model, so as to realize the diagnosis and evaluation of ASD, improve the accuracy of diagnosis and evaluation, increase the support of objective biomarkers such as standardized medical images or biological molecules, strengthen the coverage of objects of different severity and early childhood, reduce the influence of subjectivity, and reduce the diagnosis and evaluation time.

[0082] Reference below Figure 8 and Fig. 9 The data processing method according to the embodiment of the present invention is described in detail.

[0083] like Figure 8 FIG. 1 is a flow chart of a data processing method according to another embodiment of the present invention. The data processing method according to the embodiment of the present invention at least includes steps S1 to S5.

[0084] Step S1, acquiring scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data.

[0085] Step S2, preprocessing the structural MRI data, functional MRI data and diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values.

[0086] Step S3, performing statistical analysis on the gray matter volume values, functional connectivity values, and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values.

[0087] Step S4, determining the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value according to the preset regression model, the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value.

[0088] Step S5, determining the data evaluation result and the evaluation result level according to the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value and the machine learning model.

[0089] like Fig. 9 FIG. 1 is a detailed flow chart of a data processing method according to another embodiment of the present invention, which is a refinement of the data processing method according to the above embodiment. The detailed flow chart of the data processing method according to the embodiment of the present invention at least includes steps S50 to S72.

[0090] Step S50, obtaining a first parameter setting of structural magnetic resonance imaging, a second parameter setting of functional magnetic resonance imaging, and a third parameter setting of diffusion tensor imaging.

[0091] Step S51 : acquiring structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data according to the first parameter setting, the second parameter setting and the third parameter setting.

[0092] Step S52: removing artifact data and head motion data from the structural magnetic resonance data to obtain structural magnetic resonance data to be processed.

[0093] Step S53: determining a tissue probability map of a set age group according to the magnetic resonance data of the structure to be processed.

[0094] Step S54: performing tissue segmentation on the structural magnetic resonance data to be processed according to the tissue probability map to determine a tissue segmentation image.

[0095] Step S55, normalizing the tissue segmentation image to a standard coordinate space to obtain an initial gray matter volume value.

[0096] Step S56, smoothing the initial gray matter volume value based on a Gaussian smoothing kernel to obtain a gray matter volume value.

[0097] Step S57, removing a preset preceding frame of the functional magnetic resonance data to obtain the functional magnetic resonance data to be processed.

[0098] Step S58, performing slice time correction and head motion calibration on the functional magnetic resonance data to be processed to obtain corrected functional magnetic resonance data.

[0099] Step S59, detecting and removing abnormal points of the corrected functional MRI data to obtain target functional MRI data.

[0100] Step S60 , normalizing the target functional MRI data to a standard coordinate space, and performing Gaussian smoothing on the target functional MRI data according to a Gaussian kernel to obtain smoothed functional MRI data.

[0101] Step S61, de-linearizing, removing covariates and band-pass filtering the smoothed functional magnetic resonance data to obtain target pre-processed functional magnetic resonance data.

[0102] Step S62, extracting the time series of the preprocessed functional magnetic resonance data according to the preset atlas, and calculating the Pearson correlation coefficient between each region in the preset atlas.

[0103] Step S63, determining the functional connectivity value according to the Pearson correlation coefficient.

[0104] Step S64: performing head motion and eddy current distortion correction on the diffusion tensor imaging data to obtain corrected diffusion tensor imaging data.

[0105] Step S65: All corrected diffusion tensor imaging data are registered and transformed into a standard space to obtain standardized tensor imaging data.

[0106] Step S66, creating an average tensor map based on the standardized tensor imaging data, and creating an average white matter skeleton.

[0107] Step S67, performing permutation-based nonparametric statistics according to the randomization tool to obtain statistically significant white matter microstructural differences and extract white matter microstructural values.

[0108] Step S68, determining the level grouping.

[0109] Step S69, grouping the gray matter volume values, functional connectivity values, and white matter microstructure values ​​in order to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values.

[0110] Step S70, inputting the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value into a preset regression model to correspondingly extract representative difference gray matter volume value, representative difference functional connectivity value and representative difference white matter microstructure value.

[0111] Step S71, determining the data diagnosis result according to the representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value and the machine learning model.

[0112] Step S72, obtaining the characteristic value input by the user, and outputting the data diagnosis result according to the characteristic value, wherein the data diagnosis result includes the diagnosis classification and the degree prediction value corresponding to the diagnosis classification.

[0113] According to the data processing method of the embodiment of the present invention, structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data are acquired through a data acquisition module, the data processing module preprocesses the above three types of data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values, the analysis and statistics module performs statistical analysis on the three types of preprocessed data to obtain difference gray matter volume values, difference functional connectivity values ​​and difference white matter microstructure values, the first determination module determines the most representative difference gray matter volume values, representative difference functional connectivity values ​​and representative difference white matter microstructure values ​​according to a preset regression model and the three statistically analyzed values, the second determination module determines the data evaluation results and the evaluation result level according to the three most representative values ​​and the machine learning model, so as to realize the diagnosis and evaluation of ASD, improve accuracy, increase the support of objective biomarkers such as standardized medical images or biological molecules, strengthen the coverage of severity and early childhood objects, reduce subjective influence, reduce the diagnosis and evaluation time, etc.

[0114] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example.

[0115] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A data processing device, characterized in that: include: A data acquisition module, used to acquire scanned structural magnetic resonance data, functional magnetic resonance data and diffusion tensor imaging data; A data processing module, used to preprocess the structural magnetic resonance data, the functional magnetic resonance data and the diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values; An analysis and statistics module is used to perform statistical analysis on the gray matter volume value, the functional connectivity value and the white matter microstructure value to obtain a difference gray matter volume value, a difference functional connectivity value and a difference white matter microstructure value; A first determination module is used to determine the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value according to a preset regression model, the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value; The second determination module is used to determine the data evaluation results and the evaluation result level according to the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value and the machine learning model.

2. The data processing device according to claim 1, characterized in that: The data acquisition module is specifically used for: acquiring a first parameter setting for structural magnetic resonance imaging, a second parameter setting for functional magnetic resonance imaging, and a third parameter setting for diffusion tensor imaging; The structural magnetic resonance data, the functional magnetic resonance data and the diffusion tensor imaging data are correspondingly acquired according to the first parameter setting, the second parameter setting and the third parameter setting.

3. The data processing device according to claim 1, characterized in that: The data processing module is specifically used for: removing artifact data and head motion data from the structural magnetic resonance data to obtain structural magnetic resonance data to be processed; Determine a tissue probability map of a set age group according to the magnetic resonance data of the structure to be processed; Performing tissue segmentation on the magnetic resonance data of the structure to be processed according to the tissue probability map to determine a tissue segmentation image; Normalizing the tissue segmentation image to a standard coordinate space to obtain an initial gray matter volume value; The initial gray matter volume value is smoothed based on a Gaussian smoothing kernel to obtain the gray matter volume value.

4. The data processing device according to claim 1, characterized in that: The data processing module is specifically used for: Removing a preset preceding frame of the functional magnetic resonance data to obtain functional magnetic resonance data to be processed; Performing slice time correction and head motion calibration on the functional magnetic resonance data to be processed to obtain corrected functional magnetic resonance data; Detecting and removing abnormal points of the corrected functional magnetic resonance data to obtain target functional magnetic resonance data; Normalizing the target functional magnetic resonance data to a standard coordinate space, and performing Gaussian smoothing on the target functional magnetic resonance data according to a Gaussian kernel to obtain smoothed functional magnetic resonance data; Performing de-linear trending, covariate removal and bandpass filtering processing on the smoothed functional magnetic resonance data to obtain target pre-processed functional magnetic resonance data; Extracting the time series of the preprocessed functional magnetic resonance data according to a preset atlas, and calculating the Pearson correlation coefficient between each region in the preset atlas; The functional connectivity value is determined based on the Pearson correlation coefficient.

5. The data processing device according to claim 1, characterized in that: The data processing module is specifically used for: Performing head motion and eddy current distortion correction on the diffusion tensor imaging data to obtain corrected diffusion tensor imaging data; Registering and transforming all the corrected diffusion tensor imaging data into a standard space to obtain standardized tensor imaging data; creating an average tensor map based on the normalized tensor imaging data, and creating an average white matter skeleton; Permutation-based nonparametric statistics were performed according to the randomization tool to obtain statistically significant white matter microstructural differences, and white matter microstructural values ​​were extracted.

6. The data processing device according to claim 1, characterized in that: The analysis and statistics module is specifically used for: Determine the level of grouping; The gray matter volume values, functional connectivity values ​​and white matter microstructure values ​​are grouped according to the degree of difference to obtain differential gray matter volume values, differential functional connectivity values ​​and differential white matter microstructure values.

7. The data processing device according to claim 1, characterized in that: The first determining module is specifically used for: The different gray matter volume values, the different functional connectivity values, and the different white matter microstructure values ​​are input into the preset regression model to correspondingly extract the representative different gray matter volume values, the representative different functional connectivity values, and the representative different white matter microstructure values.

8. The data processing device according to claim 1, characterized in that: The data processing device further includes a data output module, which is specifically used to: Acquire the feature value input by the user, and output the data evaluation result according to the feature value, wherein the data evaluation result includes a data evaluation classification and an evaluation level prediction value corresponding to the data evaluation classification.

9. The data processing device according to claim 1, characterized in that: The data processing device further comprises: a user interaction interface, wherein the user interaction interface is used to receive the magnetic resonance data characteristic value input by the user and output the data evaluation result and the evaluation result grade.

10. A data processing method, characterized in that: Used in a data processing device according to any one of claims 1 to 9, the method comprising: Acquire scanned structural MRI data, functional MRI data, and diffusion tensor imaging data; Preprocessing the structural magnetic resonance data, the functional magnetic resonance data and the diffusion tensor imaging data respectively to obtain gray matter volume values, functional connectivity values ​​and white matter microstructure values; Performing statistical analysis on the gray matter volume values, functional connectivity values, and white matter microstructure values ​​to obtain differential gray matter volume values, differential functional connectivity values, and differential white matter microstructure values; Determine the most representative difference gray matter volume value, the representative difference functional connectivity value and the representative difference white matter microstructure value according to the preset regression model, the difference gray matter volume value, the difference functional connectivity value and the difference white matter microstructure value; The data evaluation results and the evaluation result levels are determined based on the most representative difference gray matter volume values, the representative difference functional connectivity values, the representative difference white matter microstructure values ​​and the machine learning model.

Images