A magnetic resonance imaging method for evaluating the efficacy of rehabilitation for children with cerebral palsy

Abstract

This invention discloses a magnetic resonance imaging (MRI) method for evaluating the therapeutic effect of cerebral palsy rehabilitation in children, relating to the field of medical image processing technology. The method includes: acquiring structural imaging data (T1-weighted imaging and diffusion tensor imaging) and resting-state imaging data for each child patient; extracting brain and tissue boundaries from T1-weighted imaging using a 3D-Unet segmentation model; preprocessing the resting-state and structural imaging data and aligning them to brain boundaries, then registering these data to a standard brain template for children; acquiring and normalizing multiple clinical indicators based on the standard brain template, splicing them into multi-index vector data, inputting them into a pre-trained support vector machine model, and outputting differences in clinical indicators; determining abnormal brain regions for each child based on the differences, and finally generating a multimodal fusion index analysis report to observe changes in brain function and white matter integrity before and after rehabilitation.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a magnetic resonance imaging method for evaluating the therapeutic effect of cerebral palsy rehabilitation in children. Background Technology

[0002] Cerebral palsy is a prevalent birth defect, ranking first among childhood disabilities and placing a heavy burden on affected children's families and society. Rehabilitation for children with cerebral palsy has always focused on improving motor function, with the core being the reshaping of the neural circuits corresponding to motor function in the child's brain through rehabilitation therapy. Currently, the assessment of rehabilitation efficacy for children with cerebral palsy mainly relies on clinical observation and neuropsychological testing. While these methods can provide some information on rehabilitation effects, they typically require a long time to observe significant changes, and the results are easily influenced by the child's condition and the assessor's experience, thus possessing a degree of subjectivity.

[0003] In current technologies, researchers have been exploring more precise and sensitive methods to assess the effectiveness of cerebral palsy rehabilitation. With the development of imaging technology, the application of advanced techniques such as resting-state functional magnetic resonance imaging (fMRI) and diffusion imaging helps to assess the dynamic changes in neuronal activity and white matter integrity, and has been widely used in research on other neurological diseases. Resting-state fMRI reflects changes in brain activation patterns by measuring changes in hemodynamics caused by neuronal activity; diffusion imaging effectively assesses changes in the integrity of white matter pathways by measuring the diffusion motion of water molecules, providing evidence for rehabilitation treatment.

[0004] However, most existing studies on resting-state functional magnetic resonance imaging and diffusion imaging are group studies using conventional techniques, which do not fully consider the actual level of individual pediatric patients, resulting in an incomplete understanding of pediatric brain development and pathological changes. Summary of the Invention

[0005] Therefore, it is necessary to provide a magnetic resonance imaging method for evaluating the rehabilitation efficacy of children with cerebral palsy, in order to address the problem that existing technologies do not fully consider the actual level of individual children and thus lead to an incomplete understanding of children's brain development and pathological changes.

[0006] The present invention adopts the following technical solution:

[0007] This invention provides a magnetic resonance imaging method for evaluating the therapeutic effect of cerebral palsy rehabilitation in children, comprising:

[0008] Obtain structural and resting-state imaging data for the individual pediatric patients to be evaluated;

[0009] Resting-state and structural image data were preprocessed, and the preprocessed resting-state and structural image data were registered to standard brain templates for children.

[0010] Multiple clinical indicators were obtained and normalized from the registered resting-state imaging data and structural imaging data in the standard brain template of children. The normalized clinical indicators were then concatenated into multi-index vector data and input into a pre-trained support vector machine model to output the differences in clinical indicators. Based on the differences in clinical indicators, the abnormal brain regions of the individual children to be evaluated were obtained.

[0011] A multimodal fusion index analysis report was generated to determine the rehabilitation efficacy of children with cerebral palsy based on the abnormal brain regions of the individual patients to be evaluated.

[0012] Preferably, the resting-state image data is preprocessed, specifically including:

[0013] Convert resting-state image data from DICOM format to NIFTI format;

[0014] Temporal correction was performed on the NIFTI format resting-state image data; and head motion correction was performed on the corrected resting-state image data.

[0015] The aligned resting-state image data was smoothed using a Gaussian smoothing kernel.

[0016] Remove head motion-related artifacts from smoothed resting-state image data;

[0017] Time filtering was used to remove physiological signal noise from the resting-state image data after artifact removal, resulting in preprocessed resting-state image data.

[0018] Preferably, the structural image data includes: diffusion imaging data and T1-weighted imaging data;

[0019] Preprocessing of diffusion imaging data specifically includes:

[0020] Convert resting-state image data from DICOM format to NIFTI format;

[0021] Slice outlier detection is performed on the diffusion imaging data, and Gibbs-ringing artifacts and noise are removed from the diffusion imaging data to obtain denoised diffusion imaging data;

[0022] The undistorted undiffusion-weighted image is synthesized from the denoised diffusion imaging data and combined with the original undiffusion-weighted image. Then the diffusion imaging data is corrected.

[0023] Bias field correction is performed on the corrected diffusion imaging data;

[0024] A self-supervised learning algorithm is used to filter the diffusion imaging data after bias field correction to obtain preprocessed diffusion imaging data.

[0025] Preprocessing of TI-weighted imaging data includes:

[0026] Resample the T1-weighted imaging data to a specified size space;

[0027] Using a trained 3DUnet segmentation model, brain tissue of the pediatric patients to be evaluated was extracted from space;

[0028] The brain tissue was then resampled to a space of another specified size.

[0029] Preferably, the clinical indicators include functional indicators and structural indicators; multiple clinical indicators are obtained from the registered resting-state imaging data and structural imaging data of the standard pediatric brain template, specifically including:

[0030] Based on registered resting-state image data in standard brain templates for children, functional indicators were calculated, including the whole-brain functional connectivity matrix, regional homogeneity and its average value, and low-frequency signal amplitude and its average value.

[0031] Based on the registered structural image data in the standard brain template of children, structural indices were calculated, including the whole brain structural connectivity matrix, anisotropy score and its mean, and average diffusion rate and its mean.

[0032] Preferably, the whole-brain functional connectivity matrix is ​​constructed by dividing a standard brain template of a child into multiple regions of interest, calculating the functional connectivity strength between each region of interest, and constructing the matrix based on the functional connectivity strength between the regions of interest.

[0033] Regional homogeneity was obtained by calculating the coordination coefficient between each voxel and its neighboring voxels in a standard brain template of children.

[0034] The low-frequency signal amplitude was converted into a frequency domain signal by Fourier transform of the registered resting-state image data, and the power spectral density of each voxel in the low-frequency band of the frequency domain signal in the standard brain template of children was calculated.

[0035] Preferably, the whole-brain structural connectivity matrix is ​​constructed by reconstructing white matter fiber bundles from registered structural image data using fiber tracking technology, calculating the connectivity strength of each fiber bundle, and constructing the matrix based on the connectivity strength of each fiber bundle.

[0036] Preferably, the multimodal fusion index analysis report for determining the rehabilitation efficacy of children with cerebral palsy specifically includes:

[0037] Multimodal fusion indices were extracted from the abnormal brain regions of individual children to be evaluated. The multimodal fusion indices included functional connectivity of the abnormal brain regions, average regional homogeneity of the abnormal brain regions, average low-frequency signal amplitude of the abnormal brain regions, structural connectivity of the abnormal brain regions, average anisotropy fraction of the abnormal brain regions, and average diffusion rate of the abnormal brain regions.

[0038] Based on the multimodal fusion index, an analysis report of the multimodal fusion index is obtained through automated software.

[0039] The above-mentioned at least one technical solution adopted in this invention can achieve the following beneficial effects:

[0040] In a magnetic resonance imaging (MRI) method for evaluating the rehabilitation efficacy of cerebral palsy in children provided by this invention, resting-state imaging data and diffusion imaging data are preprocessed and aligned to the brain boundaries of the child's structural images. The preprocessed MRI data and the child's T1-weighted brain imaging data are registered to a standard brain template using the Riemann manifold registration method. Based on the standard brain template, clinical indicators of the registered MRI data are calculated. Based on different anatomical regions of the template, tools such as FSL are used to calculate the registered clinical indicators. Structural and functional indicators are processed through canonical correlation analysis, and the relevant components of structural and functional indicators are output and normalized. The normalized indicators are concatenated into multi-index vector data and input into a pre-trained support vector machine model for indicator analysis. Abnormal brain regions of each child patient are obtained based on the differences in indicators. Based on the abnormal brain regions of each child patient, a multimodal fusion indicator analysis report on the rehabilitation efficacy of cerebral palsy is generated to observe the changes in brain function and white matter integrity of each child patient before and after rehabilitation.

[0041] This invention preprocesses the resting-state imaging and diffusion imaging data of each child and uses the Riemannian manifold brain registration algorithm to register the child's individual MRI data with a standard brain template. Through the calculation of clinical indicators and canonical correlation analysis, and using a support vector machine model for data analysis, it considers individual differences in the patient's brain structure, making the post-processing of MRI data simpler. It enables data analysis based on the specific situation of each patient, improving the accuracy of diagnosis and the targeted nature of rehabilitation efficacy assessment. It achieves precise image analysis for each individual child patient, avoiding the problem of averaging individual differences in group studies, ensuring that the assessment results are closer to the true condition of each child with cerebral palsy. This helps clinicians optimize treatment plans, achieve precise and personalized rehabilitation treatment, improve the symptom relief rate of children with cerebral palsy, and effectively reduce the economic burden on the children's families and society. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0043] Figure 1 A schematic flowchart of a magnetic resonance imaging method for evaluating the therapeutic effect of cerebral palsy rehabilitation in children, provided by the present invention;

[0044] Figure 2 T1-weighted imaging, diffusion tensor imaging, and resting-state functional imaging data of a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, provided by this invention;

[0045] Figure 3 The image shows the segmentation and resampling effect of a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, provided by this invention.

[0046] Figure 4 This invention demonstrates the effect of registering resting-state imaging data to a standard brain template of children using a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy.

[0047] Figure 5 The image shows the effect of T1WI registration to a standard brain template of a child using a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, as provided by this invention.

[0048] Figure 6 The image shows the effect of T1WI registration to a standard brain template of a child using a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, as provided by this invention.

[0049] Figure 7 A flowchart of the calculation of functional MRI indicators for a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, provided by the present invention;

[0050] Figure 8 Whole-brain functional connectivity matrix diagram of a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, provided by the present invention;

[0051] Figure 9 The diffusion imaging data of the magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children provided by the present invention calculates FA and MD parameter maps.

[0052] Figure 10 A flowchart of structural index calculation for a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children, provided by the present invention;

[0053] Figure 11 A whole-brain structural connectivity matrix diagram for a magnetic resonance imaging method to evaluate the rehabilitation efficacy of cerebral palsy in children, provided by the present invention;

[0054] Figure 12 The automated software for the magnetic resonance imaging method for evaluating the rehabilitation efficacy of children with cerebral palsy provided by this invention generates an analysis report graph of the rehabilitation efficacy of children with cerebral palsy. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in the specification without creative effort are within the scope of protection of this application.

[0056] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0057] Figure 1 This is a schematic diagram of a magnetic resonance imaging method for evaluating the rehabilitation efficacy of cerebral palsy in children according to the present invention, which specifically includes the following steps:

[0058] S101: Obtain structural and resting-state imaging data of the individual pediatric patient to be evaluated.

[0059] Specifically, scans were performed on at least 3T MRI scanners to acquire high-quality structural imaging data of children, including T1-weighted imaging (T1WI), diffusion tensor imaging (DTI), and resting-state functional MRI sequences with a scan time of at least 8 minutes. T1WI slice thickness was ≤1 mm. Diffusion tensor imaging sequences were used to assess white matter integrity, and resting-state functional MRI sequences were used to assess changes in brain functional activity and connectivity. For scan parameters and some images obtained after the scans, please refer to [link to relevant documentation]. Figure 2 Table 1 shows the T1-weighted imaging, diffusion tensor imaging, and resting-state functional imaging data, along with a list of scanning parameters.

[0060] Table 1. List of Scanning Parameters

[0061]

[0062] Specifically, this invention follows the diagnostic recommendations of senior radiologists, and all data were interpreted by two physicians with 10 years of experience in brain MRI diagnosis. The example used is a 6-year-old boy clinically diagnosed with spastic diplegia. Table 2 shows the clinical data of the MRI dataset of children with cerebral palsy in this invention. The first column represents the serial number, the second column represents the age, the third column represents the gender, the fourth column represents the clinical diagnosis, the fifth column represents the gross motor score, and the sixth column represents the hand function score.

[0063] Table 2 Clinical data from MRI datasets of children with cerebral palsy

[0064]

[0065] S102: Preprocess the resting-state image data and structural image data; and register the preprocessed resting-state image data and structural image data to the standard brain template for children.

[0066] Optionally, the resting-state image data is preprocessed, specifically including: converting the resting-state image data from DICOM format to NIFTI format; performing temporal correction on the NIFTI format resting-state image data; performing head motion correction on the corrected resting-state image data; smoothing the aligned resting-state image data using a Gaussian smoothing kernel; removing head motion-related artifacts from the smoothed resting-state image data; and using temporal filtering to remove physiological signal noise from the artifact-removed resting-state image data to obtain preprocessed resting-state image data.

[0067] Specifically, the resting-state imaging data was converted from DICOM format to NIFTI format using dcm2niix software (version v1.0.20241211); temporal correction was performed on the NIFTI format resting-state imaging data using the nipy.algorithms.slicetiming.timefuncs function of fmriprep software, with correction parameters determined based on the individual resting-state image sliceetiming and TR; the resting-state imaging data was smoothed using a [6*6*6]mm Gaussian smoothing kernel of fmriprep software; artifacts related to head movement were removed from the smoothed resting-state functional magnetic resonance imaging sequence data using the ICA-AROMA algorithm; physiological signal noise was removed using temporal filtering to obtain preprocessed resting-state imaging data; and the preprocessed resting-state imaging data was aligned to the child's brain boundary using the Riemannian manifold brain registration algorithm.

[0068] Optionally, the diffusion imaging data is preprocessed, specifically including: performing slice outlier detection on the diffusion imaging data to remove Gibbs-ringing artifacts; removing noise from the diffusion imaging data; for the noise-removed diffusion imaging data, synthesizing an undistorted undiffusion-weighted image from the structural image and combining it with the original undiffusion-weighted image, and then correcting the diffusion imaging data; performing bias field correction on the corrected diffusion imaging data; and using a self-supervised learning algorithm to filter the bias field-corrected diffusion imaging data to obtain the preprocessed diffusion imaging data.

[0069] Specifically, the data format of the diffusion imaging sequence was converted from DICOM to NIFTI format; outlier detection was performed on the NIFTI format diffusion imaging data, and Gibbs-ringing artifacts were removed using mrdegibbs from the mrtrix3 toolkit; dwidenoise was used to remove noise from the diffusion imaging data; for the diffusion imaging data, the Synb0-DisCo algorithm was used to synthesize an undistorted undiffusion-weighted image from the structural image and combine it with the original b0 image, and then the diffusion imaging data was corrected using dwifslpreproc in mrtrix3 software to remove head motion, eddy current correction, and distortion artifacts caused by magnetic susceptibility; the corrected diffusion imaging data was then subjected to bias field correction; the self-supervised learning algorithm patch2self was used to filter the bias field-corrected diffusion imaging data to remove noise, resulting in preprocessed diffusion imaging data;

[0070] The Riemannian manifold brain registration algorithm was used to register preprocessed pediatric brain structural data (T1WI) and resting-state functional MRI data to a standard pediatric brain template. The results of registering resting-state imaging data to the standard pediatric brain template can be found in [link to documentation]. Figure 4 .

[0071] Optionally, the T1-weighted imaging data is preprocessed, including: resampling the T1-weighted imaging data to a space of a specified size; extracting the brain tissue of the individual child patient to be evaluated from the space using a trained 3DUnet segmentation model; and resampling the brain tissue to another space of a specified size.

[0072] Specifically, see Figure 3 The image shows the segmentation and resampling results of T1-weighted imaging data. The preprocessing of T1-weighted imaging data includes: resampling the T1-weighted imaging data to a space of 1mm×1mm×1m and 256×256×256 layers, and then using a 3D-Unet segmentation model trained on children's structural images to extract children's brain tissue from T1WI images and resample it to a space of 3mm×3mm×3mm.

[0073] The preprocessed diffusion imaging data and T1-weighted imaging data were registered to a standard pediatric brain template using the Riemannian manifold brain registration algorithm. See the results for the image. Figure 5 and Figure 6 .

[0074] S103: Obtain and normalize multiple clinical indicators from the registered resting-state imaging data and structural imaging data in the standard brain template for children; concatenate the normalized multiple clinical indicators into multi-index vector data and input them into a pre-trained support vector machine model to output the differences in clinical indicators; based on the differences in clinical indicators, obtain the abnormal brain regions of the individual children to be evaluated.

[0075] Optionally, the clinical indicators include functional indicators and structural indicators; the acquisition of multiple clinical indicators from registered resting-state imaging data and diffusion imaging data in the standard brain template of children specifically includes: calculating functional indicators based on registered resting-state imaging data in the standard brain template of children, including the whole-brain functional connectivity matrix, regional homogeneity and its average value, and low-frequency signal amplitude and its average value; and calculating structural indicators based on registered diffusion imaging data in the standard brain template of children, including the whole-brain structural connectivity matrix, anisotropy score and its average value, and average diffusion rate and its average value.

[0076] Optionally, the whole-brain functional connectivity matrix is ​​constructed by dividing a standard brain template for children into multiple regions of interest, calculating the functional connectivity strength between each region of interest, and then calculating the functional connectivity strength between the regions of interest. The region homogeneity is obtained by calculating the coordination coefficient between each voxel in the standard brain template for children and its neighboring voxels. The low-frequency signal amplitude is obtained by converting the registered resting-state image data into a frequency domain signal through Fourier transform, and calculating the power spectral density of each voxel in the low-frequency band of the frequency domain signal in the standard brain template for children.

[0077] Among these, the whole-brain functional connectivity matrix, regional homogeneity, and low-frequency signal amplitude are calculated based on the registered resting-state image data. The calculation process is described in [link to calculation procedure]. Figure 7 For the functional MRI connectivity matrix diagram, see [link to diagram]. Figure 8 .

[0078] Specifically, based on canonical correlation analysis, the registered resting-state magnetic resonance imaging (MRI) data were used to calculate the resting-state MRI indices (low-frequency signal amplitude and regional homogeneity). After standard normalization, the average brain region values ​​of regional homogeneity and low-frequency signal amplitude were calculated based on the brain template. The results of the fusion analysis of the unnormalized resting-state MRI indices are shown in Table 3.

[0079] Table 3. Results of Unnormalized Resting-State MRI Index Fusion Analysis

[0080]

[0081] Based on canonical correlation analysis, clinical indicators FA and MD of diffusion tensor imaging data were calculated using preprocessed diffusion imaging data. After standard normalization, the brain region averages of FA and MD were calculated based on brain templates, as shown in Table 4.

[0082] Table 4 Results of index fusion analysis of unnormalized diffusion tensor sequences

[0083]

[0084] The brain region averages of resting-state index region homogeneity, low-frequency signal amplitude, FA, and MD are combined with the whole-brain functional connectivity matrix and the whole-brain structural connectivity matrix to form a multi-index vector, and a pre-trained support vector machine is used to identify abnormal brain regions.

[0085] Using a dataset of children with cerebral palsy and a dataset of normal children that have undergone the same preprocessing steps, a support vector machine model trained on a multi-indexed vector dataset was built after calculating resting-state indices and DTI indices. The multi-indexed vector fused from the patients' multimodal imaging indices was used as the original input matrix, and statistically significant features in the algorithm output were selected as abnormal brain regions.

[0086] Specifically, the FA parameter map and MD parameter map are calculated from the preprocessed diffusion imaging data. See [link to relevant documentation]. Figure 9 Fiber orientation density (FOD) was calculated using the CSD (spherical deconvolution) method. Based on gray and white matter tissues extracted using the 3D-Unet segmentation model, whole-brain fiber tracts were deterministically traced, and the whole-brain connectivity matrix was calculated. The calculation process for structural MRI metrics is described in [link to documentation]. Figure 10 The whole-brain structural connectivity matrix diagram can be found here. Figure 11 .

[0087] S104: A multimodal fusion index analysis report on the efficacy of cerebral palsy rehabilitation in children, based on the abnormal brain regions of the individual children to be evaluated.

[0088] Optionally, a multimodal fusion index analysis report for determining the efficacy of rehabilitation for children with cerebral palsy may be included, specifically:

[0089] Multimodal fusion indices were extracted from abnormal brain regions of individual children to be evaluated. These indices included functional connectivity, average regional homogeneity, average low-frequency signal amplitude, structural connectivity, average anisotropy fraction, and average diffusion rate of the abnormal brain regions. Based on these indices, an automated software-generated multimodal fusion index analysis report was generated. (See attached document). Figure 12 This is a schematic diagram of a multimodal fusion index analysis report.

[0090] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.

Claims

1. A magnetic resonance method for evaluating the efficacy of rehabilitation of cerebral palsy in children, characterized in that, The method comprises the following steps: obtaining structural image data and resting state image data of an individual child patient to be evaluated; the structural image data comprises diffusion imaging data and T1-weighted imaging data; converting the resting state image data from DICOM format to NIFTI format; performing time layer correction on the NIFTI format resting state image data; and performing head motion correction on the corrected resting state image data; performing smoothing processing on the aligned resting state image data using a Gaussian smoothing kernel; removing artifacts related to head motion from the smoothed resting state image data; filtering out physiological signal noise from the resting state image data after removing artifacts using time filtering to obtain preprocessed resting state image data; converting the diffusion imaging data from DICOM format to NIFTI format; performing slice outlier detection on the diffusion imaging data, and removing Gibbs-ringing artifacts and noise from the diffusion imaging data to obtain denoised diffusion imaging data; combining the non-diffusion weighted image synthesized from the denoised diffusion imaging data with the original non-diffusion weighted image, and then correcting the diffusion imaging data; performing bias field correction on the corrected diffusion imaging data; filtering the diffusion imaging data after bias field correction using a self-supervised learning algorithm to obtain preprocessed diffusion imaging data; resampling the T1-weighted imaging data to a specified size of space; extracting the brain tissue of the individual child patient to be evaluated from the space using a trained 3DUnet segmentation model; and resampling the brain tissue to another specified size of space; and using a Riemannian manifold brain registration algorithm to register the preprocessed resting state image data and the structural image data to a child standard brain template, respectively; obtaining and normalizing a plurality of clinical indicators of the registered resting state image data and the structural image data in the child standard brain template, respectively, including: calculating functional indicators based on the registered resting state image data in the child standard brain template, the functional indicators including a whole brain functional connectivity matrix, regional homogeneity and its average value, and low frequency signal amplitude and its average value; calculating structural indicators based on the registered structural image data in the child standard brain template, the structural indicators including a whole brain structural connectivity matrix, anisotropy fraction and its average value, and mean diffusion rate and its average value; the clinical indicators include the functional indicators and the structural indicators; the whole brain functional connectivity matrix is constructed by dividing the child standard brain template into a plurality of regions of interest, calculating the functional connection strength between each region of interest, and according to the functional connection strength between the regions of interest; the regional homogeneity is obtained by calculating the coordination coefficient between each voxel and its adjacent voxel in the child standard brain template; the low frequency signal amplitude is obtained by converting the registered resting state image data into frequency domain signal through Fourier transform, and calculating the power spectral density of each voxel with low frequency in the frequency domain signal of the child standard brain template; the whole brain structural connectivity matrix is constructed by reconstructing the white matter fiber bundle of the registered structural image data through fiber tracking technology, and calculating the connection strength of each fiber bundle; The normalized multiple clinical indicators are spliced into multiple index vector data and input into a pre-trained support vector machine model to output a clinical indicator difference; and according to the clinical indicator difference, an abnormal brain area of the individual child patient to be evaluated is obtained; A multi-modal fusion index of the abnormal brain area of the individual child patient to be evaluated is extracted; and according to the multi-modal fusion index, a multi-modal fusion index analysis report of the rehabilitation effect of the child cerebral palsy is determined; the multi-modal fusion index includes abnormal brain area functional connection, abnormal brain area average regional homogeneity, abnormal brain area average low-frequency signal amplitude, abnormal brain area structural connection, abnormal brain area average fractional anisotropy and abnormal brain area average diffusion rate.

Images