A rehabilitation degree evaluation system and method for neurology rehabilitation

Abstract

The application provides a rehabilitation degree evaluation system and method for neurology rehabilitation, and belongs to the technical field of rehabilitation evaluation. Functional magnetic resonance imaging data and diffusion tensor imaging data of rehabilitation personnel at multiple rehabilitation evaluation time points are collected to construct multi-modal brain network data; sparse coding and time consistency constraints are performed by using the modal scale parameters and the cross-modal shared sparse dictionary determined by a historical training sample set to obtain enhanced brain network data; neighborhood structure preserving dimension reduction is performed on the enhanced brain network data, representative features of each rehabilitation evaluation time point are connected in time sequence to obtain a rehabilitation progress state trajectory; evaluation weights are determined according to quality parameters of the functional magnetic resonance imaging data and the diffusion tensor imaging data to form a quality weighted state trajectory, and a rehabilitation degree score is determined accordingly. By using the above scheme, a stable quantitative basis can be provided for rehabilitation degree evaluation reflecting multi-stage brain network changes.

Description

Technical Field

[0001] This application relates to the field of rehabilitation assessment technology, and more specifically, to a rehabilitation assessment system and method for neurological rehabilitation. Background Technology

[0002] Rehabilitation evaluation for neurological diseases is a crucial aspect of clinical diagnosis and treatment. Currently, clinical practice largely relies on subjective assessments using scales, which struggles to objectively and accurately capture the brain network remodeling process during the rehabilitation period. Quantifying the dynamic changes in brain networks can provide objective auxiliary evidence for adjusting rehabilitation plans and evaluating treatment effectiveness. With the development of neuroimaging technology, functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) can characterize brain network features from two dimensions: functional activity and white matter structure, respectively, providing a feasible technical approach for longitudinal rehabilitation assessment.

[0003] However, most existing brain network analysis methods are designed for single-time-point and single-modal data, and cannot simultaneously track the coordinated changes in functional and structural connectivity during the rehabilitation process. At the same time, multi-time-point data lacks a unified processing benchmark, which can easily confuse data collection and scale differences with actual brain network changes, resulting in insufficient reliability of evaluation results. Therefore, there is currently a lack of rehabilitation assessment methods that can stably quantify changes in brain networks across multiple stages. Summary of the Invention

[0004] This application provides a rehabilitation assessment system and method for neurological rehabilitation, which can provide a stable quantitative basis for assessing rehabilitation progress that reflects changes in multi-stage brain networks.

[0005] Firstly, this application provides a method for evaluating the degree of rehabilitation in neurological rehabilitation, comprising the following steps: Brain imaging data were obtained by collecting resting-state functional magnetic resonance imaging (fMRI) data and diffusion tensor imaging data of rehabilitation participants at multiple rehabilitation assessment points during the rehabilitation cycle. The brain imaging data is divided into brain atlases and functional connections based on time windows to obtain multimodal brain network data at each rehabilitation assessment time point; The functional connectivity matrix and structural connectivity matrix in the multimodal brain network data are normalized according to the modal scaling parameters determined based on the historical training sample set. The cross-modal shared sparse dictionary constructed based on the historical training sample set is called to obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point. Temporal consistency constraints are applied to the functional sparse coding coefficient vector to obtain the enhanced brain network data. The enhanced brain network data is subjected to neighborhood structure preservation dimensionality reduction to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point. Based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, the corresponding representative features are determined, and the representative features are connected in the time order of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory. Based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point, the corresponding data quality confidence index is determined. Based on the data quality confidence index, the assessment weight at each rehabilitation assessment time point is determined. The assessment weight is associated with the rehabilitation process state trajectory to obtain a quality-weighted state trajectory. Based on the quality-weighted state trajectory, the rehabilitation degree score of the rehabilitation personnel is determined.

[0006] In some embodiments, performing time-window-based brain mapping and functional connectivity on the brain imaging data to obtain multimodal brain network data at each rehabilitation assessment time point specifically includes: The functional magnetic resonance imaging data in the brain imaging data at each rehabilitation assessment time point were divided into multiple time windows along the acquisition time axis to obtain multiple functional magnetic resonance signals. Brain regions are divided based on functional magnetic resonance imaging signals within each time window using brain atlas templates; Determine the Pearson correlation coefficients of each brain region within each time window, and construct the functional connectivity matrix corresponding to each time window; Brain regions were divided from diffusion tensor imaging data in brain imaging data at each rehabilitation assessment time point; Determine the connectivity strength of each brain region at each rehabilitation assessment time point, and construct the structural connectivity matrix corresponding to each rehabilitation assessment time point; By associating the structural connectivity matrix with all functional connectivity matrices at each rehabilitation assessment time point, multimodal brain network data for each rehabilitation assessment time point is obtained.

[0007] In some embodiments, the cross-modal shared sparse dictionary is pre-constructed in the following manner: Obtain the functional connectivity matrix and structural connectivity matrix at each rehabilitation assessment time point in the historical training sample set. Determine the functional modal scaling parameters and structural modal scaling parameters based on the training set statistics of the functional connectivity matrix and structural connectivity matrix, respectively, and perform normalization processing using the corresponding modal scaling parameters. The normalized functional connectivity matrix and structural connectivity matrix are used as input data for dictionary learning. The goal is to optimize the updating of dictionary atoms and sparse coding coefficients by minimizing the reconstruction error of the functional connectivity matrix and structural connectivity matrix under shared dictionary atoms. A cross-modal consistency constraint is applied to the mean functional connectivity matrix and structural connectivity matrix at the same rehabilitation assessment time point in the longitudinal training samples of the same rehabilitation patient. The optimization is terminated when the reconstruction error converges or reaches the preset number of iterations, thus obtaining the cross-modal shared sparse dictionary.

[0008] In some embodiments, applying a time consistency constraint to the functionally sparse coding coefficient vector to obtain enhanced brain network data specifically includes: Based on the cross-modal shared sparse dictionary, the normalized functional connectivity matrix of each time window within each rehabilitation assessment time point and the normalized structural connectivity matrix of each rehabilitation assessment time point are sparsely decomposed to obtain the original functional sparse coding coefficient vector corresponding to each time window and the structural sparse coding coefficient vector corresponding to each rehabilitation assessment time point. The original functional sparse coding coefficient vectors of each time window are set as the corresponding functional sparse coding coefficient vectors to be optimized. The mean coding to be optimized at each rehabilitation assessment time point is determined according to the functional sparse coding coefficient vectors to be optimized. The rate of change of the mean coding to be optimized between adjacent rehabilitation assessment time points is determined according to the time interval between adjacent rehabilitation assessment time points. While maintaining the data fit between the sparse coding coefficient vector to be optimized and the original sparse coding coefficient vector and preserving sparsity, a time consistency penalty is applied to the difference in the rate of change of adjacent mean coding to be optimized, and the sparse coding coefficient vectors to be optimized in each time window are jointly optimized to obtain the constrained sparse coding coefficient vector. According to the rehabilitation assessment time point and time window, the restricted functional sparse coding coefficient vector is associated with the corresponding structural sparse coding coefficient vector to obtain the enhanced brain network data.

[0009] In some embodiments, performing neighborhood structure preservation dimensionality reduction on the enhanced brain network data to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point specifically includes: Obtain a neighborhood structure-preserving autoencoder trained based on a historical training sample set; The constrained functional sparse coding coefficient vectors corresponding to each time window in the enhanced brain network data are concatenated with the structural sparse coding coefficient vectors corresponding to the rehabilitation assessment time points. The concatenated multimodal coding coefficient vectors are then input into the neighborhood structure-preserving autoencoder to encode the reconstruction error and the neighborhood preservation error to jointly constrain the dimensionality reduction process. Output the low-dimensional feature vectors corresponding to each time window, and use the low-dimensional feature vectors as the neighborhood structure preservation features of the corresponding rehabilitation assessment time points.

[0010] In some embodiments, determining corresponding representative features based on multiple neighborhood structure preservation features at each rehabilitation assessment time point, and connecting the representative features according to the chronological order of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory specifically includes: The mean of all neighborhood structures within each rehabilitation assessment time point is aggregated to obtain the representative features of the corresponding rehabilitation assessment time point. The representative features of each rehabilitation assessment time point are connected in a common low-dimensional feature space according to the time sequence. Calculate the Euclidean distance between representative features of any adjacent rehabilitation assessment time points, and determine the rehabilitation process state trajectory by combining the representative features of each rehabilitation assessment time point, the Euclidean distance between adjacent representative features, and the time interval between adjacent rehabilitation assessment time points.

[0011] In some embodiments, determining the corresponding data quality confidence index based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point specifically includes: The average frame shift and effective time point ratio were extracted from the functional magnetic resonance imaging data at each rehabilitation assessment time point, and the effective diffusion direction ratio, diffusion direction coverage and spatial registration error were extracted from the diffusion tensor imaging data. The average frame displacement and spatial registration error are compared with the corresponding upper limit thresholds determined based on the historical training sample set, and the effective time point ratio, effective diffusion direction ratio and diffusion direction coverage are compared with the corresponding lower limit thresholds determined based on the historical training sample set. For the average frame displacement and spatial registration error not exceeding the corresponding upper limit threshold, and the proportion of effective time points, the proportion of effective diffusion directions, and the coverage of diffusion directions not lower than the corresponding lower limit threshold, the directional interval normalization is performed in the same direction to obtain normalized quality parameters with values ​​ranging from zero to one that increase with the improvement of the corresponding data quality. The normalized quality parameters are weighted using non-negative combination coefficients that minimize the mean square error of the weighted result of the normalized quality parameters in the validation set of the historical training sample set relative to the image quality qualified labels with values ​​of zero or one. The sum of the non-negative combination coefficients is one, thus obtaining the data quality confidence index for the corresponding rehabilitation assessment time point.

[0012] In some embodiments, determining the assessment weight of each rehabilitation assessment time point based on the data quality confidence index, and associating the assessment weight with the rehabilitation process state trajectory to obtain a quality-weighted state trajectory specifically includes: Rehabilitation assessment time points whose average frame displacement or spatial registration error is higher than the corresponding upper limit threshold, or whose effective time point ratio, effective diffusion direction ratio, or diffusion direction coverage is lower than the corresponding lower limit threshold, are marked as invalid rehabilitation assessment time points. Rehabilitation assessment time points that are not marked as invalid rehabilitation assessment time points are determined as valid rehabilitation assessment time points. When there are invalid rehabilitation assessment time points or the number of valid rehabilitation assessment time points is less than the preset fixed number of assessment time points, a data shortage mark is generated and the determination of rehabilitation degree score is stopped. A preset minimum positive value is added to the data quality confidence index at each effective rehabilitation assessment time point. The corresponding assessment weight is determined based on the data quality confidence index after adding the preset minimum positive value, and all assessment weights are normalized. By keeping the representative feature coordinates, time sequence, and Euclidean distance between adjacent representative features of each effective rehabilitation assessment time point in the rehabilitation process state trajectory unchanged, the assessment weight of each effective rehabilitation assessment time point is associated with the corresponding representative feature to obtain the quality-weighted state trajectory.

[0013] In some embodiments, determining the rehabilitation level score of the rehabilitated person based on the quality-weighted state trajectory specifically includes: The representative features of each effective rehabilitation assessment time point, the Euclidean distance between adjacent representative features, the time interval between adjacent rehabilitation assessment time points, and the assessment weight of each effective rehabilitation assessment time point are extracted from the quality-weighted state trajectory and combined to obtain the trajectory features. The regression model is trained by inputting the trajectory features into the historical training sample set as training input and using the change in baseline to the last score after unifying the improvement direction according to the clinical rehabilitation evaluation scale as training label, and thus obtaining a rehabilitation degree score representing the predicted score improvement.

[0014] Secondly, this application provides a rehabilitation assessment system for neurological rehabilitation, comprising: The acquisition module is used to collect resting-state functional magnetic resonance imaging data and diffusion tensor imaging data of rehabilitation personnel at multiple rehabilitation assessment time points during the rehabilitation cycle to obtain brain imaging data. The processing module is used to perform brain atlas partitioning and functional connectivity based on time windows on the brain imaging data to obtain multimodal brain network data at each rehabilitation assessment time point; The processing module is further configured to normalize the functional connectivity matrix and structural connectivity matrix in the multimodal brain network data according to the modal scale parameters determined based on the historical training sample set, call the cross-modal shared sparse dictionary constructed based on the historical training sample set, obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point, and apply time consistency constraints to the functional sparse coding coefficient vector to obtain the enhanced brain network data. The processing module is also used to perform neighborhood structure preservation dimensionality reduction on the enhanced brain network data to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point, determine the corresponding representative features based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, and connect the representative features in the time order of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory. The execution module is used to determine the corresponding data quality confidence index based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point, determine the assessment weight at each rehabilitation assessment time point based on the data quality confidence index, associate the assessment weight with the rehabilitation process state trajectory to obtain the quality-weighted state trajectory, and determine the rehabilitation degree score of the rehabilitation personnel based on the quality-weighted state trajectory.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application processes the functional connectivity matrix and structural connectivity matrix separately according to modal scale parameters determined based on historical training sample sets, ensuring that brain network data from different rehabilitation patients and at different rehabilitation assessment time points are at a consistent data scale. A cross-modal shared sparse dictionary is used to form comparable sparse codes, and temporal consistency constraints limit isolated changes at adjacent assessment time points while retaining the fitting terms of the original coded data. Neighborhood structure-preserving features and Euclidean distances between adjacent representative features are used to form a rehabilitation process state trajectory. Quality parameters of functional magnetic resonance imaging data and diffusion tensor imaging data are used to determine the assessment weights associated with the original trajectory, and the coordinates, temporal order, and adjacent differences of the original representative features are all preserved. Therefore, this scheme can simultaneously utilize structural connectivity, functional connectivity, and multi-stage change information to provide a stable quantitative basis for evaluating the degree of rehabilitation that reflects multi-stage brain network changes. Attached Figure Description

[0016] Figure 1 This is an exemplary flowchart of a method for evaluating the degree of rehabilitation in neurological rehabilitation, according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the pre-construction of a cross-modal shared sparse dictionary according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the determination of the recovery process status trajectory according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a rehabilitation assessment system for neurological rehabilitation, as shown in some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing a method for evaluating the degree of rehabilitation in neurological rehabilitation, according to some embodiments of this application. Detailed Implementation

[0017] To better understand the above technical solutions, the following description, in conjunction with the accompanying drawings and specific embodiments, will illustrate the above technical solutions.

[0018] refer to Figure 1 , Figure 1 This is an exemplary flowchart of a method for evaluating the degree of rehabilitation in neurological rehabilitation, according to some embodiments of this application. The method includes the following steps: In step 101, resting-state functional magnetic resonance imaging (fMRI) data and diffusion tensor imaging (DTI) data of the rehabilitation personnel are collected at multiple rehabilitation assessment time points during the rehabilitation cycle to obtain brain imaging data.

[0019] In this application, the functional magnetic resonance imaging data are blood oxygen level-dependent signals continuously acquired in a resting state, the diffusion tensor imaging data are data acquired under different diffusion gradient directions and obtained by tensor fitting, and the brain imaging data are a set of data used to construct the functional connectivity matrix and structural connectivity matrix after the above correction and spatial registration.

[0020] In practice, rehabilitation assessment time points are determined according to the assessment schedule in the rehabilitation treatment plan. These time points may include the baseline, intermediate, and final stages of rehabilitation. At each assessment time point, resting-state functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) scans are performed. Resting-state fMRI continuously acquires oxygenation-dependent signals, while DTI scans acquire diffusion-weighted signals across multiple diffusion gradient directions. A consistent scanning protocol must be used for each assessment time point. FMRI data undergoes temporal correction, head motion correction, spatial registration, and bandpass filtering. DTI data undergoes eddy current correction, head motion correction, tensor fitting, and spatial registration. When head motion parameters, effective scan duration, or spatial registration quality do not meet preset quality conditions, the corresponding assessment time point is marked as having insufficient data, and the determination of the rehabilitation level score is stopped.

[0021] In step 102, the brain imaging data is partitioned and functionally connected based on time windows to obtain multimodal brain network data at each rehabilitation assessment time point.

[0022] In some embodiments, performing brain atlas partitioning and functional connectivity based on time windows on the brain imaging data to obtain multimodal brain network data at each rehabilitation assessment time point can be achieved through the following steps: The functional magnetic resonance imaging data in the brain imaging data at each rehabilitation assessment time point were divided into multiple time windows along the acquisition time axis to obtain multiple functional magnetic resonance signals. Brain regions are divided based on functional magnetic resonance imaging signals within each time window using brain atlas templates; Determine the Pearson correlation coefficients of each brain region within each time window, and construct the functional connectivity matrix corresponding to each time window; Brain regions were divided from diffusion tensor imaging data in brain imaging data at each rehabilitation assessment time point; Determine the connectivity strength of each brain region at each rehabilitation assessment time point, and construct the structural connectivity matrix corresponding to each rehabilitation assessment time point; By associating the structural connectivity matrix with all functional connectivity matrices at each rehabilitation assessment time point, multimodal brain network data for each rehabilitation assessment time point is obtained.

[0023] In practice, a sliding window can be used to divide the time window along the functional magnetic resonance imaging (fMRI) data acquisition time axis at each rehabilitation assessment time point. The window length and step size are predetermined based on the scan repetition time, the lowest frequency retained after preprocessing, and the stability verification results in the historical training sample set, and are kept consistent during the training and application phases. For example, under a scan protocol with a repetition time of 2 seconds, the window length can be set to 30 time points and the step size to 7 time points, with the blood oxygen level-dependent signal sequence within each window serving as the corresponding fMRI signal. After head movement rejection, only time windows containing a preset minimum number of consecutive effective time points are retained, without splicing signals across rejected time points. When the number of effective time windows is less than a preset fixed number, the corresponding rehabilitation assessment time point is marked as insufficient data.

[0024] In addition, in specific implementation, an automatic anatomical atlas can be used as a brain atlas template to assign voxel signals within each time window to the corresponding brain region and calculate the average time series of the brain region. The Pearson correlation coefficient between any two average time series of brain regions is calculated to form the functional connectivity matrix of the corresponding time window. White matter fiber tracking is performed on diffusion tensor imaging data, and the fiber tracking results are mapped to the same brain atlas template. The structural connectivity matrix of the corresponding rehabilitation assessment time point is formed by using the number of fibers between brain region pairs or the fiber connection strength normalized by a preset method.

[0025] In addition, in practical implementation, the rehabilitation assessment time point can be used as an index to associate and store the functional connectivity matrix and structural connectivity matrix of each time window corresponding to the rehabilitation assessment time point, so that the functional connectivity matrix and structural connectivity matrix have a consistent brain region order and matrix dimension, forming multimodal brain network data corresponding to the rehabilitation assessment time point.

[0026] It should be noted that in this application, the functional connectivity matrix is ​​used to represent the correlation between the average time series of each brain region within a time window, and the structural connectivity matrix is ​​used to represent the white matter fiber connectivity strength between each brain region within a rehabilitation assessment time point. The multimodal brain network data includes multiple functional connectivity matrices and one structural connectivity matrix at the same rehabilitation assessment time point.

[0027] In step 103, the functional connectivity matrix and structural connectivity matrix in the multimodal brain network data are normalized according to the modal scaling parameters determined based on the historical training sample set. The cross-modal shared sparse dictionary constructed based on the historical training sample set is called to obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point. Temporal consistency constraints are applied to the functional sparse coding coefficient vector to obtain the enhanced brain network data.

[0028] In some embodiments, reference Figure 2 As shown, the pre-construction of the cross-modal shared sparse dictionary can be achieved through the following steps: In step 1031, the functional connectivity matrix and structural connectivity matrix of each rehabilitation assessment time point in the historical training sample set are obtained. The functional modal scaling parameters and structural modal scaling parameters are determined according to the training set statistics of the functional connectivity matrix and structural connectivity matrix, respectively, and normalization is performed using the corresponding modal scaling parameters. In step 1032, the normalized functional connectivity matrix and structural connectivity matrix are used as input data for dictionary learning. The goal is to minimize the reconstruction error of the functional connectivity matrix and structural connectivity matrix under shared dictionary atoms, and to optimize and update the dictionary atoms and sparse coding coefficients. In step 1033, cross-modal consistency constraints are applied to the mean functional connectivity matrix and structural connectivity matrix at the same rehabilitation assessment time point in the longitudinal training samples of the same rehabilitation personnel. The optimization is terminated when the reconstruction error converges or reaches the preset number of iterations, and the cross-modal shared sparse dictionary is obtained.

[0029] In practice, the historical training sample set includes functional connectivity matrices and structural connectivity matrices obtained by multiple rehabilitation personnel at multiple rehabilitation assessment time points. First, the mean and standard deviation of the functional connectivity matrices and structural connectivity matrices in the training set are calculated respectively, and the vectorized results of the two types of matrices are standardized using the corresponding statistics. Dictionary learning takes the standardized matrix vectors as input, alternately updates dictionary atoms and sparse coding coefficients, and applies cross-modal consistency regularization terms to the sparse codes corresponding to the mean functional connectivity matrices and structural connectivity matrices of all functional connectivity matrices in the longitudinal training samples of the same rehabilitation personnel at the same rehabilitation assessment time point. The number of dictionary atoms, sparsity penalty coefficient, and cross-modal consistency coefficient are all determined on the validation set divided by rehabilitation personnel. In the application phase, the modal scale parameters obtained from the training are fixed and the cross-modal shared sparse dictionary is used.

[0030] The number of dictionary atoms K can be determined based on the number of rehabilitation patients and the matrix dimension in the historical training sample set. For example, when a longitudinal training sample of a rehabilitation patient includes 3 rehabilitation assessment time points and each rehabilitation assessment time point includes 10 functional connectivity matrices and 1 structural connectivity matrix, the longitudinal training sample corresponds to 33 matrices. The sparsity penalty coefficient is used to adjust the trade-off between reconstruction error and sparsity of sparse coding coefficients, and the L1 norm constraint is not interpreted as an upper limit on the number of non-zero dictionary atoms.

[0031] Cross-modal shared sparse dictionary is a set of dictionary atoms formed by using functional connectivity matrices and structural connectivity matrices under a fixed brain region order as training objects. Its function is to convert the two types of matrices into sparse codes in the same dictionary coordinate system, and does not directly equate functional connectivity strength with structural connectivity strength.

[0032] In some embodiments, applying a time consistency constraint to the functionally sparse coding coefficient vector to obtain enhanced brain network data can be achieved through the following steps: Based on the cross-modal shared sparse dictionary, the normalized functional connectivity matrix of each time window within each rehabilitation assessment time point and the normalized structural connectivity matrix of each rehabilitation assessment time point are sparsely decomposed to obtain the original functional sparse coding coefficient vector corresponding to each time window and the structural sparse coding coefficient vector corresponding to each rehabilitation assessment time point. The original functional sparse coding coefficient vectors of each time window are set as the corresponding functional sparse coding coefficient vectors to be optimized. The mean coding to be optimized at each rehabilitation assessment time point is determined according to the functional sparse coding coefficient vectors to be optimized. The rate of change of the mean coding to be optimized between adjacent rehabilitation assessment time points is determined according to the time interval between adjacent rehabilitation assessment time points. While maintaining the data fit between the sparse coding coefficient vector to be optimized and the original sparse coding coefficient vector and preserving sparsity, a time consistency penalty is applied to the difference in the rate of change of adjacent mean coding to be optimized, and the sparse coding coefficient vectors to be optimized in each time window are jointly optimized to obtain the constrained sparse coding coefficient vector. According to the rehabilitation assessment time point and time window, the restricted functional sparse coding coefficient vector is associated with the corresponding structural sparse coding coefficient vector to obtain the enhanced brain network data.

[0033] In specific implementation, the functional connectivity matrix of each time window and the structural connectivity matrix of the corresponding rehabilitation assessment time point are normalized using the functional modal scale parameters and structural modal scale parameters fixed in the application stage, respectively. The normalized functional connectivity matrix and the normalized structural connectivity matrix are vectorized and input into the cross-modal shared sparse dictionary to solve the original functional sparse coding coefficient vector a(i,t) corresponding to each time window and the structural sparse coding coefficient vector s(i) corresponding to each rehabilitation assessment time point.

[0034] Let a'(i,t) be the sparse coding coefficient vector of the function to be optimized at the t-th time window of the i-th rehabilitation assessment time point. The mean of all a'(i,t) within the same rehabilitation assessment time point is taken as the mean coding to be optimized, m(i). The time interval between the (i-1)-th and i-th rehabilitation assessment time points is taken as ΔT(i). r(i) = [m(i) - m(i-1)] / ΔT(i) is defined as the rate of change per unit time of adjacent mean codings to be optimized. The objective function includes the sum of the squared Euclidean distances between all a'(i,t) and their corresponding a(i,t), η multiplied by the sum of the L1 norms (also known as 1-norms) of all a'(i,t), and the number of rehabilitation assessment time points. When M is not less than 3, λ is multiplied by the sum of ||r(i+1)-r(i)||2² from i from 2 to M-1, where ||·||2² represents the square Euclidean norm, η is the sparsity coefficient, and λ is the time consistency coefficient. The candidate values ​​for the time consistency coefficient include zero, and the change from the baseline to the last score after unifying the symbols according to the improvement direction specified in the clinical rehabilitation evaluation scale is used as the verification label. The candidate value that minimizes the verification error of the rehabilitation degree score is selected. The data fitting term is used to avoid forcing adjacent rehabilitation assessment time points to the same state. When the optimization does not converge, a'(i,t) is set to a(i,t) and output as the constrained function sparse coding coefficient vector. When the number of rehabilitation assessment time points is less than 3, λ is set to zero.

[0035] The brain network data is enhanced by associating constrained functional sparse coding coefficient vector a'(i,t) and structural sparse coding coefficient vector s(i) according to rehabilitation assessment time points and time windows. This ensures that each time window corresponds to a functional code generated by the current functional connectivity matrix and a structural code generated by the current structural connectivity matrix. Temporal consistency constraints are used to limit isolated changes in the rate of change per unit time without deleting the fitted terms of the original coding data. The structural sparse coding coefficient vector, as the structural information of the corresponding rehabilitation assessment time point, is directly fed into the subsequent multimodal coding coefficient vector.

[0036] In step 104, the enhanced brain network data is subjected to neighborhood structure preservation dimensionality reduction to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point. Based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, the corresponding representative features are determined, and the representative features are connected in chronological order according to the rehabilitation assessment time points to obtain the rehabilitation process state trajectory.

[0037] In some embodiments, performing neighborhood structure-preserving dimensionality reduction on the enhanced brain network data to obtain multiple neighborhood structure-preserving features at each rehabilitation assessment time point can be achieved using the following steps: Obtain a neighborhood structure-preserving autoencoder trained based on a historical training sample set; The constrained functional sparse coding coefficient vectors corresponding to each time window in the enhanced brain network data are concatenated with the structural sparse coding coefficient vectors corresponding to the rehabilitation assessment time points. The concatenated multimodal coding coefficient vectors are then input into the neighborhood structure-preserving autoencoder to encode the reconstruction error and the neighborhood preservation error to jointly constrain the dimensionality reduction process. Output the low-dimensional feature vectors corresponding to each time window, and use the low-dimensional feature vectors as the neighborhood structure preservation features of the corresponding rehabilitation assessment time points.

[0038] In specific implementation, the constrained functional sparse coding coefficient vector a'(i,t) corresponding to each time window and the structural sparse coding coefficient vector s(i) corresponding to the rehabilitation assessment time point are concatenated in a fixed order to form a multimodal coding coefficient vector. The neighborhood structure-preserving autoencoder uses the multimodal coding coefficient vector in the historical training sample set as training input. The encoder maps the input vector to a low-dimensional feature vector, and the decoder reconstructs the multimodal coding coefficient vector based on the low-dimensional feature vector. The coding reconstruction error is used to limit the difference between the reconstructed vector and the input vector, and the neighborhood preservation error is used to limit the change of the distance relationship between adjacent input vectors in the low-dimensional feature space within the training batch. The two together constitute the training loss according to the weights determined by the validation set.

[0039] The neighborhood structure maintains that the input dimension of the autoencoder is twice the number of atoms in the cross-modal shared sparse dictionary, and the output dimension is smaller than the input dimension. The training set and validation set are divided by rehabilitation personnel. During the application phase, the encoder parameters are fixed so that the low-dimensional feature vectors of different rehabilitation personnel and different rehabilitation assessment time points are located in the same feature space.

[0040] The neighborhood structure preservation feature is a low-dimensional feature vector output by the neighborhood structure preservation autoencoder. The neighborhood structure refers to the distance adjacency relationship formed by the multimodal coding coefficient vector in the historical training sample set according to a preset number of nearest neighbors. The neighborhood structure preservation feature is a low-dimensional feature vector and does not have the property of a symmetric positive definite matrix.

[0041] In some embodiments, reference Figure 3 As shown, the representative features corresponding to the multiple neighborhood structure preservation features at each rehabilitation assessment time point are determined, and the representative features are connected in chronological order according to the rehabilitation assessment time points to obtain the rehabilitation process state trajectory. This can be achieved through the following steps: In step 1041, the mean aggregation of all neighborhood structure features within each rehabilitation assessment time point is performed to obtain the representative features of the corresponding rehabilitation assessment time point; In step 1042, the representative features of each rehabilitation assessment time point are connected in a common low-dimensional feature space according to the time order; In step 1043, the Euclidean distance between representative features of any adjacent rehabilitation assessment time points is calculated, and the representative features of each rehabilitation assessment time point, the Euclidean distance between adjacent representative features, and the time interval between adjacent rehabilitation assessment time points are collectively determined as the rehabilitation process state trajectory.

[0042] In practice, the arithmetic mean of the neighborhood structure corresponding to all time windows within each rehabilitation assessment point is calculated dimension by dimension to obtain the representative features of that rehabilitation assessment point. Each representative feature is located in the same low-dimensional Euclidean feature space, connected in the time order of rehabilitation assessment points, and the Euclidean distance between adjacent representative features is calculated.

[0043] It should be noted that the rehabilitation process trajectory is a feature space path composed of representative features at each rehabilitation assessment time point, Euclidean distance between adjacent representative features, and time interval between adjacent rehabilitation assessment time points. The Euclidean distance between adjacent representative features represents the differences in brain network features, and does not represent the rehabilitation direction or the degree of clinical improvement on its own.

[0044] In step 105, the corresponding data quality confidence index is determined based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point. The assessment weight at each rehabilitation assessment time point is determined based on the data quality confidence index. The assessment weight is associated with the rehabilitation process state trajectory to obtain a quality-weighted state trajectory. The rehabilitation degree score of the rehabilitation personnel is determined based on the quality-weighted state trajectory.

[0045] In some embodiments, determining the corresponding data quality confidence index based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point can be achieved using the following steps: The average frame shift and effective time point ratio were extracted from the functional magnetic resonance imaging data at each rehabilitation assessment time point, and the effective diffusion direction ratio, diffusion direction coverage and spatial registration error were extracted from the diffusion tensor imaging data. The average frame displacement and spatial registration error are compared with the corresponding upper limit thresholds determined based on the historical training sample set, and the effective time point ratio, effective diffusion direction ratio and diffusion direction coverage are compared with the corresponding lower limit thresholds determined based on the historical training sample set. For the average frame displacement and spatial registration error not exceeding the corresponding upper limit threshold, and the proportion of effective time points, the proportion of effective diffusion directions, and the coverage of diffusion directions not lower than the corresponding lower limit threshold, the directional interval normalization is performed in the same direction to obtain normalized quality parameters with values ​​ranging from zero to one that increase with the improvement of the corresponding data quality. The normalized quality parameters are weighted using non-negative combination coefficients that minimize the mean square error of the weighted result of the normalized quality parameters in the validation set of the historical training sample set relative to the image quality qualified labels with values ​​of zero or one. The sum of the non-negative combination coefficients is one, thus obtaining the data quality confidence index for the corresponding rehabilitation assessment time point.

[0046] The aforementioned quality parameters cover both functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI) data, reflecting the reliability of the image data from dimensions such as acquisition stability and spatial registration accuracy. Specifically, the average frame displacement characterizes the head movement amplitude during fMRI acquisition, representing the average of the sum of head translation and rotation calculated using a preset head radius at adjacent time points; the effective time point ratio measures the effective acquisition percentage of functional data, calculated as the ratio of the number of remaining effective time points after removing time points with frame displacement exceeding a threshold to the original total number of time points.

[0047] For diffusion tensor imaging data, the effective diffusion direction ratio reflects the qualified proportion of diffusion gradient data, which is the ratio of the number of diffusion directions that pass the signal intensity and motion quality check to the total number of original diffusion directions; diffusion direction coverage is used to characterize the spatial distribution completeness of diffusion gradient, which is the ratio of the number of preset spherical orientation partitions that contain at least one effective diffusion direction to the total number of orientation partitions; spatial registration error is used to measure the alignment accuracy between the image and the standard brain atlas, which is the average distance between corresponding anatomical landmarks after the data is registered to the brain atlas template.

[0048] In addition, the image quality pass label is determined by image quality reviewers who did not participate in model training based on head motion artifacts, the number and coverage of effective diffusion directions, tensor fitting residuals, and spatial registration results. Each quality parameter is mapped to zero to one according to the corresponding threshold. The average frame shift and spatial registration error are normalized in a way that decreases as the value increases, while the other quality parameters are normalized in a way that increases as the value increases. The image quality pass label takes the value of zero or one, and the combination coefficient is limited to non-negative and summed to one. It is determined on the validation set of the historical training sample set divided by rehabilitation personnel according to the mean square error between the weighted result of the normalized quality parameters and the image quality pass label.

[0049] In some embodiments, determining the assessment weight of each rehabilitation assessment time point based on the data quality confidence index, and associating the assessment weight with the rehabilitation process state trajectory to obtain a quality-weighted state trajectory can be achieved through the following steps: Rehabilitation assessment time points whose average frame displacement or spatial registration error is higher than the corresponding upper limit threshold, or whose effective time point ratio, effective diffusion direction ratio, or diffusion direction coverage is lower than the corresponding lower limit threshold, are marked as invalid rehabilitation assessment time points. Rehabilitation assessment time points that are not marked as invalid rehabilitation assessment time points are determined as valid rehabilitation assessment time points. When there are invalid rehabilitation assessment time points or the number of valid rehabilitation assessment time points is less than the preset fixed number of assessment time points, a data shortage mark is generated and the determination of rehabilitation degree score is stopped. A preset minimum positive value is added to the data quality confidence index at each effective rehabilitation assessment time point. The corresponding assessment weight is determined based on the data quality confidence index after adding the preset minimum positive value, and all assessment weights are normalized. By keeping the representative feature coordinates, time sequence, and Euclidean distance between adjacent representative features of each effective rehabilitation assessment time point in the rehabilitation process state trajectory unchanged, the assessment weight of each effective rehabilitation assessment time point is associated with the corresponding representative feature to obtain the quality-weighted state trajectory.

[0050] In practice, if the average frame displacement or spatial registration error of any rehabilitation assessment time point is greater than the corresponding upper limit threshold, or the proportion of effective time points, the proportion of effective diffusion directions, or the coverage of diffusion directions is less than the corresponding lower limit threshold, then that rehabilitation assessment time point is marked as invalid. Rehabilitation assessment time points not marked as invalid are determined as valid. If there are invalid rehabilitation assessment time points or the number of valid rehabilitation assessment time points is less than the preset fixed number of assessment time points, a data insufficiency flag is generated and the determination of the rehabilitation level score is stopped. When all preset rehabilitation assessment time points are valid, the preset maximum... Small positive values ​​ε are added to the data quality confidence index at each rehabilitation assessment time point, and the proportion of each data quality confidence index after adding ε to the sum of all data quality confidence indices after adding ε is used as the corresponding assessment weight, where ε is greater than zero; when forming the quality-weighted state trajectory, the original representative feature coordinates, time order and adjacent Euclidean distance of each valid rehabilitation assessment time point are retained, and only the assessment weight is used as an additional quantity associated with the corresponding representative feature, without weighted averaging of all trajectory point coordinates; the fixed number of assessment time points and the relative order of each assessment time point are consistent with the historical training sample set of the regression model.

[0051] In some embodiments, determining the rehabilitation level score of a recovering individual based on the quality-weighted state trajectory can be achieved using the following steps: The representative features of each effective rehabilitation assessment time point, the Euclidean distance between adjacent representative features, the time interval between adjacent rehabilitation assessment time points, and the assessment weight of each effective rehabilitation assessment time point are extracted from the quality-weighted state trajectory and combined to obtain the trajectory features. The regression model is trained by inputting the trajectory features into the historical training sample set as training input and using the change in baseline to the last score after unifying the improvement direction according to the clinical rehabilitation evaluation scale as training label, and thus obtaining a rehabilitation degree score representing the predicted score improvement.

[0052] In specific implementation, the historical training sample set includes trajectory features formed according to the same data processing flow as the application phase, a fixed number of assessment time points, and the order of assessment time points, as well as scale scores obtained by medical staff at the baseline and final rehabilitation assessment time points using the same clinical rehabilitation evaluation scale. The difference between the final scale score and the baseline scale score is standardized with a unified sign according to the improvement direction specified by the clinical rehabilitation evaluation scale to obtain a score change quantity where an increase in the numerical value indicates an increase in the degree of improvement, and this score change quantity is used as the training label for the regression model. The training set and validation set are divided by rehabilitation personnel. The regression model is fitted using the training set, and the model parameters and applicable assessment criteria are determined using the validation set. The model is divided into several categories. Each regression model is only used for application subjects that have the same target disease type, scanning protocol, clinical rehabilitation evaluation scale, number of assessment time points, and assessment time point order as the historical training sample set. During the application phase, representative features, adjacent Euclidean distances, time intervals, and assessment weights are input into a fixed regression model, and the output is a rehabilitation degree score that is within the training label range and represents the predicted score improvement. When there are invalid rehabilitation assessment time points, the application subject does not meet the above applicable conditions, or the input features exceed the training range, no score is output and a data insufficiency or inapplicability label is generated. The rehabilitation degree score is used to provide auxiliary quantitative information and does not replace the clinical judgment of medical staff.

[0053] In some embodiments, this application provides a rehabilitation level evaluation system for neurological rehabilitation, with reference to... Figure 4 The system includes a data acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire resting-state functional magnetic resonance imaging data and diffusion tensor imaging data of rehabilitation personnel at multiple rehabilitation assessment time points during the rehabilitation cycle, so as to obtain brain imaging data.

[0054] Processing module 402, in this application, is mainly used to perform brain atlas partitioning and functional connectivity based on time windows on the brain imaging data to obtain multimodal brain network data at each rehabilitation assessment time point.

[0055] In a specific implementation, the processing module 402 is further used to normalize the functional connectivity matrix and structural connectivity matrix in the multimodal brain network data according to the modal scale parameters determined based on the historical training sample set, call the cross-modal shared sparse dictionary constructed based on the historical training sample set, obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point, and apply time consistency constraints to the functional sparse coding coefficient vector to obtain the enhanced brain network data.

[0056] In addition, in a specific implementation, the processing module 402 is also used to perform neighborhood structure preservation dimensionality reduction on the enhanced brain network data to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point, determine the corresponding representative features based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, and connect the representative features according to the time sequence of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory.

[0057] The execution module 403 in this application is mainly used to determine the corresponding data quality confidence index based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point, determine the assessment weight at each rehabilitation assessment time point based on the data quality confidence index, associate the assessment weight with the rehabilitation process state trajectory to obtain the quality-weighted state trajectory, and determine the rehabilitation degree score of the rehabilitation personnel based on the quality-weighted state trajectory.

[0058] This application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described method for evaluating the degree of rehabilitation in neurological rehabilitation.

[0059] In some embodiments, reference Figure 5 The method for evaluating the degree of rehabilitation in neurological rehabilitation is as follows: Figure 5 The computer device shown is implemented, and the computer device includes at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0060] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0061] The communication bus 502 can be used to transmit information between the aforementioned components.

[0062] The memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto; the memory 503 may exist independently and be connected to the processor 501 via the communication bus 502; the memory 503 may also be integrated with the processor 501.

[0063] The memory 503 is used to store program code that executes the scheme of this application, and the execution is controlled by the processor 501; the processor 501 is used to execute the program code stored in the memory 503; the program code may include one or more software modules; the method used in the above embodiments can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0064] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0065] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor; a processor here may refer to one or more devices, circuits and / or processing cores for processing data (e.g., computer program instructions).

[0066] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device; in specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device; the embodiments of this application do not limit the type of computer device.

[0067] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for evaluating the degree of rehabilitation in neurological rehabilitation.

[0068] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments; therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0069] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application; thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for evaluating the degree of rehabilitation in neurological rehabilitation, characterized in that, Includes the following steps: Brain imaging data were obtained by collecting resting-state functional magnetic resonance imaging (fMRI) data and diffusion tensor imaging data of rehabilitation participants at multiple rehabilitation assessment points during the rehabilitation cycle. The brain imaging data is divided into brain atlases and functional connections based on time windows to obtain multimodal brain network data at each rehabilitation assessment time point; The functional connectivity matrix and structural connectivity matrix in the multimodal brain network data are normalized according to the modal scaling parameters determined based on the historical training sample set. The cross-modal shared sparse dictionary constructed based on the historical training sample set is called to obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point. Temporal consistency constraints are applied to the functional sparse coding coefficient vector to obtain the enhanced brain network data. The enhanced brain network data is subjected to neighborhood structure preservation dimensionality reduction to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point. Based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, the corresponding representative features are determined, and the representative features are connected in the time order of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory. Based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point, the corresponding data quality confidence index is determined. Based on the data quality confidence index, the assessment weight at each rehabilitation assessment time point is determined. The assessment weight is associated with the rehabilitation process state trajectory to obtain a quality-weighted state trajectory. Based on the quality-weighted state trajectory, the rehabilitation degree score of the rehabilitation personnel is determined.

2. The method as described in claim 1, characterized in that, The brain imaging data is subjected to time-window-based brain mapping and functional connectivity analysis to obtain multimodal brain network data at each rehabilitation assessment time point, specifically including: The functional magnetic resonance imaging data in the brain imaging data at each rehabilitation assessment time point were divided into multiple time windows along the acquisition time axis to obtain multiple functional magnetic resonance signals. Brain regions are divided based on functional magnetic resonance imaging signals within each time window using brain atlas templates; Determine the Pearson correlation coefficients of each brain region within each time window, and construct the functional connectivity matrix corresponding to each time window; Brain regions were divided from diffusion tensor imaging data in brain imaging data at each rehabilitation assessment time point; Determine the connectivity strength of each brain region at each rehabilitation assessment time point, and construct the structural connectivity matrix corresponding to each rehabilitation assessment time point; By associating the structural connectivity matrix with all functional connectivity matrices at each rehabilitation assessment time point, multimodal brain network data for each rehabilitation assessment time point is obtained.

3. The method as described in claim 1, characterized in that, The cross-modal shared sparse dictionary is pre-constructed in the following manner: Obtain the functional connectivity matrix and structural connectivity matrix at each rehabilitation assessment time point in the historical training sample set. Determine the functional modal scaling parameters and structural modal scaling parameters based on the training set statistics of the functional connectivity matrix and structural connectivity matrix, respectively, and perform normalization processing using the corresponding modal scaling parameters. The normalized functional connectivity matrix and structural connectivity matrix are used as input data for dictionary learning. The goal is to optimize the updating of dictionary atoms and sparse coding coefficients by minimizing the reconstruction error of the functional connectivity matrix and structural connectivity matrix under shared dictionary atoms. A cross-modal consistency constraint is applied to the mean functional connectivity matrix and structural connectivity matrix at the same rehabilitation assessment time point in the longitudinal training samples of the same rehabilitation patient. The optimization is terminated when the reconstruction error converges or reaches the preset number of iterations, thus obtaining the cross-modal shared sparse dictionary.

4. The method as described in claim 1, characterized in that, Applying a time consistency constraint to the functionally sparse coding coefficient vector to obtain enhanced brain network data specifically includes: Based on the cross-modal shared sparse dictionary, the normalized functional connectivity matrix of each time window within each rehabilitation assessment time point and the normalized structural connectivity matrix of each rehabilitation assessment time point are sparsely decomposed to obtain the original functional sparse coding coefficient vector corresponding to each time window and the structural sparse coding coefficient vector corresponding to each rehabilitation assessment time point. The original functional sparse coding coefficient vectors of each time window are set as the corresponding functional sparse coding coefficient vectors to be optimized. The mean coding to be optimized at each rehabilitation assessment time point is determined according to the functional sparse coding coefficient vectors to be optimized. The rate of change of the mean coding to be optimized between adjacent rehabilitation assessment time points is determined according to the time interval between adjacent rehabilitation assessment time points. While maintaining the data fit between the sparse coding coefficient vector to be optimized and the original sparse coding coefficient vector and preserving sparsity, a time consistency penalty is applied to the difference in the rate of change of adjacent mean coding to be optimized, and the sparse coding coefficient vectors to be optimized in each time window are jointly optimized to obtain the constrained sparse coding coefficient vector. According to the rehabilitation assessment time point and time window, the restricted functional sparse coding coefficient vector is associated with the corresponding structural sparse coding coefficient vector to obtain the enhanced brain network data.

5. The method as described in claim 1, characterized in that, The enhanced brain network data is subjected to neighborhood structure preservation dimensionality reduction to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point, specifically including: Obtain a neighborhood structure-preserving autoencoder trained based on a historical training sample set; The constrained functional sparse coding coefficient vectors corresponding to each time window in the enhanced brain network data are concatenated with the structural sparse coding coefficient vectors corresponding to the rehabilitation assessment time points. The concatenated multimodal coding coefficient vectors are then input into the neighborhood structure-preserving autoencoder to encode the reconstruction error and the neighborhood preservation error to jointly constrain the dimensionality reduction process. Output the low-dimensional feature vectors corresponding to each time window, and use the low-dimensional feature vectors as the neighborhood structure preservation features of the corresponding rehabilitation assessment time points.

6. The method as described in claim 1, characterized in that, Based on the neighborhood structure preservation features of each rehabilitation assessment time point, corresponding representative features are determined, and the representative features are connected in chronological order according to the rehabilitation assessment time points to obtain the rehabilitation process state trajectory, which specifically includes: The mean of all neighborhood structures within each rehabilitation assessment time point is aggregated to obtain the representative features of the corresponding rehabilitation assessment time point. The representative features of each rehabilitation assessment time point are connected in a common low-dimensional feature space according to the time sequence. Calculate the Euclidean distance between representative features of any adjacent rehabilitation assessment time points, and determine the rehabilitation process state trajectory by combining the representative features of each rehabilitation assessment time point, the Euclidean distance between adjacent representative features, and the time interval between adjacent rehabilitation assessment time points.

7. The method as described in claim 1, characterized in that, Based on the functional magnetic resonance imaging (fMRI) data quality parameters and diffusion tensor imaging (DTI) data quality parameters at each rehabilitation assessment time point, the corresponding data quality confidence indices are determined, specifically including: The average frame shift and effective time point ratio were extracted from the functional magnetic resonance imaging data at each rehabilitation assessment time point, and the effective diffusion direction ratio, diffusion direction coverage and spatial registration error were extracted from the diffusion tensor imaging data. The average frame displacement and spatial registration error are compared with the corresponding upper limit thresholds determined based on the historical training sample set, and the effective time point ratio, effective diffusion direction ratio and diffusion direction coverage are compared with the corresponding lower limit thresholds determined based on the historical training sample set. For the average frame displacement and spatial registration error not exceeding the corresponding upper limit threshold, and the proportion of effective time points, the proportion of effective diffusion directions, and the coverage of diffusion directions not lower than the corresponding lower limit threshold, the directional interval normalization is performed in the same direction to obtain normalized quality parameters with values ​​ranging from zero to one that increase with the improvement of the corresponding data quality. The normalized quality parameters are weighted using non-negative combination coefficients that minimize the mean square error of the weighted result of the normalized quality parameters in the validation set of the historical training sample set relative to the image quality qualified labels with values ​​of zero or one. The sum of the non-negative combination coefficients is one, thus obtaining the data quality confidence index for the corresponding rehabilitation assessment time point.

8. The method as described in claim 1, characterized in that, Based on the data quality confidence index, the assessment weights for each rehabilitation assessment time point are determined, and the assessment weights are associated with the rehabilitation process state trajectory to obtain the quality-weighted state trajectory, which specifically includes: Rehabilitation assessment time points whose average frame displacement or spatial registration error is higher than the corresponding upper limit threshold, or whose effective time point ratio, effective diffusion direction ratio, or diffusion direction coverage is lower than the corresponding lower limit threshold, are marked as invalid rehabilitation assessment time points. Rehabilitation assessment time points that are not marked as invalid rehabilitation assessment time points are determined as valid rehabilitation assessment time points. When there are invalid rehabilitation assessment time points or the number of valid rehabilitation assessment time points is less than the preset fixed number of assessment time points, a data shortage mark is generated and the determination of rehabilitation degree score is stopped. A preset minimum positive value is added to the data quality confidence index at each effective rehabilitation assessment time point. The corresponding assessment weight is determined based on the data quality confidence index after adding the preset minimum positive value, and all assessment weights are normalized. By keeping the representative feature coordinates, time sequence, and Euclidean distance between adjacent representative features of each effective rehabilitation assessment time point in the rehabilitation process state trajectory unchanged, the assessment weight of each effective rehabilitation assessment time point is associated with the corresponding representative feature to obtain the quality-weighted state trajectory.

9. The method as described in claim 1, characterized in that, The specific steps for determining the rehabilitation level score of recovered individuals based on the aforementioned quality-weighted state trajectory include: The representative features of each effective rehabilitation assessment time point, the Euclidean distance between adjacent representative features, the time interval between adjacent rehabilitation assessment time points, and the assessment weight of each effective rehabilitation assessment time point are extracted from the quality-weighted state trajectory and combined to obtain the trajectory features. The regression model is trained by inputting the trajectory features into the historical training sample set as training input and using the change in baseline to the last score after unifying the improvement direction according to the clinical rehabilitation evaluation scale as training label, and thus obtaining a rehabilitation degree score representing the predicted score improvement.

10. A rehabilitation assessment system for neurological rehabilitation, characterized in that, include: The acquisition module is used to collect resting-state functional magnetic resonance imaging data and diffusion tensor imaging data of rehabilitation personnel at multiple rehabilitation assessment time points during the rehabilitation cycle to obtain brain imaging data. The processing module is used to perform brain atlas partitioning and functional connectivity based on time windows on the brain imaging data to obtain multimodal brain network data at each rehabilitation assessment time point; The processing module is further configured to normalize the functional connectivity matrix and structural connectivity matrix in the multimodal brain network data according to the modal scale parameters determined based on the historical training sample set, call the cross-modal shared sparse dictionary constructed based on the historical training sample set, obtain the functional sparse coding coefficient vector of each time window within each rehabilitation assessment time point and the structural sparse coding coefficient vector of each rehabilitation assessment time point, and apply time consistency constraints to the functional sparse coding coefficient vector to obtain the enhanced brain network data. The processing module is also used to perform neighborhood structure preservation dimensionality reduction on the enhanced brain network data to obtain multiple neighborhood structure preservation features at each rehabilitation assessment time point, determine the corresponding representative features based on the multiple neighborhood structure preservation features at each rehabilitation assessment time point, and connect the representative features in the time order of the rehabilitation assessment time points to obtain the rehabilitation process state trajectory. The execution module is used to determine the corresponding data quality confidence index based on the functional magnetic resonance imaging data quality parameters and diffusion tensor imaging data quality parameters at each rehabilitation assessment time point, determine the assessment weight at each rehabilitation assessment time point based on the data quality confidence index, associate the assessment weight with the rehabilitation process state trajectory to obtain the quality-weighted state trajectory, and determine the rehabilitation degree score of the rehabilitation personnel based on the quality-weighted state trajectory.

Images